IBM's Eight-Core, 4-GHz Power7 Chip

pacopico writes "The first details on IBM's upcoming Power7 chip have emerged. The Register is reporting that IBM will ship an eight-core chip running at 4.0 GHz. The chip will support four threads per core and fit into some huge systems. For example, University of Illinois is going to house a 300,000-core machine that can hit 10 petaflops. It'll have 620 TB of memory and support 5 PB/s of memory bandwidth. Optical interconnects anyone?"

Comments

Comment removed

[account_deleted]

Comment removed based on user account deletion

Sorry Mac fans . . .

[Anonymous Coward]

No AltiVec for you!

Toasty.

[Anonymous Coward]

In other news, temperatures on the University of Illinois campus have mysteriously risen ten degrees. Scientists are still examining possible causes..

Re:Toasty.

[Yvan256]

Good thing they have a brand-new supercomputer, analyzing this temperature anomaly will be much faster!

Re:Toasty.

[jmorris42]

> Scientists are still examining possible causes..

Nah. If something gets warmer it is caused by Global Warming and the solution is to eliminate Western industrial civilization.

If something gets colder it is Global Climate Change and the solution is to eliminate Western industrial civilization.

If we have more hurricanes it is Global Warming. Fewer and it is Climate Change. More tornadoes? Global Warming. Floods caused by increased snowfall? Somehow that was also Global Warming, I'd have thought they would have went with Global Climate Change, but every rule seems to need an exception.

Re:Toasty.

[NICK SPACEE]

SUPERCOMPUTERS! YES!

Re:Toasty.

[Anonymous Coward]

Yet another clueless idiot babbling about stuff real scientists have devoted their lives to investigating.

Re:Toasty.

[iminplaya]

Guess they should have kept the reactor running.

Re:Toasty.

[DilutedImage]

Yet another Anonymous Coward talking crap

Re:Toasty.

[smart.id]

There's a difference between scientists and the mainstream media's coverage of that science. And I don't speak for jmorris here, but that just might be what he's satirizing.

Re:Toasty.

[Hal_Porter]

I think the people that complain about Global Warming hype are not complaining about the science but the dumbed down and politically motivated 'summary of what the vast majority of respectable scientists believe' from people who are activists not scientists.

Re:Toasty.

[Anonymous Coward]

No no no, obviously all the shit we put into the atmosphere is benign!! Fox news says so, so it MUST be true! Who are you going to beleive, thousands of climate scientists or Ann Coulter? I thought so!

Re:Toasty.

[Nutria]

Yet another clueless idiot babbling about stuff real scientists have devoted their lives to investigating.

Argumentum ad verecundiam doesn't impress me very much.
http://www.don-lindsay-archive.org/skeptic/arguments.html#authority

Re:Toasty.

[jmorris42]

> not to mention, in 2007 that the northwest passage was completely ice free for the first time in recorded history.

Yea, right. Pull the other one. First time in recorded history huh? Except for 1906, 1944, 1957, 1969, 1977, 1984, 1985, 1988 and 2000 in wooden ships, catamarans, naval vessels, cruise ships, etc.

Stop beliving the propaganda and do some googling before you open yer piehole and up looking like a retard.

btw, here is the link I got from Google searching for "northwest passage ice free"
Classically Liberal: Bad reporting about the Northwest Passage issue

Re:Toasty.

[TapeCutter]

Take a deep breath, the scientists won the science argument years ago and apart from quoting the economic 'alarmist' solution there is nothing technically wrong in his sarcastic generalizations. I have been posting on slashdot for ~8yrs defending the scientific strength of the IPCC's reports but I can still manage to appreciate well thought out sarcasm.

BTW: If you think a post is clueless then put some information in your reply, AC ad-homs won't convince anyone of anything except perhaps that your a witless arsehole.

Re:Toasty.

[noidentity]

Floods caused by increased snowfall? Somehow that was also Global Warming, I'd have thought they would have went with Global Climate Change, but every rule seems to need an exception.

What do you think melted the snow? Global Warming of course!

Re:Toasty.

[Anonymous Coward]

[insert Heisenberg uncertainty principle joke here]

Re:Toasty.

[Anonymous Coward]

Real credible blog.

I guess I should believe that over the vast majority of intellectuals.

Re:Toasty.

[Anonymous Coward]

In the end, it will be attributed to kids having no respect for the environment and driving SUVs.

Re:Toasty.

[jmorris42]

Sigh. This is what faith looks like, not reason.

So not only is that blog a lie cooked up by the vast right wing conspiracy, so are all of the links on it... So the US Coast Guard is in on it, as is Radio Canada, the Vancouver Maritime Museum, the BBC, etc. I bet you will bring up the CFR, the Rothchilds and perhaps the Freemasons before you admit that a piece of evidence supporting your religious beliefs is bogus.

News flash, big powerful enviromental groups have no monopoly on the truth, even the smallest blog can link to facts. But you wouldn't know about those, Faith trumps Reason after all. Behold the might Anonymous Coward, so certain in his Faith in Gaia that he posts snarky comments on slashdot but hasn't the stones to put his name to them. Gaia ain't going to let you into Heaven for that.

Re:Toasty.

[Anonymous Coward]

You almost have a point. But if you are going to question the scientists, you better do some research (and get some specialized education) yourself. You can't base your facts on news papers and crazy articles on the web.

Otherwise, it will be pointless to argue against them. It's like saying the world is flat, but refuse to accept that you don't have the mathematical skills to prove it.

Re:Toasty.

[Geak]

What really causes global warming? I for one beleive that the earth is still trying to figure out the question that arrives at the answer 42. It's just doing some really complex number crunching right now and the CPU is getting hot as a result.

Re:Toasty.

[Anonymous Coward]

yet another anonymous coward who has drunk the kool-aid. I for one believe everything that my cnn overlords tell me and all the agenda minded people that fund those "real scientists"

Re:Toasty.

[Anonymous Coward]

no, its subject to the heisenburger effect ..

Re:Toasty.

[Anonymous Coward]

Nah. If something gets warmer it is caused by Global Warming and the solution is to eliminate Western industrial civilization.

If "Western industrial civilization", means be a fatty freedom fries eater who rides his Hummer to the drive-inn 10 miles away to get a triple burger.
Yes, the real civilization doesn't need them.

Re:Toasty.

[hostyle]

An African or European Heisenberg?

Re:Toasty.

[Artuir]

Hey, try not to get him too riled up. He might start talking about how flu shots are really designed by big government for the true purpose of mind control.

Having argued with someone who quite literally believed in this tripe, I can safely tell you that you're wasting your time. Though I do admire the fact you are posting the information for the rest of us to read as well. It is very fascinating!

Re:Toasty.

[Anonymous Coward]

Scientists will use this Super Computer to simulate the effects of this super computer on Global Warming.

Re:Toasty.

[moonbender]

Right, since it's the environmental groups that are powerful, and not those whose interests are directly opposed to theirs. That small so-called liberal blog is fighting the big environmental establishment! Never mind that the BBC post they viciously attack contains essentially all the information about past voyages through the passage, which they so painstakenly researched using a Google search.

Re:Toasty.

[rxmd]

not to mention, in 2007 that the northwest passage was completely ice free for the first time in recorded history.

Yea, right. Pull the other one. First time in recorded history huh? Except for 1906, 1944, 1957, 1969, 1977, 1984, 1985, 1988 and 2000 in wooden ships, catamarans, naval vessels, cruise ships, etc.

One should note that in 1906 at least it wasn't exactly ice free, which is why it took Amundsen and his Gjoa three years to pass through (1903-1906). Your list is basically a list of years when some vessel finished sailing through the North-West Passage, but it doesn't really say anything about how much ice they encountered on the way.

That's also the basic fallacy of the blog you're linking to - it mentions an ice-free North-West Passage, but only for 2000. For the other years it just mentions a couple of vessels, while not really saying anything ice (except for 1984, where it says that the ice was "in retreat", implying that there still was some ice there).

So, for a comment like "Stop beliving the propaganda and do some googling before you open yer piehole and up looking like a retard." you are leaning out of the window a bit too much for my taste.

Re:Toasty.

[Jedi Alec]

What really causes global warming? I for one beleive that the earth is still trying to figure out the question that arrives at the answer 42.

Just wait till it figures out the optimal situation to determine the question would be to raise the surface temperature to 42 degrees centigrade.

Re:Toasty.

[pangloss]

Argumentum ad verecundiam doesn't impress me very much.
http://www.don-lindsay-archive.org/skeptic/arguments.html#authority

Appealing to the authority of an obscure website to make an argument against an appeal to authority is either terribly clever or terribly obtuse. Well done!

Re:Toasty.

[Muad'Dave]

Is that the hamburger that does/doesn't have cheese until you look?

Re:Toasty.

[Muad'Dave]

I have 150 slashdot freaks!

Are they hanging from hooks in your basement?

Re:Toasty.

[Muad'Dave]

An awful lot of 'real scientists' (in their day) devoted their lives to alchemy and Platonic solids. Ignorance and misguided research can pop up in any age, regardless of your perceived level of 'scientific' sophistication.

Re:Toasty.

[i_ate_god]

Did Global Warming give me the second worst winter in history with 300+cm of snowfall?

Re:Toasty.

[Danathar]

Should be a nice way to keep warm during those cold Illinois winters.

Re:Toasty.

[Anonymous Coward]

There's a huge difference between the Passage being navigated by a single ship and the passage being ice-free. I'm pretty amazed that so many people have tried to pull this card without realizing this.

Re:Toasty.

[limaxray]

I hate to tell you this, but the 'recorded history' of the Northwest Passage doesn't cover a very lengthy period of time. In fact its 'recorded history' only began in 1978 and in terms of global trends I'd say 30 years of data doesn't mean jack. But of course you already knew that because you actually read the article you linked to and not just some /.er's slant.

Re:Toasty.

[davolfman]

As are the police.

Re:Toasty.

[davolfman]

No-one's successfully conveyed to them yet that the Kyoto Treaty means they have to either give up their Lear jets and Mercedes or get to like Nuclear power.

Re:Toasty.

[Anonymous Coward]

That's why it's called global disruption now:
http://www.democracynow.org/2008/7/3/global_disruption_more_accurately_describes_climate

Re:Toasty.

[kesuki]

exactly, the northwest passage is a huge channel of sea water, for that channel to become completely ice free is a huge difference from ships getting through there over a course of 3 years...

i imagine that if the 2000, voyage was ice free, satellite images still showed a considerable amount of the passage still having ice shelves. the point of my post was that things are getting worse, and the evidence he linked to actually confirmed that at one point, it took 3 years to sail a ship through because it would get ice locked, 1/3 of the way through every winter, and then in 2000 it could be sailed through in 1 year with no ice lock problem..

the problem is simple, none of the energy companies want to be held responsible for what they've done, they can pinpoint exactly what happened to big tobacco, even though the tobacco industry knew for decades before anyone else that smoking was killing people... so energy companies are fighting with all the dirty underhanded tactics available to them. making pseudo science about global cooling, etc.

Re:Toasty.

[spun]

You are acting as if the parent post was replying to a reasoned argument. Please explain how the grandparent post was in any way valid.

Re:Toasty.

[spun]

You may want to read up a bit about how climate works in order not to appear the fool with statements like this. Pumping more heat into the climate makes it more extreme, more water evaporates and falls back as snow. Global warming can cause local cooling, as patterns of heat transfer change.

Re:Toasty.

[Anonymous Coward]

Yea except that the IBM systems are not only the fastest but also the most energy efficient.

Of course it will use some juice but it isn't like it'll need it's own powerplant.

Re:Toasty.

[spun]

So who shipped the most goods over the northwest passage? I mean, there must have been tones of companies doing it, right? Oh, yeah, we don't have any history about that, riiight.

No one shipped anything over the northwest passage. Why do you suppose that was?

Re:Toasty.

[Anonymous Coward]

You're thinking of the Schrodinburger.

Re:Toasty.

[Anonymous Coward]

its already to freaking hot up there as it is

Re:Toasty.

[whoisisis]

And will thus become self-aware

Re:Toasty.

[Artuir]

I see someone's enjoying the vast powers of over-generalization! Is your horse nice and comfortable?

Re:Toasty.

[limaxray]

Hey, don't hate on me, I just clarified what the article he linked to said. It's plain 100% unbiased fact that the 'recorded history' he is referring to is the satellite data recorded since 1978, and not some long historical record he's trying to lead you to believe. Just RTFA, and not just the blatantly slanted /. summery that conveniently leaves out that little tid bit.

And the reason the NW Passage isn't used as a shipping route is it is still iced over a good part of the year. Because of this, there are no ports or any other such infrastructure to facilitate using such a route. Regardless, there have been a number of people using plain-old-run-of-the-mill vessels to sail the passage over the past century; something that wouldn't be possible if there has always been a persistent ice pack as you AGW nuts like to suggest.

Re:Toasty.

[spun]

We all know that it has been possible under certain conditions to traverse the northwest passage, by going around the ice. It wasn't always fully frozen, occasionally one could find a channel. That's a lot different than it being totally free from ice.

Sorry, you guys are the nuts. Global warming is real, destructive to the global economy, and caused by human beings. You don't want it to be true, so you will find 'evidence' to support your desired conclusions, no matter how scanty, poorly supported, anecdotal, or thoroughly debunked it is.

But it is true, the stupid hippies are right, you are wrong, it will impact you, and you will have to change your lifestyle. Your laziness and smug selfishness will not be allowed to destroy the planet, sorry.

Re:Toasty.

[TapeCutter]

"An awful lot of 'real scientists' (in their day) devoted their lives to alchemy and Platonic solids. Ignorance and misguided research can pop up in any age, regardless of your perceived level of 'scientific' sophistication."

Thus our recent ancestors found the best choice was to formalize and constantly refine a USEFULL scientific methodology that demands the highest standards of emprical evidence and repeatability but never claims truth with absolute certainty.

Re:Toasty.

[Muad'Dave]

Absolutely. Those who claim that anthropogenic global warming is an 'absolute certainty' as are not being very rigorous, IMHO. As has been stated here, their models have such large error bars that using them to predict the future of such a large, complex, and chaotic system is hubris.

I feel kinda lonely...

[Anonymous Coward]

...in here all by myself.

Anyway...that's some killer processing power. When do I get one on my desktop?

Re:I feel kinda lonely...

[von_rick]

The processor is gonna make humans obsolete. Resistance is futile, you will be assimilated.

Core pron

[Anonymous Coward]

"For example, University of Illinois is going to house a 300,000-core machine that can hit 10 petaflops. It'll have 620 TB of memory and support 5 PB/s of memory bandwidth."

I came.

Re:Core pron

[mcpkaaos]

I came.

I saw.

Re:Core pron

[vbraga]

I won.

Re:Core pron

[EkriirkE]

That's dirty.

Re:Core pron

[EkriirkE]

Err, I conquered.

Re:Core pron

[McGiraf]

Veni, vini, vomi.

Re:Core pron

[Anonymous Coward]

I kicked its ass

Re:Core pron

[LittleBigScript]

I came.

I saw.

I compiled.

Re:Core pron

[Anonymous Coward]

I came.

I saw.

I conquered.

Re:Core pron

[rhyno46]

iCame

Is that a new Apple product?

Re:Core pron

[Kingrames]

I...

KHaaaaaaaaaaaaaaaaaaaaaaaaaaaNnnnn!!!!!!!!!!!!

Re:Core pron

[Andrew Kismet]

I lost (the game).

Re:Core pron

[Anonymous Coward]

I, Caesar, when I learned of the fame
Of Cleopatra, I straightway laid claim.
Ahead of my legions,
I invaded her regions,
I saw, I conquered, I came.

Re:Core pron

[Kligat]

I concurred.

Re:Core pron

[Quicksilver_Johny]

I came, of the wine, to be thrown up?

Re:Core pron

[Anonymous Coward]

I conquRRRRRGHHHHHHHHHH

Re:Core pron

[Alpha830RulZ]

Need a washcloth?

Re:Core pron

[tietokone-olmi]

I came.

I saw.

And I had to clean it up. Damn your bones.

Re:Core pron

[The Yuckinator]

It says "Romans Go Home"

Re:Core pron

[Anonymous Coward]

Ceiling cat?

Re:Core pron

[Daimanta]

"I came.

I saw."

iPhone

Re:Core pron

[m.ducharme]

I came, I drank wine, I threw up.

Re:Core pron

[Veni Vidi Dormi]

I came.

I saw.

I slept.

Re:Core pron

[Xocet_00]

Burma Shave.

Re:Core pron

[MistrBlank]

I came.

I saw.

I Konquered.

There, fixed it for you

Re:Core pron

[Anonymous Coward]

I lost the game.

Re:Core pron

[Anonymous Coward]

I will be able to use it. (I go there) :)
I came. I wrote. I executed.

Re:Core pron

[Anonymous Coward]

I thought it was:

I saw.
(Then)
I came.

Re:Core pron

[Anonymous Coward]

I like how this was modded informative.

Re:Core pron

[Quicksilver_Johny]

Domus? Nominative? But "go home", it is motion towards, isn't it, boy?
Dative, sir!
[knife on throat]
No, not dative! Not the dative, sir! No! The... accusative, accusative! Domum, sir, ad domum!
Except that domus takes the...?
The locative, sir!
Which is?
Domum!

Re:Core pron

[Anonymous Coward]

Coregasm?

Re:Core pron

[Anonymous Coward]

I came.

I saw.

I compiled.

I thought it was "I kicked its ass"

Re:Core pron

[Tsaroth]

Ghostbusters, FTW!

Re:Vista capable?

[clang_jangle]

Yes, but I think you'll still have to disable the aero if you want to get DNF to work right.

iPhone

[giorgist]

When can I get an iphone with it ?

G

Re:iPhone

[neuromancer23]

You can have an iBone

PPC Linux

[Doc Ruby]

I'd be a lot more excited about these PPC lines if Ubuntu 8.04 would install and run properly on the PS3, whose PPC+6xDSP architecture would be a great entry level platform for coming up with parallel techniques for the bigger and more parallel PPC chips.

Re:PPC Linux

[rbanffy]

The problem is Sony cripled the environment in ways that make it very hard to use a PS3 as a computer.

