Cell-based Server Blade Demonstrated

slashflood writes "Only a few clients in a hotel room near Los Angeles had the chance to see the first Cell based server blade running Linux 2.6.11. 'We demonstrated the prototype to show that Cell continues to mature. The product is expected to have several times higher performance compared to conventional servers,' said an IBM engineer."

Comments

Great

[pHatidic]

Now we'll have to put up with people's web servers ringing in movie theaters.

Re:Great

[lheal]

Surely IBM, with its vast army of technical wizards, will put a vibe-only feature in the firmware.

I donno, though, those cards look awfully big to hang on a belt clip. Maybe they come with a nice IBM tote bag or something.

Re:Great

[Anonymous Coward]

Probably because Linux is about the only OS (or kernel, rather) nimble enough to be ported to an entirely new architecture in a few days or weeks.

Re:Great

[Anonymous Coward]

Oh yeah like FreeBSD, right? Oh oh, maybe you're a solaris zealot?

Or is it Windows XP? Yeah that would be so great!

No wait, it would be Windows Longhorn 2007 XP special cell HPC supercomputer enterprise reboot-instead-of-BSOD XP awesome home edition.

Cool.

Re:Great

[Anonymous Coward]

stfu n00b

Re:Great

[AmberBlackCat]

When I was in some of my CS classes, we would occasionally tell jokes and I'd tell my friend, "nobody outside of this classroom could possibly find that funny".

That was before I knew about slashdot.

Re:Great

[Anonymous Coward]

How about NetBSD?

Porting Linux to new archs tends to be fast, but incredibly sloppy compared to NetBSD.

Re:Great

[DrSkwid]

lol, clueless = you

Re:Great

[Anonymous Coward]

Err, no.

Re:Great

[Anonymous Coward]

Anyone who uses the "word" 'n00b' can lick my anus and suck on my balls.

Re:Great

[Anonymous Coward]

Yeah, I'm sure NetBSD would scale to the enterprise on a Cell based enterprise class server.

What's BSD upto now with SMP? 2, maybe 4 CPU's? Watch out 1995!

Re:Great

[Anonymous Coward]

You are trolley an ass.

Re:Great

[yukonc]

+ 80 points

Re:Great

[Anonymous Coward]

Last time benchmarks were posted, FreeBSD 6.0 scaled to about 3.8 CPUs worth of useful work with a very parallel workload on a 12-CPU system, after being extensively tuned by a FreeBSD kernel hacker.

And it scales to 2 CPUs on filesystem heavy workloads thanks to the "mp safe VFS" work.

NetBSD I think completely single threaded, so it will probably be worse than FreeBSD. It will probably scale to between 1 and 2 CPUs on most workloads. There are some I have seen (eg. recent MySQL benchmarks) where going from 1 CPU to 2 results in performance decrease too, though.

Is an IBM response about Apple isue?

[Sr. Malagon]

Or to say, "hey, powerpc is not only for games"

Re:Is an IBM response about Apple isue?

[Anonymous Coward]

He is talking about iSue, Apples new software for lawyers. Either that or it's Excited First Post Syndrome.

Re:Is an IBM response about Apple isue?

[wezzy]

probably he's talking about rumors about the apple swtich to intel's processors. Many observers thinks that apple create that rumors to make pression on IBM because they move many attention from PPC970 to console's tecnology.

Re:Is an IBM response about Apple isue?

[Anonymous Coward]

Why not put CELL in iPod? Since iPod already is UNIX.

huh?

[TimeForGuinness]

Only a few clients in a hotel room near Los Angeles had the chance to see the first Cell based server blade running Linux 2.6.11.

sounds like a drug deal going down.

Re:huh?

[mix_master_mike]

Not a drug deal - this was an illegal exchange of human embryonic stem cells.

Re:huh?

[TimeForGuinness]

Not a drug deal - this was an illegal exchange of human embryonic stem cells.

lol. that's right,it's in LA. Maybe they were looking for a cure for Arnold's tumor.

[Arnold voice] IT'S NOT A TOOMOR!!! [/Arnold voice]

Re:huh?