I still think one could build a cheap computer with a Cell processor and make a decent profit. Those über-DSPs could do a whole lot to help the relatively puny PPC cores and having such a box readily available would foster a lot of research with asymmetric multi-processors. It's really sad to see future compsci graduates who never really used anything not descending from an IBM 5150

That said, I think there is some interesting stuff coming to the x86 world. That Larrabee x86 thing Intel is readying could be very interesting in itself and even generate some more interesting spin-offs. Imagine having a couple of cores that could, in an emergency, run the same binaries but that were tuned for different applications. One out-of-order core plus 4 in-order multi-threading cores would make a very interesting desktop processor.

Re:PPC Linux

[Gothmolly]

"In an emergency" ??? This is not LaForge rerouting plasma conduits to the warp drive - PCs do not rewire themselves on the fly in the presence of 'danger'.

Re:PPC Linux

[Anonymous Coward]

Yeah, because we all know that Ubuntu is the epitome of supercomputer class linux distributions.

Re:PPC Linux

[ianare]

Try yellow dog linux. And yes they can cluster them.

Re:PPC Linux

[Doc Ruby]

No, that is not a problem. Linux, including Ubuntu, has been running on PS3 since the PS3 was released. Every Ubuntu since 6.10 has run on it. And current releases of other Linux (PPC) distros usually do install, especially the Yellow Dog that is the one officially supported by Sony. But the problem is that the Ubuntu team has too few developers, and new features in Ubuntu releases break the installation in ways that the small Cell/PPC team can't keep up with.

Also, there's nothing really "puny" about the Cell's PPC core, which is a 3.2GHz dual-hyperthread Power core.

The Cell/Linux platform has already got video drivers that offload graphics from the PPC to the DSPs the same way most distros run graphics on separate VGA chips. It's a little buggy, in beta, but that's why the project just needs more developers. Not more FUD.

I don't think you really know anything about how Linux actually runs on the Cell, on the PS3. I think you're just repeating the most whiny posts about it you've heard. Because the reality is very different from what you describe, even if it's still got problems. Problems that don't require waiting for more x86 HW revisions, but rather just a little more work on the Cell Linux that Ubuntu is releasing.

Re:PPC Linux

[Doc Ruby]

Because of course I said that Ubuntu is the epitome of supercomputer class Linux distributions.

Funny how I thought I said that Ubuntu/PS3 is a great entry level platform for parallel processing development. Of course the Anonymous liar Coward is correct.

Re:PPC Linux

[Doc Ruby]

Well, I'd rather see the solutions that YDL develops make it into Ubuntu, which is overall a superior "Desktop" distro. Especially for media playing. And with its Debian-derived APT, it's overall a much easier system to maintain.

Re:PPC Linux

[Anonymous Coward]

It's really sad to see future compsci graduates who never really used anything not descending from an IBM 5150

Always thought that was an appropriate model number. You can label me 5150 if you like, but IMO if Linux ever gains dominance in the OS market, look for x86 to finally start fading.

Re:PPC Linux

[Anonymous Coward]

Also PPC optimizations on the gcc is the suxxorz.

Re:PPC Linux

[zsau]

It's really sad to see future compsci graduates who never really used anything not descending from an IBM 5150

Not really. There's enough higher level stuff lots of people will see no reason to use other tools. I have a comp. sci. degree I got without needing to care what architecture the computers I was using used. Sure I got a theoretical understanding of the way computers work and we did a bit of programming in a made-up assembly language, but really? all that low level stuff? means nothing to me. In any case, anyone who really is interested in the differences between x86 and other machines will learn it anyway.

(I did, however do much of my programming on a 64-bit PPC running 32-bit Debian, but as long as I coded standards-compliant C I never had an issue with compiling it on my uni's x86 servers for submissions.)

Re:PPC Linux

[Anonymous Coward]

The techniques required for successful SPE utilization have very little in common with the techniques required for a multi-core PPC processor.

Re:PPC Linux

[Ilgaz]

Lets not forget 1 fact, Ubuntu guys dropped official PPC support all by themselves stating Apple's switch to Intel. There is no "Evil Sony" intervention there. Zero. You drop a fairly modern architecture which is supported by the vendor of OS X, the "evil" Apple in Leopard all by your choice.

https://wiki.ubuntu.com/PowerPCReview

It can't get more developers or even more users because of that simple fact: There is no Ubuntu official support for PPC. FUD will end once the decision makers say they made a MISTAKE on that very same page and apologise PowerPC users like me.

I was not very interested in Linux as a Quad G5 OS X user but I remember the lack of logic in that decision made me nuts and smile same time.

Yellow Dog Linux on Cell really works fine, fine enough that people can even run Java applets inside their browser or run some sort of Flash (Gnash).

I don't think actual POWER users are remotely interested in Ubuntu like distros but Ubuntu people should see that the PowerPC is not only Apple. People saying "Altivec optimisation is pointless, look Apple dropped PPC" should also remember even that monster POWER processor will have Altivec.

Re:PPC Linux

[Anonymous Coward]

So, is this the first step on our way from "But, does it run Linux" to "But, does it run Ubuntu"?

This is worrying in a way.

Re:PPC Linux

[Anonymous Coward]

The PPE uses a simpler, in-order architecture, so it will be a lot slower than a POWER4-7 running at the same speed. But yeah, plenty fast for most uses.

Running your graphics on the SPEs means they're not available for other tasks, and IIRC there's a big bottleneck when transferring the graphics into the video memory at least on high resolutions.

Still, if one really wants to run Linux on PPC something like Terra Soft's PowerStation is a far better option. If you really want to play with Cell, bung in one of those Toshiba 4-SPE Cell cards and go to town.

Re:PPC Linux

[sanosuke001]

Actually, Linux does run quite well. However, it needs to be stripped down for the small amount of system/video RAM the PS3 has.

Re:PPC Linux

[Doc Ruby]

No, actually, Ubuntu has problems running its installer in the current version, 8.04.

I really wish people who just read old articles and half-read my posts would stay in "read-only" mode until they actually knew what they're talking about. It's distracting.

Re:PPC Linux

[Doc Ruby]

Yes, the Cell's PPE is slower than a POWER4-7, but it's not a POWER4-7, so I don't expect it to be as fast as that. It is, however, just as fast as a PowerPC running at the same 3.2GHz, including the AltiVec FPU.

Running the graphics on the SPEs does not make them unavailable for other tasks - it just uses some of their cycles. They're plenty fast enough to do lots of other tasks. The Cell's PPE + SPEs are still much faster than a PPC, or a Pentium4, plus a VGA card. The best use of a PS3 is to develop parallel processing techniques that efficiently share the SPEs as an on-demand resource. There's plenty of crunching power to go around. In fact, one simple way to use them is to compartmentalize each of the 6 SPEs to a single task, so there's six little screamers, five of which wouldn't be consumed by the graphics load if only one is dedicated to it.

Oh, and the Cell's main architectural advantage is its on-chip bus between PPE, SPEs, and RAM (including the framebufer, even the offchip VRAM) that is so fast and coherent that it's effectively "invisible", without slowing those fast SPEs down.

If you want to "play" with a Cell, the $400 PS3 is a lot easier to "bung in" than any of the dedicated Cell workstations that cost $10,000 or more.

Linix on PS3 is so cheap and easy to get started with that just reading articles about it is no substitute before people just go on posting about it as if they know what it's about.

Re:PPC Linux

[Doc Ruby]

No one said "evil Sony" until you did.

And what's stopping you from being part of the community support for PPC Linux on Cell? Just because neither Sony nor Ubuntu.com itself maintains Ubuntu for the PS3 doesn't really mean much. Most PC vendors (with any CPU) don't support Linux themselves. And lots of distro teams don't directly support all the architectures their distros are run on or get ported to. The whole point of Linux as FOSS is that people can do with it what they want without "official" support. There is indeed a community of kernel developers and distro packagers, as well as a larger community of "bleeding edge users" who all just want Ubuntu running well on their PS3, and aren't just looking to tag along with someone else's "official" development. There's nothing magic about Sony or Ubuntu.com development.

Ubuntu on PS3 runs pretty well already. Including running Java applets and Flash in a browser, and practically all the other workstation apps. And with the SPE video drivers I can already watch HD movies direct to my giant TV over the HDMI cable. Ubuntu/PS3 is nearly the best media station for my home network, and it's still only in beta.

With a few more good developers more interested in results than in being "official", Ubuntu on PS3 will be the platform of choice for people who want their media to "just play". If they don't wait for "official" support, which they don't need, that time will come soon enough.

Re:PPC Linux

[Anonymous Coward]

Most PowerPC chips are also out-of-order designs and would outperform the PPE at the same clock speed. The PPE was designed to be simple on purpose, in order to achieve a high clock frequency. Due to the nature of the code that will be running on Cell and Xenon, this is a perfectly acceptable compromise.

Re:PPC Linux

[Anonymous Coward]

Don't forget to look into Gentoo.

Re:PPC Linux

[ihuntrocks]

Actually, I've never had any issue with the available optimizations in the GCC for the PPC architecture. This may be, in part, due to my understanding of those features though. The PPC doesn't necessarily need all the "optimization features" that the x86 line has, partly due to being a more robust architecture. It's much like the "if it ain't broke, don't fix it" strategy. x86 just needs a little more help. It's more out of need that it has more optimization options than PPC than out of any failing of the PPC or the gcc team to provide optimizations for PPC.

Re:PPC Linux

[Anonymous Coward]

that's ubuntu's failing, clearly. IBM has been supporting SuSE and RHEL on power chips for a while now. These machines are powerhouses meant for real work, huge memory, huge core count and massive data backing them. You really shouldn't be using linux in the first place :)

Re:PPC Linux

[Doc Ruby]

Like any "Ubuntu failing", that's just a task for the community to fill, which is why I posted in public. Where people can see it and join the community. Rather than just whine about whose "fault" it might be. WHich is about as useful as saying someone shouldn't really be using Linux in the first place, when Linux is perfectly useful on these machines, including the PS3. As IBM's support on powerhouse machines amply demonstrates. And which shows how useful a Linux PS3 can be as an entry level, since the PS3's Cell is a powerhouse in its own right, much more than merely a PPC, even if the rest of the HW is too small for more than entry level work.

People have to start somewhere. Starting by upgrading a working Linux distro on a $400 working multimedia PC is a great place to start, even if it never leads to development on "powerhouse" machines like fully-equipped POWER or Cell machines.

Re:PPC Linux

[rbanffy]

"The Cell/Linux platform has already got video drivers that offload graphics from the PPC" ok. That's cool and a major progress.. Still, there is no access to the GPU because Sony decided there wouldn't.

BTW, can you upgrade the memory?

You do think it is a great computer. I say it could be an amazing computer if only Sony hadn't decided it would be a videogame console.

And no. I don't own a PS3. I considered one but the FUD, as you describe it, convinced it would not be the dream machine it could be.

Re:PPC Linux

[Doc Ruby]

The memory can't be upgraded, which limits its use for some apps. But not for media playback apps, especially streaming media. It's a great media console in the making.

The memory can be worked around with Flash drives mapped as swap, which is an adequate compromise to add general purpose tasks to its media player strengths. And there's probably a way to use a SATA-RAM device to actually add real RAM, though not anywhere full speed (but as fast as lots of PC memory). And real Cell processing on real datasets is only as fast as the SATA or Gb-e can deliver streams of data, so it's a barely adequate platform for real computation apps.

But all that leaves the PS3 a very interesting machine both for research and for real use for media. Which is why I would like to see Ubuntu installers work properly on it, so the rest of the projects don't get halted at that essential enabling stage. Even though there's a kernel and distro team working on that, their work would be even more productive if they could start with a working system, instead of spending time catching up to square one.

Re:PPC Linux

[Ilgaz]

I remember that time when OS X was 10.4.x level and we were getting some sort of clues about future of PPC on OS X. I thought I better try the most close distro to OS X and ended up with Ubuntu. Listening fans blow full speed (all 9 of them), I even did my best to report a kernel crash issue to an application developer. I also figured that time why Debian has become a legend since that Debian guy actually _cared_ for that bug report on such a weird machine (Quad G5) running unsupported Ubuntu (didn't know then) and actually fixed the issue even while I state "I can't run any kind of linux in this period to test".
I checked Ubuntu pages for support and saw that statement saying they drop official support because my vendor dropped PPC production.
I had to buy a commercial partition manager just to get rid of Ubuntu distro because having a video edit workstation, I really can't handle the downtime and amazing price of getting a liquid cooling system replaced by Apple service center.
I am not saying anything about PS3, I don't own a PS3 but that is what Ubuntu guys made me to lose my belief to Linux. Getting ignored at some PPC linux channels have also made situation worse. I ended up buying a Family license for Leopard and if Apple _really_ become a total asshole to their powerpc customers, I will buy Yellow Dog Linux. Yes, buy it.
I can't trust to Ubuntu after reading that page. I explained all details to you including fans, getting ignored, buying a partition manager just because it could be easier to understand my stance against them.

Re:PPC Linux

[Doc Ruby]

Well, that was when Ubuntu was still a lot less mature. The reason Ubuntu is so popular now is because those kinds of bugs are mostly gone (which is the best that anyone can say about any OS, including MacOSX). And of course since it's so much more popular, there are more people and bug reports making it better, faster.

In fact, I'm trying to figure out what to do with an iBook G4 (PPC) running MacOSX Jaguar, because the iPod I just plugged into it is demanding v10.4.8 or later. Apple seems to insist that I buy a v10.5.2 upgrade for the privilege of using the iPod I bought with the Mac I'd already bought. If I can't find a way to drag my CDs onto that iPod on that Mac without paying for the OS upgrade, then I'm thinking about installing Linux on it, if Linux will support that iPod use without buying extra SW. And I'm thinking that Linux will be Ubuntu, because Ubuntu has generally the easiest and most complete support for peripherals (like the iPod). And because I use Ubuntu on my other machines, so I have the simplicity of using a single distro for similar tasks rather than special cases on different machines for the same task.

I have a serious question:

[inject_hotmail.com]

Seriously:

I have a couple dual-core PCs. I notice that some won't ever use 100% CPU even though they easily could. I check "set affinity" in task manager, which says the process should use both cores...but it only ever hits 50% of total CPU. Looking at the CPU graph, it shows that as usage goes up on one core, usage goes down on the other.

Is there any way to force a process to run over 2 cores at 100%?

If not...how would 300,000 cores help unless you are running 300,000 processes, or an app that you know will scale over that many cores?

The preceding was in fact a serious question.

Re:I have a serious question:

[Anonymous Coward]

The applications that are going to be run on this type of machine are designed to be run on this kind of machine.

Re:I have a serious question:

[badboy_tw2002]

If your process looks like this:

int main()
{

while (something)
{
      doSometing();
}

}

It will hit 100% on one core and that's it. Its not multithreaded - one CPU will churn on it forever and the others will sit around waiting for a task from the OS. 2 course, 200,000 cores the results will be the same. These machines are made for tasks that are broken up into lots of smaller jobs and processed individually. Its not magic - more cores won't get a single threaded process done faster.

Seriously.

Re:I have a serious question:

[Tom9729]

Seriously:

I have a couple dual-core PCs. I notice that some won't ever use 100% CPU even though they easily could. I check "set affinity" in task manager, which says the process should use both cores...but it only ever hits 50% of total CPU. Looking at the CPU graph, it shows that as usage goes up on one core, usage goes down on the other.

Is there any way to force a process to run over 2 cores at 100%?

If not...how would 300,000 cores help unless you are running 300,000 processes, or an app that you know will scale over that many cores?

The preceding was in fact a serious question.

My guess is that the Windows scheduler is doing it's job and preventing your processes from eating the CPU.

Have you tried playing with process priorities?

Re:I have a serious question:

[encoderer]

You figured it out: it's all in the programming.

Multi-threading is difficult, and until just recently, it was a niche market. Most people just didn't have dual processors.

That'll change. Until then, you're still benefiting by running OS and background processes on one core and the foreground process on another.

Re:I have a serious question:

[Anonymous Coward]

Try here. And here.

Re:I have a serious question:

[QuantumRiff]

Yes, run a multithreaded app on the system, or run two single threaded apps. (photoshop is multithreaded, as are many rendering software, etc) Not much software is yet multi-threaded, at least not on windows.

Re:I have a serious question:

[inject_hotmail.com]

The applications that are going to be run on this type of machine are designed to be run on this kind of machine.

Thanks for the reply...

I posted so quickly, I should have written "...some processes won't ever..."

Anyway...yeah, it's hit and miss on the processes that do this. I guess it's up to the author of the app to make it multi-core friendly.

Re:I have a serious question:

[WilliamBaughman]

If not...how would 300,000 cores help unless you are running 300,000 processes, or an app that you know will scale over that many cores?

The preceding was in fact a serious question.

Having 300,000 cores wouldn't help if you didn't have enough cores. However, University of Illinois probably won't be using it to run one instance of McAfee and one instance of Word. Chances are, they'll be using it for meteorological simulations.

Re:I have a serious question:

[inject_hotmail.com]

If your process looks like this:

int main()
{

while (something)
{

      doSometing();
}

}

It will hit 100% on one core and that's it. Its not multithreaded - one CPU will churn on it forever and the others will sit around waiting for a task from the OS. 2 course, 200,000 cores the results will be the same. These machines are made for tasks that are broken up into lots of smaller jobs and processed individually. Its not magic - more cores won't get a single threaded process done faster.

Seriously.

Thanks for the quick reply.

I posted so quickly, I should have written "...some processes won't ever..."

Aren't a lot of games and apps single-threaded? Hmmm. I figured that dual/quad-core wasn't all it's cracked up to be. So, essentially, if I have a single-threaded app on a quad core, it'll perform at 1/4th the potential speed.

That doesn't leave me with a warm and fuzzy feeling inside.

The funny thing is that it teeter-totters back and forth from one core to the other. I wish I knew what made it do that.

Re:I have a serious question:

[friedman101]

you're assuming that doSomething() is entirely unparallelizable and that every line depends on the previous line.... not the case for generic code. a smart compiler or informed programmer can make even the most trivial programs work on multiple cores.

Re:I have a serious question:

[WilliamBaughman]

Having 300,000 cores wouldn't help if you didn't have enough cores. However, University of Illinois probably won't be using it to run one instance of McAfee and one instance of Word. Chances are, they'll be using it for meteorological simulations.

Sorry, that should read 'Having 300,000 cores wouldn't help if you weren't running enough processes.'

Re:I have a serious question:

[inject_hotmail.com]

My guess is that the Windows scheduler is doing it's job and preventing your processes from eating the CPU.

Have you tried playing with process priorities?

I posted so quickly, I should have written "some processes won't ever..."

Yep, that's the "set affinity" part. And that's what doesn't make sense. If I 'set affinity' to both CPUs on a dual-core system, I say it should max 'em both out...but it never does.

Re:I have a serious question:

[tfranzese]

The problem is that software is ultimately limited by how much of the algorithm is sequential. You cannot parallelize (I'll make that word up if I have to) every portion of software, there is always some sequential portion of code that cannot be split up that will prevent that. Certain software is limited more than others. It is up to the software developer to design their algorithm to exploit as much parallelism as possible be it by running routines that can be done concurrently into two threads to a point where they must communicate. The applications typical on a desktop environment have not always been designed with this in mind, and at times it can be a challenge. Now, there are other classes of problems unlike those typically seen on the desktop. These problems are what's called embarrassingly parallel. A lot of scientific problems fall under this category and can reap huge rewards out of a cluster. If the software is largely sequential (either by design or necessity), the process is unlikely to see any benefit from multiple cores and is at the mercy of the operating system's scheduler as far as I know. I think this gives some idea, but I've only had one course in parallel computing and it's been a while. It's pretty interesting stuff, and hopefully I didn't butcher too much.

Re:I have a serious question:

[jcarkeys]

You sir, are correct. Most things aren't set to run in parallel and thus don't get the gains. Gains come from optimized code (obviously), but also doing multiple tasks. Leave Photoshop to render an HDR image and play a game, if you want. Though to be fair, it's not "just 1/4th" performance, the other cores are able to handle some of the other CPU tasks, such as running hardware controllers.

And the reason that it kind've oscillates between cores is because "Set Affinity" tells the process that it's allowed to use that core, not that it has to or even should. If you want something to use both cores, open up two processes, set the first to core 1 and the second to core 2. Most of the time that's unusable like that, but I recently transcoded my entire music library and set one process to do songs from A-M, and the other from N-Z. It really helped

Re:I have a serious question:

[Firehed]

It's not that simple. While one single task generally is not coded to take advantage of the entire system (single threaded on a dual system, dual thread on a quad system, whatever), you are able to actually use your computer while said task is underway. Ever encoded a DVD on a single core machine? Not so fun - half the time, you can't even use your mouse. Slap the same task on a dual-core box, and suddenly you can continue to work (or play) while that goes on in the background. Alternately, you can encode two DVDs simultaneously and be done in the speed it would normally take to finish one. Parallelism in its most literal sense.

Of course, many video-related apps these days are multi-threaded, but you get the general idea.

Re:I have a serious question:

[inject_hotmail.com]

Having 300,000 cores wouldn't help if you didn't have enough cores. However, University of Illinois probably won't be using it to run one instance of McAfee and one instance of Word. Chances are, they'll be using it for meteorological simulations.

Sorry, that should read 'Having 300,000 cores wouldn't help if you weren't running enough processes.'

So if this thing is 4GHz, 8 cores would mean 500MHz per? Honestly, it doesn't sound so appealing to me knowing that some apps won't ever use 4GHz. Scaling it up, I hear that Intel plans to have 1000s of cores on a CPU...does this mean ultra slow apps?

Re:I have a serious question:

[tfranzese]

I should add/correct myself that the designer should not limit themselves to a particular number of threads. If the algorithm can only be made concurrent with two threads, that may be the case, but generally if the problem can utilize all 99 to get the problem done in a reasonably more efficient time, then it should utilize them.

That said, and this is somewhat off topic, I think it would be good (if it hasn't already been done) to allow the user to limit the number of cores the system uses or customize the power options so that they can choose when to bring certain cores out of sleep for the OS to utilize. I hate wasting power when all I want to do is browse the web when I'm not busy with more demanding tasks.

Re:I have a serious question:

[tchuladdiass]

The funny thing is that it teeter-totters back and forth from one core to the other. I wish I knew what made it do that.

The OS runs the process a few milliseconds at a time, then kicks the process of the cpu for another process to run (if there is one, including OS tasks such as I/O routines). When the OS starts up the process again for a few more milliseconds, it may start it up on a different core. That is why both cores will show 50% average utilization.

Now if you set CPU affinity for that process to be on one core, then it will max that core out at 100% and the other core will be idle. This may result in better performance, because you get better cache utilization if the process stays on the same core.

On a related topic, this can also be the case if the app is multithreaded -- sometimes it is more efficient to run multiple threads on the same CPU instead of across CPUs, if each thread is accessing the same region of memory. Otherwise, if the threads are on different CPUs or cores, then the threads are constantly invalidating the cache on the other core, causing more (expensive) reads/writes to main memory.

Re:I have a serious question:

[KillerBob]

On the other hand... if your process looks like this:

int main() {
    while (TRUE) {
        fork();
    }
}

It might do a pretty good job of tying up resources. :)

Re:I have a serious question:

[Annymouse Cowherd]

No, each core is running at 4Ghz. That does not total up to 16 Ghz processing power though, because only multithreaded programs can take advantage of more than one core at once, and they still have to wait if they're sharing data.

Re:I have a serious question:

[vux984]

Aren't a lot of games and apps single-threaded? Hmmm. I figured that dual/quad-core wasn't all it's cracked up to be. So, essentially, if I have a single-threaded app on a quad core, it'll perform at 1/4th the potential speed.

Yes, although, most high end games and game engines actually are multi-threaded. Few are designed to take advantage of more than 2 cores though, and none that I know of will use 8 or 300,000...

So, essentially, if I have a single-threaded app on a quad core, it'll perform at 1/4th the potential speed.

Not necessarily. If you have 3 women can you make a baby in 3 months instead of 9? Given that it still takes 9 months and 2 of the women are idle, would you say that these women are performing at 1/3rd the potential speed? Same sort of logic applies here. If the task is inherently sequential, having more cores (or ladies) won't make it any faster.

Somethings -are- highly parellizable, like ray-tracing or cutting down all the trees in a forest.. and other things are partly parallelizable... like changing tires (a pit crew can change 4 tires at once... but adding more staff to allow you to change 5 tires at once doesn't make your team any faster...)

That doesn't leave me with a warm and fuzzy feeling inside.

Yes, in general computing applications, an 8GHz CPU would be faster than a quad core 2GHz. (And even under optimal parallilizable situations the 2ghz quadcore would just barely surpass the 8ghz cpu due to lower task switching overhead.) So the faster single cpu is almost always better. The reason we have quad core 2Ghz cpus is that they are much much more practical to actually make, and a lot of the stuff that takes a long time (rendering 3d, encoding movies, etc is actually highly parellizable so we do see a benefit. And much of the single threaded sequential stuff we see is waiting on hard drive performance, network bandedith, or user input... so cpu isn't the bottleneck there anyway.

The funny thing is that it teeter-totters back and forth from one core to the other. I wish I knew what made it do that.

If you look at task manager, there what? some 40+ processes running. The OS rotates them onto an off of the 2 cores based on what they all need in terms of cpu time. So your 'cpu heavy task' gets pulled off a core to give another task a timeslice, and then once its off, it can be scheduled back onto either core. Ideally should stay on one core to maximize level one cache hits, etc, but if its been off the core long enough for the other processes to cache all new memory it doesn't really matter which one it gets assigned to, and in any case flipping from one to the other every now and then makes a almost immeasurably small performance difference.

btw - the 'set processor affinity' feature tells the OS that you really want this process to run on a given cpu/core, instead of hopping around. But in most cases its not something one needs (or gains any benefit) from doing.

Re:I have a serious question:

[prod-you]

AFAIK, set affinity only means that it'll only use that one CPU. Theoretically, with an intense program, that cpu should get to 100% and leave the other one available for other things.

In windows though, a multicore processor will try to split workload by threads, not processes, so even a single program can take advantage of multiple cores; especially if multiple threads do heavy lifting.

Re:I have a serious question:

[Anonymous Coward]

If you're looking for 'common user' cases of multithreaded applicaitons, I'd say take a look at some of the video conversion. I don't believe that editing videos is very commonplace, but with all the ipod videos and other portable video players around these days people are certainly going to look for a way to put that xvid or h264 video they got off a torrent onto their ipod. This usually entails reencoding and there are various programs, free and nonfree that combine all the necessities into an easy to use gui that actually churns out a .mp4 or m4v file for the ipod or psp. Some use the x264 codec which is multithreaded, so before anyone discounts multithreading for lack of 'common user' uses think of all the ipods and psps out there.

Re:I have a serious question:

[modmans2ndcoming]

it will get you multiple single threaded processes done faster though.

Re:I have a serious question:

[Zeussy]

Or, the other thing I like about dual core and up boxes is that they appear to be more stable, back on a single core machine, when a process really wanted to lock up, in some mysterious while(1); loop it could be a real try of patience to kill the app. On a dual core machine no worries, still got the other core to save yourself with :)

Re:I have a serious question:

[mikael]

You would have to have all your programs to take advantage of multi-threading ie. pthreads

With multi-threading, there are different ways of improving the performance of applications. The simple way is to have separate threads for user input, background tasks, and display.

The more complex way is to replace each single function that would run on a CPU (ie. color adjusting an image) to run on multiple cores/threads. For any function, the application creates a whole series of tasks which are retrieved and processed by separate threads. For image processing, this would involve splitting the processing of the image into many subimages - each thread processes a particular chunk of the main image.

For these kind of supercomputers, the users are going to have massive 3D grid volumes (eg. 2048^3 or greater) with each core/thread processing a particular sub-volume. Calculations like CFD can require a good many passes to run a single time-step of the simulation. And both a current generation grid and a next generation grid need to be stored. All this information needs to be processed so that it can be rendered in real-time).

Re:I have a serious question:

[inject_hotmail.com]

The OS runs the process a few milliseconds at a time, then kicks the process of the cpu for another process to run (if there is one, including OS tasks such as I/O routines). When the OS starts up the process again for a few more milliseconds, it may start it up on a different core. That is why both cores will show 50% average utilization.

Now if you set CPU affinity for that process to be on one core, then it will max that core out at 100% and the other core will be idle. This may result in better performance, because you get better cache utilization if the process stays on the same core.

On a related topic, this can also be the case if the app is multithreaded -- sometimes it is more efficient to run multiple threads on the same CPU instead of across CPUs, if each thread is accessing the same region of memory. Otherwise, if the threads are on different CPUs or cores, then the threads are constantly invalidating the cache on the other core, causing more (expensive) reads/writes to main memory.

Quite informative -- thanks for the insight. I thought it was the OS, but didn't know those specifics. I had no idea that maxing out one core would be more efficient though.

Re:I have a serious question:

[ZuggZugg]

The "problem" you are noticing is that most software is not programmed to take advantage of multiple execution cores.

The problem in a nutshell is that writing parallel execution routines in software is not trivial.

What you point out is exactly the problem that many have been "freaking out" about for a while. That multi-core is all fine and dandy for workloads that can leverage parallelism. But for a lot of applications this is very difficult to accomplish.

In the case of this "computer" at this university, it's likely a number crunching "computer" or supercomputer. Very likely to be just a gang of machines networked together to process ridiculously parallel problems.

Not something you'll ever boot Vista on and expect to run Half Life any faster on...

Re:I have a serious question:

[NerveGas]

Only if it's written for it.

I have a Q6600 (quad core), and regularly play a video game on one monitor and an H.264-encoded movie on the other. Between the two, it will keep all four cores at 100%.

Re:I have a serious question:

[djikster]

If on the other hand, you have stuff like:

#pragma omp parallel for for (int i = 0; i

... the code would actually use as many cores as you can throw at it. (Assuming gcc 4.2+ with -fopenmp)

Re:I have a serious question:

[Anonymous Coward]

Seriously:

I have a couple dual-core PCs. I notice that some won't ever use 100% CPU even though they easily could. I check "set affinity" in task manager, which says the process should use both cores...but it only ever hits 50% of total CPU. Looking at the CPU graph, it shows that as usage goes up on one core, usage goes down on the other.

Is there any way to force a process to run over 2 cores at 100%?

Yes. By writing the application using parallel processing technologies and techniques. (MS Notepad = not so much, Seti@Home = very much so)

If not...how would 300,000 cores help unless you are running 300,000 processes, or an app that you know will scale over that many cores?

You have answered your own question. Only by having multiple execution contexts (e.g., threads, processes, fibers, vectors, etc.) can you utilize multiple executing cores. For examples of applications that can utilize this system, and for applications that this system will be built for, see the proceedings of the IEEE/ACM Supercomputing Conference (www.supercomp.org).

The preceding was in fact a serious question.

And a good one. It is important to remember that this system is *not* for running your desktop, but for solving scientific problems that cannot be solved (in a practical timeframe) using existing systems.

Another serious question is: "How much power (i.e., electricity) will this system consume for both computational *and* cooling components?" A Power6 gets about 100MFLOPS/Watt (www.green500.org); so assuming a Power7 is 4x more efficient with 2x electricity redundancy, this machine will need 50MW!

Re:I have a serious question:

[rbanffy]

"Aren't a lot of games and apps single-threaded?"

And that's one more thing we can thank Microsoft for.

Hadn't DOS and the PC-clones crippled with mono-processor/mono-threading DOS/Windows stack become the dominant architecture for most of the 90s, we would have rock-solid, secure, multi-processor, 64-bit RISC boxes running some flavor of Unix on our desktops by now.

Thanks Bill.

Re:I have a serious question:

[inject_hotmail.com]

No, each core is running at 4Ghz. That does not total up to 16 Ghz processing power though, because only multithreaded programs can take advantage of more than one core at once, and they still have to wait if they're sharing data.

Ah! Really...that's good then.

On a side note, I notice that shady computer vendors like adding GHz on multi-core PCs in their marketing. They also claimed their computers had a 1GB drive when those drives weren't even available (probably 2x500GB or 4x250GB refurbs in RAID...). MDG was doing it a while ago.

I think they got sued or something, because they stopped doing that.

Re:I have a serious question:

[forkazoo]

Is there any way to force a process to run over 2 cores at 100%?

No. Some apps are only trying to do one thing at a time. Giving them the resources to do two things at once won't change that. It'd be like giving you two cars and telling you to get someplace twice as fast.

If not...how would 300,000 cores help unless you are running 300,000 processes, or an app that you know will scale over that many cores?

The preceding was in fact a serious question.

You only bother to buy a beast like that if you have code that will scale to that many cores. That's easy because you won't be using just off the shelf code. A lot of the code that runs on that machine probably won't get run on anything else unmodified. When you are writing your code for a specific machine, it's more natural that you would take care to make it run well on that machine.

Re:I have a serious question:

[amirulbahr]

If you run two of these processes on a dual core machine then each will get 100% of each core.

Re:I have a serious question:

[inject_hotmail.com]

Oh man...don't get me started. ;)

I 100% agree with you on that one. RISC processors SHOULD have taken off like wildfire. As an example, the low clocked original PlayStation was 33MHz RISC (with what...8MB RAM?), but it out-performed anything around. Imagine a 3GHz RISC on the desktop!

Ha ha! Jokes on us. CISC won.

I wonder if IA64 took off, how we'd be doing...probably not too much better I think. AMD64 has been around for how long, and we're still in 32bit mode! AFAIK, all it would take is a recompile, and any given app would run in 64bit mode natively, right? This is inertia from companies that care not, but for profits.

I think we need to thank all the vendors out there for this one, because it has taken 4 years to get any sort of coherent 64bit driver support. 64bit games -- what a joke! 64bit support is still terrible, as far as Joe Sixpack is concerned.

A few years ago I built a PC for a friend with a 64bit Intel loaded with 64bit WinXP. He formatted and reinstalled 32bit because nothing worked properly. He's still in 32bit mode. Why? I am too! Why?!? I feel like we're just heading into the 90's now.

Re:I have a serious question:

[Detritus]

Single threading, like used on old versions of Windows, does have some advantages. It avoids a lot of concurrency related problems that most programmers are not properly trained to deal with. If everyone follow the rules, it's efficient and performs well.

I was recently reading a paper on multi-core processors and the future of programming. It pointed out that many multi-threaded programs work fine until they are run on a real multi-processor system where multiple threads can actually run simultaneously. At that point, strange timing-related bugs often appear that are very difficult to replicate and diagnose.

Re:I have a serious question:

[Anonymous Coward]

Many of the scientific parallel codes are not embarassingly parallel. Stuff like rendering independent frames of a movie is embarassingly parallel. The scientific codes have communication between processors throughout the run of the application, and hence are not embarassingly parallel. However, stuff like SETI at home is embarassingly parallel.

Re:I have a serious question:

[Skrapion]

Oh man, look at what you've done. Don't you know that guys, particularly geeky guys, love to explain things? Now somebody has to go through and mark half of these responses as redundant.

PS: Amdahl's Law predicts the theoretical maximum additional performance you can gain by adding more cores.

Re:I have a serious question:

[inject_hotmail.com]

Another serious question is: "How much power (i.e., electricity) will this system consume for both computational *and* cooling components?" A Power6 gets about 100MFLOPS/Watt (www.green500.org); so assuming a Power7 is 4x more efficient with 2x electricity redundancy, this machine will need 50MW!

Wow...I didn't even think of that. That's insane!

(looks over at BFG's website for a 50MW PSU)

They might have a hard time with that one. Mind you, I find it amazing that we're measuring PSUs in kilowatts now -- thousands of watts! I think my first computer had a 50watt PSU, max.

Re:I have a serious question:

[UncleTogie]

Call me crazy, but that code looks like it'd be royally forked. ;)

Not having seriously programmed since the TRS-80, I was curious... Jus' how hard IS it to develop decent multi-threaded/multi-core code....?

Re:I have a serious question:

[Hal_Porter]

Is that really true? I've got a dual core laptop and most of the time it's doing web/email type stuff. The only times I've ever hit 100% CPU were encoding video and playing games. And if you look at Performance Monitor, both of those ended up maxing out both cores. So it certainly seems that people that write code that will max out CPUs have done a decent job of trying to make their code use multiple cores. Maybe not a 32 thread system like this one, but certainly on a 2 core one.