[Anonymous Coward]

You can always spot the idiot forum refugees by the square brackets.

Re:huh?

[grammar fascist]

Not a drug deal - this was an illegal exchange of human embryonic stem cells.

This is a fine example of a joke that is only funny to the ignorant and willfully stupid.

How many times do you need it pointed out that the only "limitation" on embryonic stem cell research is that it can't be federally funded? That's it. Nothing's illegal. It makes sense to not fund something with taxpayers' money that a significant portion of the taxpayers have severe moral issues with.

I'm offtopic, sure. Mods: make sure you mod down the parent poster as well.

Re:huh?

[ozmanjusri]

It makes sense to not fund something with taxpayers' money that a significant portion of the taxpayers have severe moral issues with.

As opposed to the billions of dollars being spent on the war in Iraq which, presumably, no taxpayers have problems with?

Re:huh?

[Anonymous Coward]

{No shit?}

Re:huh?

[hostyle]

Why would they? Americans love to sit around the TV and watch a war - live!

Re:huh?

[jo_ham]

The problem is, of course, that you can't mix and match - if you do any stem cell research in a lab, then that whole lab (and any other research unconnected with stem cells is /also/ denied federal funding).

The only way to ensure that just the stem cell research is denied the funding is to set up a separate lab, which is expensive and difficult.

The blanket ban, driven by religious motives primarily (like the whole religious crusade against science in the middle and dark ages [earth is flat, world created in 7 days etc etc]) makes it very, very difficult to carry out stem cell research alongside other federally funded projects.

Re:huh?

[Anonymous Coward]

We're not actually allowed to watch the war. We only get "sanitized" updates.

I think war, in all its bloody horror, should be broadcast. I'm tired of this "oh, we can't show that, it's too graphic" bullshit. Too many of our sons and daughters, mom's and dad's, and friends are over there fighting (and dying) for us to ignore our duty to know what is going on. Just like there's a big difference hearing a lecture on WW2 and watching Band of Brothers or Saving Private Ryan, there's a big difference hearing a newscaster tell you how bad it is, and seeing it for yourself.

But no, that would be bad, because then most people would be anti-war, all the time! Information is bad!

Re:huh?

[Anonymous Coward]

LOL.. the Op get's informative and you get offtopic for responding. True Liberal Groupthink. I love this game !!
That is why Dems continue to lose their grip on this country.

Re:huh?

[Anonymous Coward]

It makes sense to not fund something with taxpayers' money that a significant portion of the taxpayers have severe moral issues with.

The problem with this argument is that for any particular government expenditure, you can always find someone who will have a severe moral issue with that particular expenditure. Your argument therefore reduces to one where a single person should be allowed to veto the majority's decision to fund any particular program. Democracies work by majority rule subject to the protection of certain "inalienable rights" that majorities cannot take away. A personal veto over expenditures is not one of those inalienable rights. How are you gonna feel when someone who has a moral objection to locking people in prison vetos all expenditures on prisons?

Re:huh?

[petermgreen]

it seems to me that both the original poster and the response were at 50% informative 50% offtopic.

i think the kind of people who are likely to mod down are the kind that browse at +2 or +3 so if someone mods an offtopic post up to that level (especially +3 and above) or it gets there automatically becase of karma bonus its likely to get modded down

if you are making offtopic posts then use the no karma bonus option to reduce the visibility of your posts to those likely to mod down.

Re:huh?

[Bob Uhl]

The blanket ban, driven by religious motives primarily (like the whole religious crusade against science in the middle and dark ages [earth is flat, world created in 7 days etc etc]) makes it very, very difficult to carry out stem cell research alongside other federally funded projects.

And those damn pesky religious nuts are holding us back from other forms of experimentation on humans, too. Damnit, we need to test drugs out on prisoners--it for the children!

And of course science in the mediaeval period was kept alive by the Church, but why inject facts into one's bigotry?

Re:huh?

[Skjellifetti]

And of course science in the mediaeval period was kept alive by the Church, but why inject facts into one's bigotry?