And on a server it's probably even better. Serving HTTP or SMB should be embarassingly parallel since each users can have a server thread dedicated to them. In both cases I'd expect things to be IO limited, not parallelism limited.

Re:I have a serious question:

[inject_hotmail.com]

The "problem" you are noticing is that most software is not programmed to take advantage of multiple execution cores.

The problem in a nutshell is that writing parallel execution routines in software is not trivial.

What you point out is exactly the problem that many have been "freaking out" about for a while. That multi-core is all fine and dandy for workloads that can leverage parallelism. But for a lot of applications this is very difficult to accomplish.

In the case of this "computer" at this university, it's likely a number crunching "computer" or supercomputer. Very likely to be just a gang of machines networked together to process ridiculously parallel problems.

Not something you'll ever boot Vista on and expect to run Half Life any faster on...

Ah yes...but it still goes to how companies solve problems (overheating CPUs, GHz limitations) with non-solutions (2 cores @ half speed), cramming it down our throats ("surf the net! burn CDs! write your term paper! all at the same time! ! !) as wonderful, and then everyone wants it (try to find a new single-core CPU).

Now my stuff runs slower. Thanks. Intel says they plan to make CPUs with 100s or 1000s of cores...what for?! 2 is plenty...

Re:I have a serious question:

[JebusIsLord]

A couple things:

x86 chips today are 99% RISC-like (the term RISC is rarely uses today, since basically no modern CPUs are "pure RISC" in design (reduced as in not even having a multiply instruction, like older SPARCs). Sure the exposed architecture is ugly x86, but that's the compiler's job to worry about, not yours. It doesn't really affect the chip performance. Don't forget x86 chips are still the fastest out there, despite the weird interface)

Also, for Joe Sixpack, 64-bit is pointless - especially when the 32-bit version works on the same OS! If you recompile a program as 64-bit (and often that is all there is to it; a recompile), you'll notice that the binaries are larger. In fact, most pointers (memory addresses) now takes up twice the space, so your program also uses more memory. The benefit? Unless your app uses more than around 3GB of RAM, basically zero (On x86 there is a sometimes a slight performance benefit, not because 64-bits is "faster" or anything, but because AMD added some more registers to the x86-64 spec).

Anyhow, i generally view 64-bits as a waste of address space, UNLESS you're accessing large amounts of memory (>3GB per program!). This will be more of a concern in the next few years, but there isn't any rush. I use 64-bit Vista for development (Because I have 4GB of RAM) but otherwise probably wouldn't care. Even Visual Studio (the dev platform for 64-bit code) is mostly a 32-bit app, nor should they change it.

Re:I have a serious question:

[symbolset]

iometer

Properly configured it can stress all the cores on all the nodes in your cluster.

Oh you wanted to do something useful...

Intel released it as open source in 2001. Edit the source for the dynamo so that it does something useful. Compile and install. Done.

Or you could load Vista and play a light game. That ought to peg both cores.

Actually, dual core is what it's cracked up to be. While your single threaded application is grinding away you can still interact with your computer instead of staring at the hourglass like you used to do. Since you like playing with the affinity you can launch several long single threaded tasks and set their affinity for different cores. Transcode a .AVI into a DVD of the family picnic? Render an animation in POVRay? Compute a few billion prime numbers. Fold some proteins. Calculate the propagation of thermal energy through single fibers in a carbon-fiber fabric. Whatever you want.

Soon almost all non-trivial applications will be multithreaded, and then you'll be cursing the hourglass again. Until then enjoy your vacation from its tyranny.

Re:I have a serious question:

[ushimitsudoki]

AFAIK, all it would take is a recompile, and any given app would run in 64bit mode natively, right?

Not always. Some things rely on certain data types being of a certain size and will break if, say, an "int" is twice as long as expected.

Re:I have a serious question:

[Samah]

So of course what you mean is:

int main() {
for(int i=0;i<200000;i++) {
if(fork()==0) {
doSomething();
}
}
}

Re:I have a serious question:

[Anonymous Coward]

Q&D answer... Running simulations with random variables. You run the same sim on many processors with random variable generation and then statistically take the outputs to determine the trend. Had 100 processors running the same sim (96 hours per run) gave me one data point... with 3000 data points at once problem solved in a week instead of a year...

Re:I have a serious question:

[Anonymous Coward]

Looks like the problem is a crap compiler to me.

Re:I have a serious question:

[Gumbercules!!]

I have to thank you.

That was the best explanation of multi core processing I have ever read.

Re:I have a serious question:

[Firehed]

Well yeah, if you have a habit of using unstable apps ;) Gotta keep an eye out for the threaded while(1); though, that one will really kick your ass.

(not that I should talk, having locked up my machine a couple times with a bad loop trying to write a few lines of code to parse through 2.5gb of sql with some ungodly number of rows that simply can't be restored as one on any machine in the building)

Re:I have a serious question:

[Anonymous Coward]

If the task is inherently sequential, having more cores (or ladies) won't make it any faster.

There has just got to be a joke in here somewhere

Re:I have a serious question:

[Nyall]

Lol. I love literalists.
He hasn't designed an OS? Or studied it in a class?

Maybe he was just giving a very simplistic explanation to try and answer the question at hand?

But oh noes, not enough detail was provided.

Re:I have a serious question:

[jd]

Assuming you have a multithreaded or otherwise parallel program, you would probably start with a profiler for parallel programs. DAKOTA and KOJAK are two open source profilers for just that sort of work. If the program isn't truly parallel, it won't run in parallel, whatever you do.

Re:I have a serious question:

[azrider]

Jus' how hard IS it to develop decent multi-threaded/multi-core code....?

Not hard at all, as long as the problem you are trying to solve can be broken down into parallel steps.
OTOH, if the problem requires sequential steps, you are fsck'd

Re:I have a serious question:

[asc99c]

A lot of application and games writers are complaining bitterly about the move to multi-core processing, as it does mean you need to change the way the code is written to take advantage of it.

I write stuff that runs on big UNIX boxes that has been necessarily multi-process for a long time. It's just a matter of finding things that can be done independantly and then explicitly putting them in their own process.

Ideally languages and compilers will do this at some point but so far mainstream languages do not. Also when you're doing desktop GUI apps it's often tricky to do a good job of multi-processing, and the GUI toolkits don't yet do much to help.

Re:I have a serious question:

[pimpimpim]

Not necessarily. If you have 3 women can you make a baby in 3 months instead of 9? Given that it still takes 9 months and 2 of the women are idle, would you say that these women are performing at 1/3rd the potential speed?

I don't know about the specifications of your reproductive system, but I could make 3 babies with 3 women in 9 months with no physiological problems (the psychological effect will be different though).

What you mean is that a woman will take at least 27 months to make 3 babies from 3 men. (for the sake of simplicity we don't count twins or more here)

Why do you want to do this in the first place, are you planning a sitcom?

Re:I have a serious question:

[Fweeky]

The benefit? Unless your app uses more than around 3GB of RAM, basically zero

Plenty of things quite enjoy being able to perform operations in 64bits at a time, actually. Especially when it comes to media, crypto, compression, and indeed games; on top of having 2-3x the usable number of general purpose registers, which certainly isn't something to sneer at given how awful x86 has traditionally been in this area. Plenty of things you're likely to actually care about the performance of are likely to get a nice boost.

64-bits as a waste of address space, UNLESS you're accessing large amounts of memory (>3GB per program!)

Well, you generally only get 3GB when you've performed tricks to ask the OS to allow that; e.g. /3GB boot flag, fiddling with MAXDSIZ, or recompiling with a different user/kernel space split.

On top of that, it's not all about RAM, it's about address space; if you've only got 32 bits to play with, you need to be very careful about allocating it, since any wastage can lead to exhausting your virtual address space before your physical space; like with filesystems, fragmentation becomes more of a concern the closer you get to your maximum capacity.

Large virtual spaces are also useful when it comes to doing things like mmapping large files; for instance, a database might like to mmap data files to avoid unnecessary copying you get with read()/write(), but mmapping a 1GB file means you need 1GB of address space, even if you don't touch any of it. When it's common to access disk using memory addresses, 3GB starts looking small very fast.

You also very quickly eat into it using modern graphics cards; 512MB is common, having two isn't that uncommon, and things are moving towards 1GB; bang goes your 3GB, all that frame buffer needs addressing too, on top of the kernel's other needs.

Really, 32bit needs to die screaming, sooner rather than later.

Re:I have a serious question:

[Z34107]

You can't force a process to run "100%" on multiple cores if it only has one thread (as most do.)

If you have a dual-core processor and a two-threaded process, those threads can't be interdependent. And then, they have to be doing work "hard" enough to tax both cores. And that work can't be constrained by other devices - no waiting for your disk buffer to be filled, a network packet, etc.

The other part of the problem is that Windows' task scheduler doesn't let programs monopolize the CPU. If it finds a mostly non-interactive program that's a processor-cycle black hole, it's priority will get downgraded. The closest thing you can do to peg your cpu is write a for(;;) loop at "realtime" priority

Re:I have a serious question:

[KGIII]

Actually I am glad you asked. I didn't answer a single question of yours but I got to remember why I enjoyed slashdot while reading the responses so thank you. My comp-sci was 12 years ago so, well, I don't even pretend to know this stuff. Your asking led to my reading and learning. Thanks.

Re:I have a serious question:

[Macman408]

Furthermore, a single application need not use all 300,000 cores. Like your PC, the supercomputer can run multiple applications at once, using as many or few threads as each one needs. Or as many or as few threads as the researcher was able to bargain for.

That said, there's a lot of work to help your everyday PC programmers write programs that are designed to be multithreaded, too. It seems like every week there's a new library or technology to enable multithreading. Personally, I've used pthreads, OpenMP, MPI, TBB, MapReduce, and LogTM (Transactional Memory, which is a completely different paradigm from what people are used to). And now Apple has announced their Grand Central technology (whatever that may be), in addition to existing methods such as NSThreads.

Someday, one or a few of these will really win out (hopefully it's the one that makes multithreading easy and effective, though such a library has yet to be developed). Or, the multicore phenomenon may die down a bit - if you look at other parts of your computer (like the memory or I/O busses), they tend to be serial at first, then when serial can't be made much faster easily, they go parallel with similar technology. Eventually, somebody figures out that you can do something radically different, and get a really fast serial link (think ATA to SATA here). Next, it becomes feasible to make that really fast serial link parallel again. This generalizes a bit, and multicore processing seems to be very early in its lifetime yet, but failing to develop a good parallel programming language/library could cause its premature death...

Re:I have a serious question:

[KGIII]

Agreed and, well, I must ask. Did his post REALLY deserve a -1 Troll or should it have been -0 We're not that interested?

Re:I have a serious question:

[Anonymous Coward]

That had fuck-all to do with the PS1 using a RISC and everything to do with it being a frozen platform with a special-purpose design with special-purpose support chips onto which some of the most processing-intensive tasks could be offloaded.

By comparison, your typical PC at the time was a general-purpose machine running a general-purpose operating system, doing everything in software and typically at higher resolution to boot. Once 3D accelerators entered the consumer market the Playstation was left in the dust.

Re:I have a serious question:

[Ilgaz]

I got Quad G5 and I am _just_ beginning to see that they are actually used in Mac OS X Leopard (threaded spotlight, diskutil bz2) and Adobe Flash (Version 10 beta uses all). As a video guy, parallel processing is a life saver but it all depends on the developer. Even sometimes some real stupid decisions. E.g. OS X Quicktime encoder will use only 2 CPUs/Cores while encoding H264 while open source solutions or freeware can go up to 4 CPUs (I bet 8 on opto-xeon). Note that I am speaking from video perspective which is relatively simple to parallel process (divide it to number of CPUs and process)

I was watching the Computer Chronicles, a very old episode and they were talking about our issue in 1980s. It was a super computer programming problem, how to make programs actually using the massive parallelism, new development concepts, how to "teach" developers to use multiple CPUs.

http://www.archive.org/details/CC126_supercomputers

That is from 1984! It is amazing that we actually get impressed when Adobe Flash 10 beta uses 400% CPU (in a good sense) in 2008. So the problem is not solved at all :)

Re:I have a serious question:

[Mattsson]

Large memories on graphics cards need at least that amount of addressing, but sometimes use much more.

In my machine with 4GB RAM and 32bit OS, I get ~3.4GB with a 256MB card, but if I put a 8MB pci graphics card in there in order to get 3 heads instead of 2, I lose another 400MB RAM. =)

In my girlfriends machine, which is identical except that it has a 512MB card, we actually get more usable RAM than in the one with a 256MB card. (When I leave out the pci card) %)

Re:I have a serious question:

[Anonymous Coward]

excellent post

Re:I have a serious question:

[msormune]

Yes it will, if the program/whatever runs in a virtual machine environment like .NET CLR or Java VM. In that case, one of the CPUs/cores could take up garbage collection, for example, which of course would work better that one CPU/core doing all the job.

Also, one core could handle I/O operations like TCP/IP.

Seriously.

Re:I have a serious question:

[Anonymous Coward]

x86 chips are not the fastest out there. Power6 is the fastest processor. Power7 will be when it comes out I expect.

Re:I have a serious question:

[tehcyder]

Hadn't DOS and the PC-clones crippled with mono-processor/mono-threading DOS/Windows stack become the dominant architecture for most of the 90s, we would have rock-solid, secure, multi-processor, 64-bit RISC boxes running some flavor of Unix on our desktops by now.

Yes, thanks to Bill Gates, I don't have a real super-computer on my desk right now, what a bastard.

Re:I have a serious question:

[twiddlingbits]

Are we having a TECHNICAL discussion of the relative merits of OSes and multi-core processors or are we screwing around just throwing crap ideas out hoping someone won't correct them? The post was DEFINITELY not a Troll and anyone who modded it as such needs to have their mod points taken away. People who think and repeat the incorrect and incomplete generalities they read on the Internet and thus think they know something are dangerous. Geeks complain thier bosses do that sort of thing yet they don't raise hell when they have the chance to stop it.

Re:I have a serious question:

[encoderer]

No, we're just screwing around. Get over it.

Re:I have a serious question:

[Danathar]

Yes, but if you were to plow your field would you rather use 2 oxen or 1024 chickens?

Personally I'd pay good money to see 1024 chickens plow a field but two oxen would probably be more efficient.

Re:I have a serious question:

[Anonymous Coward]

'x86 chips are still the fastest out there'

Nope. That would be the IBM Power6.

"IBM ships fastest processor on Earth"
http://tech.slashdot.org/article.pl?sid=08/04/10/1152205

Re:I have a serious question:

[kabocox]

Anyhow, i generally view 64-bits as a waste of address space, UNLESS you're accessing large amounts of memory (>3GB per program!). This will be more of a concern in the next few years, but there isn't any rush. I use 64-bit Vista for development (Because I have 4GB of RAM) but otherwise probably wouldn't care. Even Visual Studio (the dev platform for 64-bit code) is mostly a 32-bit app, nor should they change it.

You know you've got a good point about 32 bit vs 64 bit apps. I'm more nervous about 32 bit vs 64 bit Windows OS. O.k. you are right unless you've got 4 Gb or more of ram and apps that make use of it, you won't see any magic benefit with 64 bit windows. My worry is that we are on the transition edge of 4Gbs in a standard desktop. O.k. it may take another 5 years, but your plain jane windows computer should be able to use much more than 4 Gb of ram. O.k. 98% of users won't notice or care that their OS is 64 bit and can handle 4Gb of ram. What matters is that it can just in case Duke Nuke'Em Forever gets released and suggests you have 10 Gb of ram to run or youtube home video editor that let's you do real time video editing of all your videos or who knows what neat awesome app that chews through massive amounts of RAM to make pretty shines for their user base. O.k. right now I can't even run the existing shiny games. I only have 512 MB of ram. It doesn't take a genius to see that shortly 1-2 GB and then 4 GB will become min. spec. for those types of apps.

Re:I have a serious question:

[jerAzevedo]

I program x86 assembly both 32 bit and 64 bit. I can tell you that most instructions as compiled by gcc/g++ are 32 bit. Whenever gcc/g++ uses 64 bit operations it is most of the time a speed improvement, excluding operations on the stack pointer. The extra registers are used when applicable (as in the 32 bit parts of the 64 bit registers %r8-%r15, eg %r8d-%r15d). Stack references have to be 64 bit but you can use 32, 16, and 8 bit operations on the actual data. Only when performing actions on the stack pointer directly do you use 64 bit operations. For example subq $8, %rsp subtracts 8 from the stack pointer, movl %eax, (%rsp) moves the value of %eax to the data region pointed by %rsp.

Where you really see speed improvements are where your program uses floats, doubles, or longs. All int variables / operations in your program use 32bit registers. Float is used a lot by opengl so there is definitely a benefit for using 64 bit over 32 bit when it comes to things like games.

So it's not as "one sided" as you're trying to make it out to be. Eg the program might be larger because opcodes for 64bit instructions are bigger but whenever gcc/g++ is using these registers it's most likely going to result in a speed improvement (minus stack operations).

Eg,

subq %rax,%rax: 48 29 c0 (64 bit operation, 3 bytes on HD)
subl %eax,%eax: 29 c0 (32 bit operation, 2 bytes on HD)


movq $0xffffffff,%rax: 48 b8 ff ff ff ff 00 00 00 00 (64bit, 10 bytes)
movl $0xffffffff,%eax: b8 ff ff ff ff (32 bit operation, 5 bytes)

Of course in the second example the two instruction do exactly the same thing. %rax and %eax look exactly the same afterwords (%rax: $0x00000000ffffffff, %eax: $0xffffffff). So in this case the compiler would automatically use the 32 bit operation. But if you were dealing with longs for example and were moving $0xffffffffffffffff into %rax it would use the 64 operation, and this would of course be a performance boost.

Re:I have a serious question:

[CrazedSanity]

So, essentially, if I have a single-threaded app on a quad core, it'll perform at 1/4th the potential speed.

Another, simpler way to look at it: now, instead of your M$ OS killing your gameplay experience, your game will (mostly) have it's own "dedicated" processor.

Re:I have a serious question:

[weepinganus]

To phrase that as a car analogy, 4 trucks driving at 60 mph is not the same as 1 truck driving at 240 mph, although there are some situations in which the output would be similar.

Re:I have a serious question:

[Anonymous Coward]

"Is there any way to force a process to run over 2 cores at 100%"

In theory, maybe. And I'm talking bout the ideal situation.

Near linear speedup is possible in systems involving a handful of processing elements. I say near linear because according to Amdahl's law, a program needs to consist of 100% parallel code in order to achieve this. That means every single line in your program must be able to be executed in parallel. And this is impossible - if there is even one line that initializes some shared resource, if the programs have any synchronization needs, or if there is even one cycle of waiting due to a communication, then you are less than 100%.

All this assumes your process was heavily optimized to run concurrently. Chances are if you are picking a random process on your machine and expecting a checkbox in the properties box of the task manager to force execution on two processing elements, then you'll never get that. In fact, a 100% non sequential process will not even use the second processing element, by it's very nature it can't even use it.

Re:I have a serious question:

[badboy_tw2002]

Seriously, I think its facinating to see how many people have responded with "yes...but" answers to a really simple response to a fairly novice question. Yes, there's lots of ways to behind the scenes parallelize something, but that's really not the point. I was really just illustrating what the OP was seeing in as simple a way possible. I guess I should have just said

int main()
{
        int i;
        while (true) i++;
}

This is C. No CLR. This will only run on one core. There are lots of ways for other things to run in the background, other processes, etc. But what he was seeing was a single process that was not threaded using one core of his CPU. That's it.

Re:I have a serious question:

[badboy_tw2002]

p.s.

Seriously. :)

Re:I have a serious question:

[Anonymous Coward]

Yes, that is true, but who's running only one computationally-intensive process at a time on these servers?

Re:I have a serious question:

[vux984]

I don't know about the specifications of your reproductive system, but I could make 3 babies with 3 women in 9 months with no physiological problems (the psychological effect will be different though).

But I only want one baby faster, not three.

What you mean is that a woman will take at least 27 months to make 3 babies from 3 men. (for the sake of simplicity we don't count twins or more here)

That's true, but I could add more women and get 3 babies in 9 months instead of 27.

Re:I have a serious question:

[Anonymous Coward]

"more cores won't get a single threaded process done faster." - by badboy_tw2002 (524611) on Monday July 14, @10:19PM (#24190701)

Oh, I'd have to say, that's NOT 'completely true'... & mainly because of the OS process scheduler kernel subsystems, & OTHER running processes + OTHER CPUs/Cores present....

Those other processes (from the OS & other apps running) MIGHT have been competing w/ said "single-threaded" designed process, on that SAME single core (or, even in an SMP/HT/SMT environs-system) in a single cpu/core system, OR otherwise...

Think about it:

E.G.-> Your process "A" there is probably doing some recursive activity, say string processing! Pretty heavy...

(Fact is, I just wrote an app along those lines for HOSTS file manipulations in fact, & it does pretty much as you describe (especially when I boost its parent thread (1 of 2) to REALTIME cpu priority (I allow this in this app) -> http://www.thenewtech.com/forums/attachment.php?attachmentid=58&stc=1&d=1214726022 , because of the nature of the work, string processing largely which is fairly CPU intensive/expensive, it actually sort of fits here, but, is multithreaded in design)...

See, you're not considering OTHER running processes (OR, threads even) that can/could be "shunted off" by the OS process scheduler kernel subsystem module, to OTHER Cpus/Cores present (emulated via H/T or otherwise, even true multi-cpu SMP arrangements - as its pretty much 'all the same' once this gets 'virtualized' to the OS via the HAL layer + drivers)... especially once the "main cpu/core" is saturated, or nearing it.

I.E.-> Once YOUR process (or mine) starts "tearing up" CPU core #1... if there are OTHER cpu cores present? The OS will send their child threads of execution off to said 'freer' cores... allowing your app to continue, while the OS + other program parent & child threads run off of the less cpu-cycle saturated/starved CPUs/Cores present...

APK

Re:I have a serious question:

[Nyall]

nope. it wasn't a troll

Re:I have a serious question:

[FernandoBR]

Yes, but with more ladies, it'll be much more funny!

Re:I have a serious question:

[Anonymous Coward]

x86 chips are RISC like in a sense that the front-end decode logic breaks up the instruction into smaller RISC like operations. However given the fact that x86 instruction sizes are variable in nature because of all the prefixes and what have you, the front end decode is usually complicated and plays a significant role in timing.

SPARC benefits from the fact that the instruction sizes are all the same, hence the decode is extremely simple and can be done in a fixed number of cycles.

x86 are fastest out there, but that depends on the situation and the type of applications they are running. x86 are extremely good at single-threaded performance, but not as much when it comes to running multi-threaded apps. One might argue that the recent SPARC processors and IBM Power are better in that regard.

Re:I have a serious question:

[Anonymous Coward]

Aren't a lot of games and apps single-threaded? Hmmm. I figured that dual/quad-core wasn't all it's cracked up to be. So, essentially, if I have a single-threaded app on a quad core, it'll perform at 1/4th the potential speed.

That would only be true if you went with the 4 core 4ghz chips over some non-existent single core 16ghz CPU.

In reality, a single threaded application on one core of this multi-core 4ghz cpu, will STILL run faster than the same app on your same #of core CPU that is under 4ghz that you own right now.

So no, it would still run your single thread apps faster than what you are using. However it will run (well made) multithreaded apps 16 times faster than you could ever hope.

In addition, I can also simply run 16 single threaded processes at 4ghz each, something your CPU couldn't do even if it was 4ghz or more.

Re:I have a serious question:

[rbanffy]

"Also, for Joe Sixpack, 64-bit is pointless"

Mostly because we use Windows or Unix-like OSs. All they can to with vast address spaces is a vast program memory. What a waste of bits.

Re:I have a serious question:

[rbanffy]

"It avoids a lot of concurrency related problems that most programmers are not properly trained to deal with."

And... Why do you think these programmers were never properly trained? Would it be because they never saw a real multi-processing box?

Re:I have a serious question:

[rbanffy]

It's a good exercise of imagination to project what a current-day Amiga, Atari ST, RISC-PC or even Lisp Machine would look like.

I bet it would blow away any gaming PC.

Re:I have a serious question:

[rbanffy]

"x86 chips today are 99% RISC-like (the term RISC is rarely uses today, since basically no modern CPUs are "pure RISC" in design (reduced as in not even having a multiply instruction, like older SPARCs). Sure the exposed architecture is ugly x86, but that's the compiler's job to worry about"

So, half the silicon of a high-performance RISC-like CPU is used to make it look like an old and clunky CISC architecture? That's progress.

Wouldn't it be nicer to dedicate that silicon to build another core, more cache or something else?

Re:I have a serious question:

[oblivionboy]

..or expensive.

Re:I have a serious question:

[Anonymous Coward]

That part takes less than 10% of the die on a modern x86. It is basically a "solved problem" and not really interesting.

Re:I have a serious question:

[rbanffy]

Well... 10% of the die of a modern x86 is a whole lot of transistors.

How many transistors a POWER7 has compared to a Core 2 Duo?

Re:I have a serious question:

[LionMage]

Besides the increased number of general purpose registers on x86-64, there's also the change in calling convention -- on 32-bit x86, function arguments are pushed onto the stack, whereas on x86-64, the arguments are passed via register. That's another reason that apps like Photoshop run faster when compiled as 64-bit x86 code.

Re:I have a serious question:

[painehope]

Of course they're screwing around - most applications using these chips will be either (a) server apps that are multi-threaded and optimized for the architecture (POWER architectures will not run WoW, geniuses) or (b) supercomputing applications (in which case they will use MPI or OpenMP and also be heavily tuned to the architecture).

About the most generic server apps these things will see will be things like NIS or LDAP. Compiled for the architecture and most likely still heavily optimized. There's a reason that IBM releases their own compiler (XL) suite for POWER chips.

And, yes, it will run Linux. And AIX. And nothing else.

Re:I have a serious question:

[Anonymous Coward]

I went back and checked the source for that number, and it was AMD's CTO estimating the transistor count for the K8 core, which had between 70 to 230 million transistors, depending on the size of the caches. The Conroe core has approx. 290 million transistors (4MB L2), while the Yorkfield XE has 820 million transistors on two dies (6MB L2 per die). I don't know the transistor counts for the POWER7, but according to Wikipedia the POWER6 has about 790 million transistors, likely due to the per-core 4MB L2 caches.

So the actual CPU logic part of the core is dwarfed by the monster caches in modern processors, and the instruction decoder part is a small part of that.

Re:I have a serious question:

[JebusIsLord]

It's true, it isn't as one-sided as I said. I think I did a pretty good job of simplifying for the target audience (OP) though. Your post will be more informative to the /. audience as a whole.

Re:I have a serious question:

[KGIII]

I have to wonder if Microsoft's quietly operating HPC team will do something about just Linux and AIX running on it. You may know but many people don't that Windows can and will cluster (for example) just fine and Microsoft actually has a rather large investment in the HPC arena.

Re:I have a serious question:

[Anonymous Coward]

(POWER architectures will not run WoW, geniuses)

The Mac client supports PPC. It'll take some work to get it to run on POWER7, though...

Re:I have a serious question:

[painehope]

Good thought, but I doubt it. I've always viewed MS's HPC as something of a joke - I mean, they've had to buy the hardware and supply the software for free for almost all major deployments. And most of those were in the educational industry, whose attitude consists of "it's free, we'll take it, who gets bragging rights? Oh, it's supposed to do something? Shit, I'll throw some grad students at it..." - and nothing ever gets done.

Re:I have a serious question:

[painehope]

And notice that Apple moved to a x86/x64 platform. So, therefore, I seriously doubt they'll be releasing any more OSes to run on PPC platforms, not to mention that PPC and POWER architectures are different beasts anyways. Related, but different in crucial ways.

Re:I have a serious question:

[poopdeville]

The hardest part will be getting OS X installed on a POWER7 machine. POWER is a slight super-set of the PowerPC ISA, so OS X PPC will run just fine, assuming you have an Apple licensed ROM.

Re:I have a serious question:

[rbanffy]

For this comparison to work, we would need the core transistor counts.

Still, I bet one could squeeze one or two 64-bit MIPS cores inside the Core 2 instruction decoder.

Re:I have a serious question:

[Anonymous Coward]

The R14000 used 7.2 million transistors according to Wikipedia, so no unless you would settle for something that ran like an 80s vintage 8-bitter.

Re:I have a serious question:

[rbanffy]

Well... You should be able to do a full R4000 core in about a million transistors. The original 4000 had 1.35 including caches.

Re:Vista capable?

[Anonymous Coward]

Y'know, it's kinda lame to make Vista vaporware jokes now that it's been released.

Tick...tick...tick...

[HaeMaker]

...tick...tick...tick...

Steve Jobs is crying in his pillow tonight.

[pecosdave]

If I would have held out just a little bit longer!

Re:Steve Jobs is crying in his pillow tonight.

[Anonymous Coward]

yeah, sure. Because THAT would fit perfectly in a laptop.

Apple changed ship because IBM was unwilling to commit to a laptop worthy chip.

and judging by the amount of laptops Apple sells, the decision was wise.

Re:Steve Jobs is crying in his pillow tonight.

[Orion Blastar]

Chances are IBM will still have a problem supplying them, plus new game consoles will get a priority in shipping in 2010, when that XBox 720 or Playstation 4 comes out.

It is also possible that the eight core chip will be really expensive, and in order to keep up with it a PowerMac would cost $4000 or more just to eliminate bottlenecks and use optical technology like super computers use to be able to use the chip properly. Not to say that nothing stops Apple from bringing out PowerPC based Macs in 2010 as Mac OSX already runs on PowerPC code and would have to be modified to run on the Power7 instruction set. Which is very doable. Apple could have Intel Macs for low cost systems for home and small businesses, and Power7 Macs for high end workstations and servers for middle to large businesses. I don't see why Apple couldn't bring back PowerMacs and sell them next to Intel Macs, unless IBM starts to have production problems again and can't supply Apple the number of PowerPC chips that they need?

Re:Steve Jobs is crying in his pillow tonight.

[pecosdave]

I've actually been wondering if it would be possible to put the PS3's Cell processor into a laptop. I would pay reasonable money for that.

Re:Steve Jobs is crying in his pillow tonight.

[Anonymous Coward]

That and the fact that this isn't a PowerPC processor, anyways.

Re:Steve Jobs is crying in his pillow tonight.

[Wesley Felter]

Look at the heatsink in a PS3 and you have your answer.

Re:Steve Jobs is crying in his pillow tonight.

[Anonymous Coward]

A geek's wet dream if there ever was one.

Toshiba actually announced such a beast (albeit with less SPEs, wich might be a way to use slightly defective cell chips to increase the yields).
The only problem is that if this is only used in one machine, noone is going to bother writing applications for it.

Re:Steve Jobs is crying in his pillow tonight.

[Anonymous Coward]

IBM didn't have production problems, they just didn't think it was all that important to cater to what a small consumer, like Apple, wanted.

Re:Steve Jobs is crying in his pillow tonight.

[Anonymous Coward]

No he isn't. Apple's market share is going up (I don't have the figures at hand, but you can look it up).

Also, this is a SERVER CHIP NOT A LAPTOP CHIP. No Power architecture laptop oriented chip exists, and there is no plan to make one. IBM is not interested, because they have no need for such a beast, and Freescale, which makes Power architecture embedded chips doesn't need it either. Apple can't afford to design and fab just for it's laptop volume. It would be foolish to have a different CPU for desktop/server machines and the laptops.

So the only real question is are you as stupid as you seem for making such an inane comment?

Re:Steve Jobs is crying in his pillow tonight.

[Anonymous Coward]

Yes, it is a PowerPC processor. PowerPC is an instruction set. POWER7 is IBM's newest implementation of the PowerPC architecture.

Re:Steve Jobs is crying in his pillow tonight.

[Hal_Porter]

Plus a Cell processor would be hopeless in Mac. The PPC core was relatively underpowered - most of the horsepower was in the SPEs and OS X would only have used those in very special circumstance with a lot of work.

E.g. you could imagine Photoshop using them for filters but given how slow Adobe has been at moving to X86 which is very similar to PPC, I wouldn't hold my breath waiting for a Photoshop which was tuned for a very different architecture like Cell.

Re:Steve Jobs is crying in his pillow tonight.

[Anonymous Coward]

AFAICT, IBM is phasing out the PowerPC brandname (makes sense, it isn't in PCs anymore). But POWER and PowerPC have meant the same thing for a while now.

Re:Steve Jobs is crying in his pillow tonight.

[TeknoHog]

A much better PPC for a laptop would be the PWRficient. How does 2 GHz at 7 W per core sound? I wonder if such a laptop ever takes off though, now that Apple has bought the chop company.

Re:Steve Jobs is crying in his pillow tonight.

[Ilgaz]

Toshiba does it. Cell is a secondary processor.

http://www.computerworld.com/action/article.do?command=viewArticleBasic&taxonomyName=laptops&articleId=9102518&taxonomyId=66&intsrc=kc_top

Re:Steve Jobs is crying in his pillow tonight.

[Hal_Porter]

A much better PPC for a laptop would be the PWRficient. How does 2 GHz at 7 W per core sound?

Much worse than Atom?

Re:Steve Jobs is crying in his pillow tonight.

[Anonymous Coward]

http://hardware.slashdot.org/article.pl?sid=08/01/11/150215 :)

Re:Steve Jobs is crying in his pillow tonight.

[jim.hansson]

toshiba has a prototype laptop with a normal intel CPU and one extra "lite" cell processor http://crave.cnet.co.uk/laptops/0,39029450,49295004,00.htm

Re:Steve Jobs is crying in his pillow tonight.

[pecosdave]

Nifty, if Toshiba hurries they may be able to outdo the next iProduct.

Re:Steve Jobs is crying in his pillow tonight.

[pecosdave]

I tend to remember a "system on a chip" experiment from the late 90's. One of my clients got one with a Compaq label. Crap. Complete and utter crap.

That doesn't mean I've dismissed PWRficient, and I am intrigued by it, even though Atom does exist. It's worth watching.

Re:Steve Jobs is crying in his pillow tonight.

[Anonymous Coward]

Apple will continue to supply the PA6T-1682M for now on an end-of-life basis (mainly to the military), but P. A. Semi will be making iPhone chips from now on.

Re:Steve Jobs is crying in his pillow tonight.

[Orion Blastar]

Much like Intel phased out the Pentium series, but the Centrino and Dual Core series still use Pentium instruction sets but are no longer called Pentium.

Re:Steve Jobs is crying in his pillow tonight.

[Anonymous Coward]

yeah, sure. Because THAT would fit perfectly in a laptop. Apple changed ship because IBM was unwilling to commit to a laptop worthy chip. and judging by the amount of laptops Apple sells, the decision was wise.

Someone needs to mod the parent up because this is bang on.

Re:Finally

[QuantumRiff]

Yeah, right, it has optical connections. they will have to be disabled to play videos, otherwise you might copy them!

Can I imagine a ...

[krkhan]

... beowulf cluster of these 300,000 core machines? I want to be able to play high-definition video without any lag on Windows 7.

Release set for 2010

[EkriirkE]

WTF? By that date, using the planned arch, it will be obsolete if not already scrapped before then.

Hmmm

[Anonymous Coward]

Isn't there still no consumer-grade CPU marketed as 4ghz? Last I heard, that clock speed race ended at 3.8ghz due to heat issues.

Can these be used in PCs?

[the0]

Can they? Probably not... but why? If IBM has the capacity to manufacture these (presumably) awesome chips, why doesn't it develop for the general PC market?

  Hint to the mods: These are insightful/interesting questions.

Re:Can these be used in PCs?

[Yvan256]

I'd guess that general PC market = low-margins, not to mention that this new processor will probably require a 5KW power supply.

Re:Can these be used in PCs?

[Wesley Felter]

Putting a server processor in PCs was tried; it was called the G5. It turned out OK, although having no laptop version was annoying.

Re:Can these be used in PCs?

[Anonymous Coward]

Back in 1991, IBM began making a POWER derived chip intended for desktop PC use. It was called, oddly enough PowerPC. But since Apple was the only manufacturer to actually use the PowerPC in desktops, IBM and Motorola eventually said fsck it, and focused on embedded uses and the desktop performance languished until Apple abandoned the chip in favor of Intel.

POWER for the desktop could kick ass, assuming you can convince anyone to care about it...

Great

[afidel]

So you can get 16 cores in a low end box but it still won't have enough I/O slots so you will have to buy a shelf at $obscene_amount, seriously why does IBM put such few I/O slots in the lower end P series boxes?

Re:Great

[Anonymous Coward]

For me, All I require are two F/C cards and two NICs. One is usually on the mainboard.The Boxes can then do computationally intensive work. Like running a clustered database stored on a SAN.

Re:Great

[Anonymous Coward]

It is called APV. Description here These boxes are designed to be virtualized. If you do not like that you can still get a D20 or the (newer) G30 tray.

Re:Great

[Just Some Guy]

you will have to buy a shelf at $obscene_amount, seriously why does IBM put such few I/O slots in the lower end P series boxes?

Asked and answered.

Re:Great

[Anonymous Coward]

There should be no general need for a large number of I/O slots. Apart from storage (FC?) and ethernet networking, what will most people need to plug in? And ethernet is used to provide storage in many cases

Re:Sorry Mac fans . . .

[imamac]

So still no G5 laptop???? AHHHHHH!!!!!!

how much closer are we now to ...

[bizitch]

"daisy, daisy ....." or "stop that dave ...."

2010 eh?

[Yvan256]

"My god, it's full of cores!"

4 Threads per core?

[jdb2]

It should be noted that previous POWER architectures had 2 threads per core. They also had SMT ( Simultaneous Multi-Threading ) support, which gave them an "effective" 4 threads per core. I wonder. Are the all the threads on the POWER7 "true" threads ( ie. 4 execution units -- 1 per thread ) or is it a 2 thread setup with SMT? On the other hand, if the POWER7 really does have 4 "true" threads, then with SMT you'd get an "effective" *8* threads per core.

jdb2

Re:4 Threads per core?

[Anonymous Coward]

Actually POWER6 is has an effective two threads per core. If I boot a system with 16 cores and SMT enabled, AIX sees 32 "processors." With SMT disabled, I see 16 "processors."

Re:4 Threads per core?

[XaXXon]

I always thought the definition of a "core" as whatever the minimal set of hardware required to run a single thread at "full power". By my logic, anytime you run more than one thread on a core, you're doing what SMT does.

Someone please tell me if I'm wrong (and how).

Re:4 Threads per core?

[mike260]

Are the all the threads on the POWER7 "true" threads ( ie. 4 execution units -- 1 per thread ) or is it a 2 thread setup with SMT?

The threads in question are almost certainly SMT-style threads; that is, parallel streams of instructions feeding into the same shared pool of execution resources.
If you're going to have separate execution resources for each thread then they're not threads, they're cores.
BTW, one 'execution unit' per thread doesn't make much sense given than modern CPUs need separate execution units for different tasks, and also want to execute multiple instructions in parallel.

Re:4 Threads per core?

[SQL Error]

It should be noted that previous POWER architectures had 2 threads per core.

Correct.

They also had SMT ( Simultaneous Multi-Threading ) support, which gave them an "effective" 4 threads per core.

No, they do not "also" have SMT. It is the SMT that gives them 2 threads per core in the first place.

Power 5 & 6 have 2-way SMT. Power 7 has 4-way SMT.

Re:4 Threads per core?

[Anonymous Coward]

It's too bad AIX sucks. With 32-way SMT it just sucks more. At 4GHz, it just sucks faster.

Re:4 Threads per core?

[Anonymous Coward]

Actually, they say each POWER6 chip has dual cores, and each core is 2-way via SMT. So a P6 "MCM" has two chips, four cores and 8 logical cpus (threads) when SMT is on.

POWER5 had chips grouped in 2 (DCM), 4 and 8 (MCMs) chip modules. Don't ask me why two chips is a DCM in P5 and MCM in P6. Marketing.

It's unlikely P7 would switch to fine-grained threading, so would expect 4 SMT threads per core.

Re:4 Threads per core?

[Macman408]

I always thought the definition of a "core" as whatever the minimal set of hardware required to run a single thread at "full power". By my logic, anytime you run more than one thread on a core, you're doing what SMT does.

Someone please tell me if I'm wrong (and how).

A core is a set of registers and function units, among other hardware. Each core is, effectively, a completely separate processor (though it likely shares some things, such as the L2 cache and FSB with other cores). Since processor usually refers to a whole chip (encased in a plastic or ceramic package, and soldered on the motherboard), the term "core" refers to when there is more than one inside a single package. The ultimate goal is usually to have all cores on a single piece of silicon, but often multi-chip modules are used (especially early in production), where a four-core processor might contain two silicon dies, each with 2 cores. This can help increase yield (by reducing the die size), and reduce production cost. After improving the yield of the processor, or changing to a reduced feature size (eg 90 nm to 65 nm), a switch back to a single die is possible, reducing the packaging cost.

Simultaneous Multithreading (SMT), on the other hand, works on a single processor/core. It is a feature that allows sharing of the processor resources, such as registers and functional units. A PowerPC 970 (which never had SMT support) could issue 4 instructions and 1 branch every clock cycle. Because of that, plus the deep pipelines, up to 216 instructions can be in various stages of completion at any given time. However, on average, a program branches every 4 instructions - this means that the processor would have to correctly predict 54 branches to keep the pipeline full, AND that the instructions would be (mostly) independent of each other. This isn't easy to do. So, what many processors do is split the available resources. One might issue 2 instructions from one thread and 2 instructions from another in each clock cycle, or alternate clock cycles issuing 4 from one, then 4 from the other. This shares most of the CPU's resources, while requiring a fairly minimal amount of extra logic to track the second thread.

So, cores are extra hardware that can perform more calculations. SMT is taking better advantage of what is already present.

However, the disadvantage of SMT is that it can slow a single-threaded program down, because now it has to share resources. Some processors actually do away with superscalar (aka issuing multiple instructions at once) and out-of-order execution and bypass logic, and instead rotate through many threads. For example, if the pipeline is 8 stages deep and supports 8 threads in hardware, it can issue one instruction from each thread in each clock cycle. Then it never needs to check if an instruction is dependent on an earlier one, because an instruction is completed before the next one from the same program is issued. Having this many threads can also minimize the cost of a branch misprediction, cache miss, or other long-latency events. Also, removing the bypass, dependency checking, multiple-issue, and instruction reordering logic can give a significant reduction in power and area on the chip. The performance hit by these eliminations can then be made up by adding many more cores than you'd see in a superscalar out-of-order processor like a Pentium or Core architecture. The catch is that a processor like this is only faster if you have enough threads to keep the processor busy. However, if your problem is big enough to need a supercomputer, you're darn well going to spend time writing it to take advantage of as many threads as you can.

Re:4 Threads per core?

[Ilgaz]

They actually let you reboot a POWER6 server to try things? No wonder you post as AC :)

Re:4 Threads per core?

[SpinyNorman]

SMT (aka hyperthreading - Intel P4 HT) is as real as multi-threading per core gets. If you instead duplicated the entire execution pipeline you'd just have another entire core!

The execution pipeline of a modern CPU consists of many specialized decode, load/store, integer/logic (ALU), floating point (FPU), etc components (often more than one of each type) that all run independently, maybe idle at any given time if not needed or maybe waiting for operand results of an earlier step in the pipeline, or waiting for memory access (latency).

What SMT essentially does is add mutiple decode streams feeding into the pipeline so that the pipeline and it's constituent components are genuinely exectuting the instruction streams from multiple threads simultaneously. Despite the fact that they are still competing for the shared components there is still genuine parallelism since in a single threaded scenarion many of the units may be sitting idle waiting for operands or memory access. Using SMT means that the execution units are more fully utilized and it prevents memory latency (cache misses) from stailling the core by allowing the pipeline to utilize the execution units (ALU, FPU, etc) for instructions for which the operands are already available.

In summary, SMT is true multi-threaded core multi-threading. The only better (but more wasteful) type of multithreading is to use multiple cores, but they are still not fully indpendent since (at least in a desktop architecture) they are going to be competing for scarce memory and bus bandwidth.

Re:4 Threads per core?

[Anonymous Coward]

And all SMT does is allow you to make better use of an otherwise idling pipeline.

There is a reason why AMD hasn't used this technique and it has a little something to do with "fewer pipeline bubbles."

Even the core2 doesn't use it. If they threw SMT on core2 it'd be SOLELY for a marketing gimmick and not because it helps.

The only reason SMT made sense on the P4's ridiculously long pipeline was because it was so riddled with bubbles it wasn't that hard to run another process right through it.

Re:4 Threads per core?

[mczak]

Even the core2 doesn't use it. If they threw SMT on core2 it'd be SOLELY for a marketing gimmick and not because it helps.

Not quite true. This is all a question of how expensive the feature is to implement vs. how much performance improvement (which can vary a lot depending on the applications used) you will get. Nehalem (the core2 successor), if you're looking only at the core architecture, is little more than exactly that, a core2 with SMT (of course it has way more differences in the "uncore" area, cache organization, integrated memory controller etc.), and I doubt it's just for marketing purposes... Sure it will not double performance but OTOH it is fairly cheap to implement.

Re:4 Threads per core?

[jdb2]

For some reason I had the idea that the current POWER architecture implementation was scalar.

Please change "4 execution units -- 1 per core" to "4 execution pipelines, supporting up to 4 simultaneous threads of execution".

jdb2

Re:4 Threads per core?

[jdb2]

SMT (aka hyperthreading - Intel P4 HT) is as real as multi-threading per core gets. If you instead duplicated the entire execution pipeline you'd just have another entire core!

Please see my reply to mike260. I don't know where I got the idea that the POWER architecture was scalar.

The execution pipeline of a modern CPU consists of many specialized decode, load/store, integer/logic (ALU), floating point (FPU), etc components (often more than one of each type) that all run independently, maybe idle at any given time if not needed or maybe waiting for operand results of an earlier step in the pipeline, or waiting for memory access (latency).

What SMT essentially does is add mutiple decode streams feeding into the pipeline so that the pipeline and it's constituent components are genuinely exectuting the instruction streams from multiple threads simultaneously. Despite the fact that they are still competing for the shared components there is still genuine parallelism since in a single threaded scenarion many of the units may be sitting idle waiting for operands or memory access. Using SMT means that the execution units are more fully utilized and it prevents memory latency (cache misses) from stailling the core by allowing the pipeline to utilize the execution units (ALU, FPU, etc) for instructions for which the operands are already available.

In summary, SMT is true multi-threaded core multi-threading. The only better (but more wasteful) type of multithreading is to use multiple cores, but they are still not fully indpendent since (at least in a desktop architecture) they are going to be competing for scarce memory and bus bandwidth.

I appreciate your thoughtfulness but I am versed in modern Digital-Logic/Microprocessor Design and Computer Architecture. I have to admit I thought you were patronizing me. ( But that's normal as I have a short temper, which can be intuited from my past posts ;) But again, the reason I said "per execution unit" was due to the strange idea that the POWER architecture was scalar. I don't know where I got that in my head -- a scalar processor with SMT is an oxymoron. The current POWER architecture implementation is in-order -- for some reason I mistook that for "scalar" - Brain Fart. :)

jdb2

Re:4 Threads per core?

[jdb2]

It should be noted that previous POWER architectures had 2 threads per core.

Correct.

They also had SMT ( Simultaneous Multi-Threading ) support, which gave them an "effective" 4 threads per core.

No, they do not "also" have SMT. It is the SMT that gives them 2 threads per core in the first place.

I'm no POWER architecture expert ( as can be seen ) but you need at least 1 execution pipeline to support an SMT thread. Without SMT the execution pipeline represents *1* thread of execution. If the processor *also* ( I don't know why you have a problem with this word. How about "in addition to" ?) has SMT support, then that gives you support for n additional threads for n execution pipelines.

Power 5 & 6 have 2-way SMT. Power 7 has 4-way SMT.

Thanks for the info, and correction.

jdb2

Re:4 Threads per core?

[Anonymous Coward]

The PPE used in the Cell and the Xbox360's Xenon is an in-order architecture, but other than that every POWER or PowerPC architecture that's on the market now that I know of is scalar, ranging from the low-end embedded variants to the supercomputer chips. A scalar architecture with SMT makes sense because the pipelines for the high-end CPUs are so ridiculously long that there can be over a hundred instructions in-flight (the PPC970/G5/POWER4 can have up to two hundred, the newer architectures likely even more). At this point it becomes a herculean task to wring out more performance out of a single instruction stream, so duplicating the state and processing another thread just makes things simpler!

Re:4 Threads per core?

[SQL Error]

And all SMT does is allow you to make better use of an otherwise idling pipeline.

Which can be very useful.

Even the core2 doesn't use it. If they threw SMT on core2 it'd be SOLELY for a marketing gimmick and not because it helps.

[cough]Nehalem[/cough]

Re:4 Threads per core?

[SQL Error]

The PPE used in the Cell and the Xbox360's Xenon is an in-order architecture, but other than that every POWER or PowerPC architecture that's on the market now that I know of is scalar, ranging from the low-end embedded variants to the supercomputer chips. A scalar architecture with SMT makes sense because the pipelines for the high-end CPUs are so ridiculously long that there can be over a hundred instructions in-flight (the PPC970/G5/POWER4 can have up to two hundred, the newer architectures likely even more). At this point it becomes a herculean task to wring out more performance out of a single instruction stream, so duplicating the state and processing another thread just makes things simpler!

Huh? Cell and Xenon are in-order superscalar. PPC970/G5/POWER4 are out-of-order superscalar. Some of the low-end embedded PowerPC chips are scalar (single-issue), but all POWER chips from the very beginning have been superscalar.

Re:4 Threads per core?

[SQL Error]

I'm no POWER architecture expert ( as can be seen ) but you need at least 1 execution pipeline to support an SMT thread.

No, in fact. A single pipeline can support an arbitrary number of threads, just taking one instruction from one of those threads in any given cycle. From memory, the POWER5 has 8 pipelines, and the POWER6 has 5. I don't think IBM have released details of the POWER7 yet.

Without SMT the execution pipeline represents *1* thread of execution.

Absolutely.

If the processor *also* ( I don't know why you have a problem with this word. How about "in addition to" ?)

How about just saying "has"?

has SMT support, then that gives you support for n additional threads for n execution pipelines.

Not exactly. You can have 2-way SMT on an 8-issue core (POWER5), or 8-way SMT on a 2-issue core. There's no direct relationship between the number of threads and the number of pipelines.

Re:4 Threads per core?

[Agripa]

Nehalem reportedly has a 4 cycle L1 cache latency instead of the 3 cycle L1 cache latency in Core2 so my guess is that SMT was necessary to make up for the lower IPC.

Re:4 Threads per core?

[jdb2]

I'm no POWER architecture expert ( as can be seen ) but you need at least 1 execution pipeline to support an SMT thread.

No, in fact. A single pipeline can support an arbitrary number of threads, just taking one instruction from one of those threads in any given cycle.

This is not SMT. It's fined-grained multithreading ( or barrel processing ) ala the UltraSparc T2. ( Niagra 2 ) I'm talking about "threads" in the context of SMT.

From memory, the POWER5 has 8 pipelines, and the POWER6 has 5. I don't think IBM have released details of the POWER7 yet.

Without SMT the execution pipeline represents *1* thread of execution.

Absolutely.

If the processor *also* ( I don't know why you have a problem with this word. How about "in addition to" ?)

How about just saying "has"?

has SMT support, then that gives you support for n additional threads for n execution pipelines.

Not exactly. You can have 2-way SMT on an 8-issue core (POWER5), or 8-way SMT on a 2-issue core.There's no direct relationship between the number of threads and the number of pipelines.

I had thought of this case, but again we're talking about "normal" SMT ie. SMT that gives you the best performance/thread-count. 2-way SMT on an 8-issue core is just wasting resources, especially in the case that the processor is in-order. 8-way SMT on a 2-issue core would result in a significant performance loss as the overhead associated with switching threads would outweigh the gain made by trying to saturate the execution resources of the core.

jdb2

Um, nope, his apps stink

[tjstork]

No, what this means is that his applications really aren't burdening the CPU. If you build an MT Windows App that genuinely scales, then, it will most certainly give all the CPUs up. What's happening in his case, most likely, is that he's i/o bound and his cpus are doing nothing. OR, his applications aren't even multithreaded. Or both. I've written some MT C++ apps for Windows for crunching insurance prices and yep, they'll peg all the CPU that you can give it. Also, I wouldn't recommend even setting the task affinity in TM.

Re:Um, nope, his apps stink

[tuomoks]

Right, except it's not always just I/O. I'm not much Windows fan but (XP at least) can be efficient. It is the bad application - I designed a comm. subsystem, queuing, en/decryption, image translation, key management, etc, tested it in 1,2,4,8 core systems (emulating the application) and could drive all cores to 100% busy, almost linear throughput increase. Now - add an application to top of that - 1-2 cores, 20%, 2-4 cores none and 8 cores -%10 throughput?

Took three months to fight the application developers (and they still don't get it?) - total misuse of threadpools in C#! And they were supposed to be the C#/.NET specialist - I'm just an OS guy (mainly MVS/Unix/Linux?) And I had a very good team writing the services for that subsystem but no saying anything about the application design?

The problem I see is that Windows is so much easier to write bad applications - the subsystem actually (excluding auditing) runs under Wine in Linux and, just for fun, I tested it. Same results, very near same throughput.

And please, if running in Intel, test your hyper-threading - not good for everything!

Re:Um, nope, his apps stink

[tjstork]

And please, if running in Intel, test your hyper-threading - not good for everything!

I'm almost to the point of recommending that hyperthreading should just be turned off because it so totally screws up floating point and more often than not, people needing loads of threads for crunching are terrible.

Took three months to fight the application developers (and they still don't get it?) - total misuse of threadpools in C#! And they were supposed to be the C#/.NET specialist - I'm just an OS guy (mainly MVS/Unix/Linux?) And I had a very good team writing the services for that subsystem but no saying anything about the application design?

By itself, Windows is actually a damned good product. I've actually had a good experience writing services for Windows. *nix stuff makes it easier to develop daemons but I think Windows has made it easier to deal with threads - at least until MS f--- up threaded debugging in Visual Studio.

But its the legion of VB RAD hackers come C# programmers that just wreck it for the rest of us.
I have found, that keeps me into Linux, is that 80% of Windows developers are utterly incompetent, whereas 80% of Linux developers are actually very good.

Windows developers are often self taught and that's a good thing, but most have never had any formal Computer Science education and completely don't know everything is ultimately some kind of a graph and as such lack the tools to properly analyze problems. Instead, what you have is a bunch of retards applying Microsoft designed cookie cutter developmental processes to arrive at solutions and lack any sort of understanding as to what it is they are actually doing, and that MS dumbs some of its APIs down to cater to these people.

So, even though I like Windows, when I come home, I find myself firing up my Linux box even when it aggravates me, because, there's only so much stupid code that I can take in one day. But still, I still like the way Windows C SDK in some ways more than I like the way that X works. Had X just given a little more thought to 2d drawing when it was developed, and they had been able to agree on a standardized widget bundle, we would not have the mess that is GTK versus Qt all sitting on top of creaky X, and still only barely have device independent drawing in the way that Windows has had for nearly 20 years.

Nah, they are just building a new kind of bomb

[tjstork]

They only -say- they are using all this computer power for weather forecasting. In reality, they are devising a new kind of anti-matter bomb, and they'll just use that sucker to take over the world. PRetty soon, the capital of Earth will be the University of Illinois, so, if you stole any books from the campus library, you'd better pony up, or you'll be on your way to the Gulag they plan on constructing in North Dakota.

Imagine a Beowulf cluster...

[ducomputergeek]

errr....wait a minute, that sounds exactly what U of I is building!

Apple Got Dumped By IBM.

[Anonymous Coward]

When IBM secured the contracts for all three console companies they no longer had any reason to bother with Apple anymore. Apple and Jobs were a nightmare to work with for IBM compared to other companies.

IBM told Apple to fuck off and that they had no intention of bothering with a mobile G5 class chip. Jobs in a panic turned to PA Semi who didn't have anything ready at the time.

So Jobs finally had to turn to Intel as his 'first choice'.

Yeah, high five Apple and Jobs you really 'made a smart move' there. Effectively being forced into a position where the main selling point of your new computers is they are actually useful to the mass market due to them being able to run your dominant competitor's OS.

Too bad for Jobs being an overpriced niche x86 OEM box manufacturer isn't viable long term.

Re:Apple Got Dumped By IBM.

[pecosdave]

I'm still a bit disappointed they didn't go with AMD. Seriously, Intel seems to "normal" at least AMD would be cheaper, at the time performed better, has an awesome mobile chip, and was doing 64 bit better before Intel. Though I'm not a big ATI fan, considering AMD and ATI have merged, and AMD is really trying to do the right thing with ATI I can see loads of benefit for Apple in the AMD camp.

Re:Apple Got Dumped By IBM.

[ya really]

IBM told Apple to fuck off and that they had no intention of bothering with a mobile G5 class chip.

Actually, it was the other way around.

Jobs stated that Apple's primary motivation for the transition was their disappointment with the progress of IBM's development of PowerPC technology, and their greater faith in Intel to meet Apple's needs. In particular, he cited the performance per watt (that is, the speed per unit of electrical power) projections in the roadmap provided by Intel. This is an especially important consideration in laptop design affecting hours of use per battery charge.

In June 2003, Jobs had introduced Macs based on the PowerPC G5 processor and promised that within a year the clock speed of the part would be up to 3 GHz. Two years later, 3 GHz G5s were still not available, and rumors continued that IBM's low yields on the POWER4-derived chip were to blame. Further, the heat produced by the chip proved an obstacle to deploying it in a laptop computer, which had become the fastest growing segment of the personal computer industry. wikipedia.org

Intel chips outperform the PowerPC cpus without a doubt. PowerPC cpus were horrible. The first MacBook pros with Intel chips were 2-3 times faster than the ones before with PowerPC chips. If anything, it was a good move for Apple to start using Intel. I'm not a huge Mac Fan. I own one Apple product, a Nano with RockBox currently on it. However, I do hate when people don't do their fact checking and simply want to troll about a company they hate without justification.

Re:Apple Got Dumped By IBM.

[Anonymous Coward]

Yeah, high five Apple and Jobs you really 'made a smart move' there. Effectively being forced into a position where the main selling point of your new computers is they are actually useful to the mass market due to them being able to run your dominant competitor's OS.

Great anecdote.. here's another (and this is from a guy running Debian on the desktop): Apple machines, especially the laptops, are selling like hotcakes and I support about a hundred of them along side a few hundred Windows and UNIX boxes. I've seen no more than ten of them using Bootcamp or a VM like parallels. Nobody wants Vista, and after discovering OS X very few have a use for XP outside from an app or two that doesn't yet have a Mac equivalent. Even if you need to run one of these apps you can run them on your Mac desktop via Parallels coherence feature.

Say what you will, but it will all happen again. Last time it was DEC. This time it will be MS. Sorry Microsoft, people aren't going to wait until 2014 for you to get your shit together.

Re:Apple Got Dumped By IBM.

[wish bot]

I remember something from that time that suggested it was simply a supply issue - AMD weren't big enough to guarantee supply. I remember looking at the figures and being surprised (about the capacity of AMD).

I also remember Jobs saying Intel had shown him _very_ exciting things, hint hint. And they were too.

Re:Apple Got Dumped By IBM.

[2ms]

Care to back any of this up with any kind of links or anything? Is this all coming out of your ass or can you show us anything to back up this idea that IBM dumped Apple? Year after year IBM's Gx processors were much much slower than their x86 competitors, so it'd appear you have things backwards.

Re:Apple Got Dumped By IBM.

[Hal_Porter]

The reason people troll is that Apple fanboys were telling us PPC was much faster than Intel right up to the switch, at which point they started telling us they were much slower.

It's like Big Brother fanboys telling you that they have always been at war with Eurasia one day and the very next day that Eurasia has always been their ally. This sort of thing invites trolling and/or rocket bombs.

Re:Apple Got Dumped By IBM.

[ya really]

Heh, that's understandable. I must have missed that argument a few years ago. I typically try to avoid reading/listening to anything written by someone that zealous to a particular brand. It sort of borders on being its own religion. I'd dump any brand in a heartbeat if a better product were issued by someone else. Always seemed somewhat odd to me to stand by a product when it's inferior. If AMD happens to be better than Intel at the time I need a CPU, im buying AMD, and if Intel is better, then there's no reason I'd buy AMD, same with Nvidia/ATI and whatever else. I'd say fanboyism is similar to rooting for your last place team in sports, but you dont necessarly have to pay to watch a team.

Re:Apple Got Dumped By IBM.

[raftpeople]

Intel chips outperform the PowerPC cpus without a doubt. PowerPC cpus were horrible

The PowerPC cpu's were not horrible. I've seen benchmarks over the years showing them outperforming intel cpu's (of the same generation) for some tasks (not all, some). The new architecture for Intel is definately impressive and Apple absolutely made the correct choice.

IBM continues to be the king of the hill at server processors like POWER5,6 and probably 7, but these are targeted at a different market than Apple's customers, and are not the same as the PowerPC cpu's.

Re:Apple Got Dumped By IBM.

[Sparky McGruff]

Desktop PPC chips were great. The performance of a G5 desktop tower was pretty impressive for the day. They just sucked for laptops. Or anything else that you didn't want to use as a space heater. The huge difference wasn't in the desktop space -- a high end G5 wasn't much different in performance from the high end intel chips. In fact, they didn't replace the high end G5's for quite a while after they started the switch; they were the last in the line to go (along with the X-serves). But the difference between a G4 powerbook and their intel replacements was huge, because the G4's were effectively antiques.

Re:Apple Got Dumped By IBM.

[IntlHarvester]

There's two relevant questions:

1) What IBM could accomplish
2) What Apple was willing to pay for

The answer to the IBM/Apple breakup question is somewhere between those two.

Re:Apple Got Dumped By IBM.

[zsau]

What you quote doesn't really backup your story. Saying Apple dumped IBM implies IBM was hurt. In reality, it was probably to IBM's advantage; they were happy to service Apple for as long as Apple was happy to buy from them, but had no particular desire to focus their attention on desktop- and laptop-class chips. IBM was happy to talk about delivering icy cold 3 GHz laptop chips, but that doesn't mean they had any particular desire to put as much money into it as they needed to to get anywhere. So Apple did leave IBM, but they didn't dump them — they just went to someone whose main focus actually is producing icy cold laptop chips. Apple leaving IBM was probably in everyone's best interest, including IBM.

Re:Apple Got Dumped By IBM.

[JamesP]

Well, take a look at this:

http://www.anandtech.com/mac/showdoc.aspx?i=2685&p=11

G5s were faster in several tasks when the 1st intel macs arrived

Re:Finally

[Anonymous Coward]

"Optical interconnects anyone?"

Oh, snap!

Re:Vista capable?

[KillerBob]

*whooosh*

DNF = Duke Nukem Forever

Memory Bandwidth

[Brad1138]

It'll have 620 TB of memory and support 5 PB/s

Is that kind of memory bandwidth possible? You could access the entire 620TB in ~120 milliseconds. I guess nothing is ever to fast, it just seems unrealistically fast.

Re:Memory Bandwidth

[irtza]

I believe the memory is aggregate and so is the bandwidth...so per core memory bandwidth is only 5PB/300K cores/s

The real question is how memory allocation is done in per core - does each core have unrestricted access to the full 620TB or is it a cluster with each machine having unrestricted access to a subset and a software interface to move data to other nodes.

if anyone here has insight on this, please fill in the giant blank.

Re:Memory Bandwidth

[777v777]

It would be incredibly unlikely that each core could directly access the full 620TB. The current largest machines on the Top500 list are all distributed memory machines(clusters). However, the trends in modern interconnect networks are to increase the capabilities for doing stuff like remote direct memory access (RDMA). In such a scheme, the remote memory is not addressable(with load/store instructions), but stuff can be transferred between memories of different nodes by the network hardware. The codes commonly run on the top500 machines are likely written in MPI or MPI/OpenMP. This means they don't need to directly access remote memories.

Re:Memory Bandwidth

[Anonymous Coward]

Is that kind of memory bandwidth possible? You could access the entire 620TB in ~120 milliseconds. I guess nothing is ever to fast, it just seems unrealistically fast.

Well, over 300 000 cores, thats about 2GB for each to transfer in 125ms, or about 16GB/sec each. Considering a high-end Athlon64 can pull about 7GB/sec, this seems very reasonable.

Re:Memory Bandwidth

[irtza]

that's what I thought. I was just curious to see if it had changed without me having heard about it - I don't work on "real" clusters. I was thinking that with 64-bit hardware it would be possible to address memory on another machine that would than halt the process and allow remote memory access hardware kick in to move the data locally - ie a page move - much like a program accessing a page in a swap file. access time across a network would be horribly slow to not move the data closer to the cpu.

Thanks for your input.

Re:Memory Bandwidth

[flaming-opus]

well, in the really sophisticated clusters like Cray or NEC MPPs, you can actually open up a window of memory on a node, and remote nodes can load/store directly to that memory. Obviously it's slower than local memory, but you can program it like a shared memory map.

IBM, however, prefers the hybrid mpi/openmp approach for their clustered smp constellations.

Re:Memory Bandwidth

[mdf356]

It'd be clustered. Here's a few numbers that I think are public domain.

The physical box can contain up to 128 CPUs, with 4 way SMT makes 512 CPUs. The box can have up to 16TB I believe, but of course that's terribly expensive. Given the numbers listed (300k cores, 620TB) it sounds like there's about 483 cpu's per TB of memory, so there's likely to be 586 boxes with a little over 1TB each.

If they're running AIX, there will be one OS instance per box, and they will have some software on top of the OS to manage the whole cluster so it can all solve one problem.

Re:Finally

[Anonymous Coward]

Still can't run Crysis, though.

Re:Vista capable?

[Anonymous Coward]

Yes, thank you, I'm well aware. DNF on Vista is an old joke about how Duke Nukem is taking so long, it will require Vista. Now that Vista is out and there have been games that require it, the joke is outdated. So go whoosh yourself.

I just Want to Cry

[Louis Savain]

Every time I read about the latest multicore processor that uses a parallel programming model based on multithreading, I just want to cry. How many times must it be repeated to the industry that multithreading is not part of the future of parallel computing? Multithreading simply sucks. Concurrent threads are non-deterministic, prone to errors and hard to program. There is an infinitely better way to design and program parallel computers that does not involves threads at all. To find out why multithreading's days are numbered, read Parallel Computing: Why the Future Is Non-Algorithmic.

Re:I just Want to Cry

[Anonymous Coward]

You've got a guy on Blogspot going up against the large number of researchers at Intel who actually designed the chips, as well as researchers who can design and assemble supercomputers and are doing so with the belief that these chips are suitable. I wonder who wins.

Re:I just Want to Cry

[Louis Savain]

You've got a guy on Blogspot going up against the large number of researchers at Intel who actually designed the chips, as well as researchers who can design and assemble supercomputers and are doing so with the belief that these chips are suitable.

Not true. IBM, Intel, AMD, Nvidia, Microsoft and the others do not believe that they have the answer. In fact, judging by the amount of money (hundreds of millions) that they are currently spending in research labs around the world trying to find a solution, it's obvious that the industry is in a state of panic.

I wonder who wins.

It's not over until it's over.

Re:I just Want to Cry

[Anonymous Coward]

I know companies trying to improve their products when they can just sell the old ones is rare, but it does happen, and that's what these companies are doing.

Re:I just Want to Cry

[Louis Savain]

I know companies trying to improve their products when they can just sell the old ones is rare, but it does happen, and that's what these companies are doing.

Nope. They are scared to death because programmers are having a hell of a hard time writing multithreaded software that uses more than just a few threads at a time. It's not easy to keep all the cores busy and balanced. If the industry can't find a solution soon, then nobody is going to buy those 16, 32, and thousand-core devices that are coming down the pike. Nobody wants to se a bunch pretty little cores piling up at the fab, all dressed up with nowhere to go.

Re:I just Want to Cry

[ceswiedler]

It's worse than that. He's the guy on blogspot. I imagine he also has a perpetual motion machine to sell.

To the grandparent: if you want people to start taking your ideas seriously, I recommend you stop talking about them like you're crackpot.

Re:I just Want to Cry

[Warll]

Thats what happens when you have to bring a new product to the market place a number of times over the course of a financial period. If they were to cut R&D right now they'd shadow of their former self long before those stock options matured.

Re:I just Want to Cry

[Max Littlemore]

You've got a guy on Blogspot going up against the large number of researchers at Intel who actually designed the chips, as well as researchers who can design and assemble supercomputers and are doing so with the belief that these chips are suitable. I wonder who wins.

History is absolutely full of people who don't follow the mainstream theory or have financial backing and end up creating the next mainstream theory which receives all of the financial backing.

History is also full of people such yourself, AC, who poor scorn on non-conformist ways of looking at things and end up looking like fools.

Maybe he has a point, maybe he hasn't, but whether or not he is in the mainstream has little or no bearing on the validity of his thought.

Re:I just Want to Cry

[Louis Savain]

Maybe he has a point, maybe he hasn't, but whether or not he is in the mainstream has little or no bearing on the validity of his thought.

Well said. I want to point out that I am not the only one who's been saying that multithreading sucks. Donald Knuth has been saying the same thing. So have a whole bunch of other people. What I am adding to the fray is that there is an easy way to do parallel programming without threads that guarantees deterministic parallelism and easy programming. I am not asking for anybody to give me money. I just want to see the industry do the right thing.

Re:I just Want to Cry

[Anonymous Coward]

So if you're such a perfect genius, why aren't you designing these chips for one of the big companies instead of running your mouth about how you're the only person who understands parallel programming properly, and that all the engineers at Intel, AMD, IBM, Sun and so forth are idiots?

Oh right, you probably have no clue what you're talking about.

Re:I just Want to Cry

[tuomoks]

Maybe not scared to death but concerned. Mostly, of course, for financial reasons but there also are some very bright people who would like to make computing more efficient.

Now - multiple processors / cores is nothing new, I hit those in 70's and loved them. Unfortunately the education and training didn't follow - more geared to basics, algorithms, structures, languages, etc than what, how and why? And it is even worse today - a product specialization and certificates are what pays - not performance, architecture, design, security, whatever!

Almost any computer problem can be broken to parallel solutions and execution - when was the last time that was in job requirements, you know those things which make you employed?

Re:I just Want to Cry

[Anonymous Coward]

I notice there are a lot of links to a parked domain from your blog. http://www.rebelscience.org/Cosas/COSA.htm goes nowhere interesting.

Also, how would self modifying code work?

Re:Finally

[Matey-O]

Naw man, my laptop has a sticker that says it's 'Vista Capable'.

Re:Sorry Mac fans . . .

[Anonymous Coward]

it (power7) probably only has VMX

5.0ghz Power6

[Anonymous Coward]

is already available.

In other news...

[billybob_jcv]

...IBM also announced a performance upgrade kit for the Power7 to enable it to meet the minimum HW requirements for Vista.

Re:In other news...

[drspliff]

While Microsoft have long since dropped Power support from their desktop & server lines, the stripped down 2K/XP kernel for the Xbox360 runs on a multi-core PowerPC chip. Considering Windows 2003 & 2008 Server both support NUMA to some extent it could well be possible.

Think about it... 64 processors with 8 cores each, you could run Vista AND Word at the same time!!!!!

Re:Finally

[Anonymous Coward]

A beowulf cluster of those should be able to...

Re:Finally

[lastomega7]

But it turns out it will only actually run the home edition smoothly.

620 TB of memory?

[xbytor]

Surely no one would ever need more...

Intel like make apple a unfair deal to get there..

[Joe The Dragon]

Intel like make apple a unfair deal to get there chips in to apple systems.

The AMD lawsuit came out at just about the same time as the Intel apple deal.

Intel was not that good at the time and some of there chips that got used in macs where not even 64BIT at the time.

Intel chip sets sucked next the one for the amd chips. The Power Mac g5 had more pci-e lanes with a better setup then Intel had.

AMD had dual cpu boards with a better cpu to chipset link with more pci-e lanes and sli as well DDR1 ECC / DDR2 ECC not FB-DIMM like intels board.

With AMD you can uses any NB/SB setup unlike the dual cpu intel boards.

AMD MB with ATI and NVIDIA chipsets had much better on board video then intel had that the mac mini g4 said was bad next to it and there where Intel MB with NVIDIA and ATI on board video at the time as well.

The intel chipsets at the time also had a 3gb ram limit that is still there even with a 64bit cpu.

The mini still uses the same on board video chip.

Numa Numa

[Gothmolly]

Where's that NUMA NUMA kid when you need him?

Re:Numa Numa

[BiggerIsBetter]

Where's that NUMA NUMA kid when you need him?

New Jersey.

Re:Numa Numa

[tietokone-olmi]

Must've been cut by razor blades.

Re:Numa Numa

[BiggerIsBetter]

Where's that NUMA NUMA kid when you need him?

New Jersey.

I weep for Slashdot, that mods do not get a NUMA joke from 2004.

UofI machine is a bit low on memory

[yorkshiredale]

That University of Illinois machine sounds like it needs more memory.

Only 620TB? Why not bump it up to 640? That should be enough for anybody.

Re:UofI machine is a bit low on memory

[hkfczrqj]

Being a grad student at Illinois I can tell you something. You really don't know about the University's accounting system. It can literally index every atom in campus (not that they need to). That's why 640 won't be enough :) Also, the supercomputer will require the construction of a new power plant. Seriously.

Re:UofI machine is a bit low on memory

[matty619]

No Body will EVER need more than 640 TB of memory.

Looks like 3D Realms

[Anonymous Coward]

.... have tipped IBM off about the minimum requirements for Duke Nukem Forever....

Re:Looks like 3D Realms

[cptnapalm]

By the time DNF ships, IBM will have this amount of processing power in your wristwatch.

Re:Finally

[marafa]

optical connections are the nerves between your eyes and your brains. if you disable them you wont be able to watch the videos let alone your girlfriend!

Re:Finally

[FlyingSquidStudios]

My margarine tells me that I can't believe it's not butter... but I can.

Re:Finally

[failedlogic]

I won't be fast enough for Windows 7.

Will it be a hybrid like Roadrunner?

[stardude82]

Using x86 processors more or less like controlers?

Will it run?

[ireallylovelinux]

Will it run lin.... oh forget it.

Re:First!!!!11

[n9hmg]

Yes, it will. And let me, for the first time in my life, say "Duh!".

Re:Finally

[RobertM1968]

optical connections are the nerves between your eyes and your brains. if you disable them you wont be able to watch the videos let alone your girlfriend!

Girlfriend? Who on Slashdot has one of those? C'mon! You must not visit here often!!!

;-)

Re:Finally

[Anonymous Coward]

Naw man, my laptop has a sticker that says it's 'Vista Capable'.

But is it 'usable'?

Sure there is...

[Nick Driver]

Is there any way to force a process to run over 2 cores at 100%?

Sure there is. Just install Oracle Database Server on it and hit it with some poorly-written queries over an ODBC connection.

What? Four threads per core?

[yellowstone]

I thought 1 core == 1 thread of execution?

Or are they talking about some kind of extra hardware support for multitasking?

The compiler

[symbolset]

And the libraries are where they'll be working this one out. They can't really expect the average programmer to handle this any more than they can count on him to implement a hashing algorithm.

Maybe we'll get Yet Another Fine Programming Language out of it -- though several of them were designed for this from their conception.

Re:The compiler

[Detritus]

Decades ago, I read an excellent book on Concurrent Pascal by Brinch Hansen. I was sure that similar features would soon become a standard part of modern high-level languages. I'm still waiting.

Maybe in a laptop...

[richardkelleher]

Can I get a pair of those in a laptop? I'll pick up a pair of Nomex shorts a the local Nascar outlet store!

Not a contradiction

[Weaselmancer]

Nah. If something gets warmer it is caused by Global Warming and the solution is to eliminate Western industrial civilization.
If something gets colder it is Global Climate Change and the solution is to eliminate Western industrial civilization.

Not a contradiction, even though it seems like one.

Study the bifurcation diagram. As you drive the system harder by turning up R (which may be analogous to global warming - i.e. more available heat energy might be described this way) notice how the system follows R, then suddenly begins oscillating between two extremes. Keep on driving R harder and it breaks into chaos.

The weather IMHO has a lot in common with the logistic map equation. It's present behavior is dependent on it's past state, it's swings are driven by the energy input to the system, etc.

I know it's a gross oversimplification, but so is a mass falling through a uniform gravitational field with no wind resistance and so on. It's still useful to think about.

Re:Not a contradiction

[jmorris42]

> Not a contradiction, even though it seems like one.

I'm not arguing that there isn't global warming. I'm not even arguing that human activity isn't influencing it. What I am arguing is that our science is nowhere near good enough to say a goddamned thing on the subject even if it were actually a scientific issue and not a political one.

How many computer models in the early 1990s predicted the ten years of flat/cooling we have been experiencing? Or better, point to a computer model that has shown accuracy and bet me $100 it's current prediction will be true in ten years. You would be a fool to attempt it. This is a very basic point, if you don't have reliable numbers it isn't science. Until we have models that can not only show the past, but make a prediction for the future that actually comes true we know zero. And one accurate prediction doesn't even come close to rising out of the error bars.

As for consensus, there ain't one and it wouldn't matter if there were, science doesn't work that way, politics does. If GW were about science Hansen would have been sacked the second he was proven to have cooked his numbers. But since this issue long ago crossed over to politics lies are par for the course, if we sacked every lying politician Washington DC and every state capitol would be depopulated.

Re:Not a contradiction

[Weaselmancer]

I'm not arguing that there isn't global warming.

Yup. I'm not arguing there is. You've hit the nail on the head exactly. Soon as this became a political talking point anything resembling science went right out the window. Now there's no way to tell. Is this latest report good science, or is it political propaganda? No way to know thanks to people in Washington needing something new to argue about.

It's a shame, too. It might really be happening and by the time the Left vs. Right noise machine quiets down about it - it could be too late.

PS: No bet on the weather prediction. Lorenz proved that impossible back in the 60's. :)

Re:Not a contradiction

[Anonymous Coward]

(Posting anon since I've used oodles of mod points this thread already)

What's always baffled me about the whole attempting to model global warming issue is that weather is the quintessential "chaotic" system: Lorenz's original studies, the butterfly effect, all that jazz. And yet there's a vast amount of effort put into studying/debunking various "models" of the climate system, despite the fact that it's likely tremendously unstable to even tiny shifts in initial conditions or unknown parameters (which seem to be turning up every few weeks). The demand for truly "reliable" predictive numbers may never be met, simply because they are impossible to produce.

Of course, the problem is that I couldn't imagine how to approach the problem from a different angle - science is about taking observations and building models. Unfortunately, in a system where even medium-term predictions rapidly become impossible, I feel the fixation in some quarters on trying to out-model one another is... counter-productive, at best.

Re:Not a contradiction

[Anonymous Coward]

Its just political. Global warming is big money and all the companies are slowly understanding that you can go green, patent it and then make the government force everyone else to go green and pay you fat sums of money for the patents you have that are required to compete in your industry.

1. Patent going green
2. Get the government to pass laws
3. Push all your competitors out of business or leech off their earnings
4. Profit

Re:Not a contradiction

[Muad'Dave]

Slightly OT, but the definitions of bifurcations look an awful lot like IIR filters, a DSP construct. They're recursive filters that can exhibit unruly behavior as compared to FIR filters.

Re:Not a contradiction

[Weaselmancer]

That's why chaos math is useful. You don't really need to have long-term predictive numbers. It's really more about studying the dynamics of the system. How it responds in the short term to this and that. Which values drive the system and what shifts of behavior you can expect to see.

It's not (IMHO anyways) really intended to be predictive. It's really more descriptive.

For instance, you can come up with a Lorenzian system that has the massive sensitive dependence on initial variables. You can let it run for weeks on end and see parts of phase space the system never heads to. Then halt the system and restart. Ten minutes from now you won't be able to tell where your system will be, but you know where it might be, and where it won't be.

Chaos math isn't classical "plug the numbers in the equation and get an answer" math. It's about dynamics. It takes a slightly different mind set to be useful. Like when you first learn calculus. Rather than using equations, you have to think about deriving them. It requires you to rethink how you approach problems.

Re:Not a contradiction

[Hes Nikke]

Or better, point to a computer model that has shown accuracy and bet me $100 it's current prediction will be true in ten years.

everyone knows that the dollar will be worthless in ten years, the same people who are pushing global warming on us are also pushing the amero on us. now if you had bet gold, you would be betting something that holds it's value ;)

Re:Not a contradiction

[Anonymous Coward]

Climate and weather are very different. Climate is influenced by things like the positions of the continents, the functioning of the ocean currents (which are themselves mostly stable for a given continental position), and many other things, including the composition of the atmosphere.

Here's a simple example about weather and climate. Weather is like having a meteor fall on your head. Climate is like the years of raining ash because of it. Climate is in fact "easy" to reason about, unlike weather. For example, these ocean currents are mostly affected by the position of the continents, wind, and thermal gradients, both in depth and latitude. In the Atlantic, the hot Gulf Stream, driven by warm winds, tends to flow north along the surface. When it reaches the Northern Atlantic, it gets dramatically colder and sinks. This is a so-called thermohaline circulation.

A rise in sea temperature slows this process, and indeed, it can even stop. This has lead to oceanic stagnation (deep water anoxia and subsequent rotting of deep water organisms) and mass extinction (from poison gas released by stagnant oceans -- just as they are released from stagnant seas and lakes today).

I'm not arguing that this is going to happen to us, or any time soon (though it probably will happen again in the next few million years). What I'm trying to get at is that there is nothing hypothetical about this. Geologically significant events have the power to change climate, and the exponential growth of the human race is certainly geologically significant. (Significant enough to call this period in geological time the 'anthropocene'.)

Re:Finally

[Anonymous Coward]

True. The entire purpose of new versions of Windows is to make people buy new computers.

a little more than 4000 sq. feet

[Profane MuthaFucka]

Clearly we have a long way to go before we have a planet sized computer. However, I know a website where we can find plenty of robots to complain about it.

University of Illinois Funds?

[tyrione]

So that's where some of that fraud money has gone to, eh? I found this envelope and all it had was a return address to Barack.

I couldn't resist.

Desktop PCs

[Lord Lode]

This CPU appears to own Intel and AMD's. Can it be used in desktop PC's?

Re:Desktop PCs

[RightSaidFred99]

It doesn't own anybody. It's competition isn't x86 processors - it's far more expensive, uses more power, and has a non-mainstream instruction set. By the time this is out it will be competing with a very mature Nehalem chip - and not looking favorable in anything but raw performance in specific tasks.

Re:Desktop PCs

[Ilgaz]

Non mainstream instruction set? The Power instruction set? Of course it will have lots and lots more and specialised commands but it is still a POWER CPU. I bet it will run gcc too.

Look how MS solved the XBox 360 early developer access issue: They ordered thousands of its distant cousin, PowerMac G5

Re:Desktop PCs

[Anonymous Coward]

Of course it will run gcc. PowerPC has been supported by gcc and Linux for at least 10 years. (The instruction set is called PowerPC. POWER7 is IBM's latest implementation)

Re:Desktop PCs

[krischik]

Note that Power without PC is a server CPU like Opteron or XEON. But still they share the same instruction set.

Martin.

ADA has it

[symbolset]

And APL always did. There are many others. Now that the architecture of the processor is moving back to the '80s, only with multiple cores, we need a technology archaeologist to dig back through back issues of the Communications of the ACM from the 70's and 60's and show us where we went astray.

Re:Finally

[Anonymous Coward]

no no no,
That's an enterprise class Power Mac ya got there. For the first time ever, Apple Mail will be a responsive application.

Re:Finally

[anexkahn]

barely

Memory

[Anonymous Coward]

620TB should be enough for anybody /me waits 20 years

Re:Intel like make apple a unfair deal to get ther

[Anonymous Coward]

You're retarded. Intel showed Apple the Core2 chips coming out and it was game over for AMD. Looking back, they kind of made the right decision.

Re:Finally

[JorDan Clock]

Right. That's the entire reason. Not to offer new features or anything.

Re:Sorry Mac fans . . .

[Ilgaz]

Funny is people actually thought IBM can't deliver 3 Ghz or cold running G5. No, they just chose not to deliver it to Apple. Their focus is enterprise, servers, massive scientific computing. The early warning came when they sold their superbly prestigious and brand advertising Laptop division to Lenovo.

Just imagine they cancel this CPU to deliver 3 Ghz G5 to Apple. For what? Apple fans turned x86 fanatics almost overnight happily buying parallels to run Windows applications on OS X and buying overpriced Windows games which are masked as OS X applications.

At least IBM and Apple took away the "endian" excuse in hands of developers and GPU vendors. They still, shamelessly sell 20-30% more expensive graphics cards to Mac users, running Intel, on standard PCI-X mainboard! New excuse is... EFI!

Re:Finally

[Ilgaz]

There is kind of a sense that MS have seen the fact and shaving off features (!) on Windows 7. Well, if they don't after Vista disaster which even their friends like Intel refuses to upgrade, there won't be Microsoft in 5 years. Serious.

Photoshop

[xbytor]

I use photoshop. Maxing out it's memory on XP is a trivial matter, especially if I have Firefox, Thunderbird, xemacs, and a few bash sessions running at the same time. This may be the one thing that pushes me to get a big expensive Mac unless I find out that XP64 has stabilized to the point that I can convert over without any problems.

-X

hmmm...

[RancidMilk]

> ./install vista
I'm sorry Dave, I can't let you do that.

> ./man why
insufficient intelligence

Re:Sorry Mac fans . . .

[Anonymous Coward]

At the time, they couldn't. Even their server/computing lines didn't run that fast. POWER4 also wasn't designed with power consumption in mind, so it's no wonder they never managed to get it down to acceptable levels. There's a reason the cooling solutions in Apple's G5 are about the same size as what you usually find cooling the reactors in nuclear power plants.

There IS a Human Cause to Global Warming!!!

[dreamchaser]

Global Warming IS real and IS caused by humans. Well, by one human. All of Al Gore's bloviating on the subject has created such a huge amount of hot air that it has upset the balance.

Save the planet! Stop Al Gore!

Re:Finally

[larry bagina]

If shaved windows is anything like shaved pussy, it'll be a good thing.

Now we know what Windows7 will need

[gelfling]

2 of these.

Reliability

[TooManyNames]

The interesting thing about this (and any other supercomputer like it) is how much time must be spent checkpointing. In order to maintain consistent state across the entire machine, up to 40% of its computational power must actually be spent checkpointing (not to mention the memory overhead associated with transferring state to stable storage). I'd imagine that the scientists who will use the new supercomputer will be slightly dismayed that they'll frequently have access only to about 60% of the processing power.

Really? Guess he switched that link on you, then

[Anonymous Coward]

Right now, that blog links to
http://news.bbc.co.uk/1/hi/world/americas/6995999.stm
and I see nothing about previous passages there.

Vista Capable Hardware

[MoonLyn]

No, it isn't Vista Capable, Moron. Vista runs on x86/x86_64 architecture when this machine is PowerPC architecture...

Re:Intel like make apple a unfair deal to get ther

[Joe The Dragon]

AMD still had the better 2 cpu + systems and they used cheaper ram. Amd also was open to useing any chip set as well.

*drool*

[Anonymous Coward]

I wonder how long it would take to compile a linux kernel on that thing...

imagine it: 'make -j 300000' *rapid scrolling of output* *Bing!* Total time: Time spent typing make command + 1 6/1000ths of a minute.

Re:Finally

[Anonymous Coward]

You know, when you get your ass handed to you like that, and you come back with, "Oh yeah, well, uhhh, yer a fag!" It makes you look like even more of a tool. Face it, you lost that exchange badly, Profane MuthaFucka.

Disagree

[TheLink]

There's at least one other reason.

If Microsoft stopped releasing new _slightly_ incompatible versions of windows for long enough, someone will likely catch up and make a Windows compatible O/S.

Then Microsoft to Microsoft Windows will be like IBM to IBM PC BIOS.

For example: if people really stuck to Windows XP + DX9 for say the next 5 years, in that time WINE will come up with very good XP and DX9 compatibility, then someone could presumably package it nicely enough to release as a "drop in" replacement for Windows - different theme etc, but works like Windows XP.

If that happens Microsoft would lose significant control, just like Intel has lost significant control over where x86 is heading - they tried to EOL x86 and push the market onto the Itanic, but AMD produced AMD64 and so Intel had to do something like AMD64 (EMT64).

WINE is already pretty darn close. Probably too close for Microsoft's comfort. Microsoft might sue for patent infringement etc, but it is not certain they'll win, or win in all countries.

Re:Disagree

[poolmeister]

You mean like ReactOS?

Re:Disagree

[TheLink]

Yes. That was what I was thinking of.

But note that: "the user-mode part of ReactOS is almost entirely WINE-based". ( http://www.reactos.org/en/about_whatisreactos.html )

So if Microsoft takes out WINE it will probably take out ReactOS too.

Re:Finally

[Profane MuthaFucka]

I never post as an AC. Why would I need to? Profane Muthafucka isn't exactly on my driver's license...

Re:What? Four threads per core?

[argent]

They're talking about hyperthreading, like the Pentium 4. If one thread is waiting on a cache miss another thread that already has everything in cache will get to run.

Re:Finally

[Anonymous Coward]

Some vista capable hardware.

Nah, it's the new Mac Pro. 300k cores, 620 TB RAM, and an ATI 2400 Pro.

Re:Finally

[Anonymous Coward]

but will it run Crisis?

Re:Finally

[painehope]

Yeah, but the real question is - do you want to watch the videos of his girlfriend?

Just give me your CC # and something can be arranged...

Re:Finally

[LWATCDR]

Or the first machine that can run FlightSimX at 60 FPS

Re:What? Four threads per core?

[yellowstone]

OK, thanks for the reply.

Re:Finally

[MoonLyn]

It is not Vista Capable Hardware, you moron! It is PowerPC architecture whilst Vista is only built for Intel x86 and x86_64 architecture! So it IS neither VISTA CAPABLE HARDWARE nor even XP CAPABLE HARDWARE