Science was not kept alive by the Mediaeval Church. Quite the contrary. The Church sought to supress anyone who dared experiment with ideas that differed from the Church's orthodoxy. What the Church did was to preserve a portion (we don't know how large a portion) of the discoveries of the Greeks and Romans and other earlier civilizations. But mere knowledge preservation is not at all the same as science.

Re:huh?

[ABaumann]

You might have misread the sentence that you quoted. See, it says "a significant portion of taxpayers."

Re:huh?

[aichpvee]

like the whole religious crusade against science in the middle and dark ages [earth is flat, world created in 7 days etc etc]

They're still doing that.

Re:huh?

[Anonymous Coward]

Yep, we in California are more than happy to watch you pinheads in the rest of the country torpedo your industries. We're pretty used to backwaters that can't manage to hold down high-tech, but we never thought we'd nail biotech too. Thanks a zillion, Georgie.

But gollll-EEE, thanks for settin' us straight mister!

Re:huh?

[Anonymous Coward]

Really? And how many Americans do you personally know that love to do that?

Re:huh?

[DeadChobi]

Actually, a lot of the science that is still considered fact and was discovered during the middle ages was originally pioneered by monks. The rudimentary study of genetics conducted by high-school students in biology(in my state, at least) was originally pioneered by a monk who spent years of his life dedicated to breeding various flowers to see if he could get different traits out. The church has actually been responsible for a lot of the foundation of what we now consider fact.

Re:huh?

[Skjellifetti]

Are you saying that the Middle Ages lasted into the 19th Century? No, the Church has been responsible for very little science and is well known for its persecution of real scientists like Galileo. Check out the list of books which Catholics were forbidden to read (which included works by Francis Bacon, the father of the modern scientific method) until Pope Paul VI finally dropped it in 1966.

Now we just need to ask it tough questions!

[Anonymous Coward]

Maybe we'll get better answers than "42".

I wonder if anyone knows how close we are to the power of the human brain yet.

Re:Now we just need to ask it tough questions!

[UniverseIsADoughnut]

42.0

Got to use the floating point power.

Re:Now we just need to ask it tough questions!

[Friar_MJK]

Ok, so people ask when we'll be able to download the contents of our brain, or when the CPU will come close to our brains capabilities and what-not. Well how exactly do we know how powerful our brain is? Who gets to judge that? Humans have love, jealousy, greed and the like; some emotions are chemical responses, while others come from our brain. But we can't mathematicaly calculate pi, or vertex shaders and polygons (really don't know what I'm talking about with the graphics, but you get the idea). What do they judge our brain power on? Could anyone provide some insight on this? I'm highly interested.

Re:Now we just need to ask it tough questions!

[r_jensen11]

I'd rather 42.00
The more significant figures, the better

Re:Now we just need to ask it tough questions!

[MyNymWasTaken]

I wonder if anyone knows how close we are to the power of the human brain yet.

How do measure the computational power of the human brain?

Here's a 6 year old napkin calculation.

They give a figure of 10^8 MIPS. Figure 1:8 for a MIPS:MFLOPS ratio. So ~13 TFLOPS.

The IBM Blue Gene/L is the current record holder at 135 TFLOPS. That puts it at the power of 10 human brains if that napkin calculation has any validity.

For average consumer computers...

The ordinary computer of Aug. 2004 performed 18,000 MIPS. Ref

Human brain power is ~12.44 Moore's law cycles away from that point. That gives 19-25 years.

So, your computer should be more powerful than your brain by 2030.

Re:Now we just need to ask it tough questions!

[PakProtector]

The Computer is a General Purpose device. The Human Brain is Not.

I think your calculations need to be revised.

Re:Now we just need to ask it tough questions!

[Anonymous Coward]

I've actually heard the complexity bruited about as more like 100PFLOPS, given that it's not just the number of cells, but the number of axons, dendrites, receptors, and neurotransmitters.

By that standard we're still ~15-20 years away from having a supercomputer that'll hit it, and another 10-15 years after that for it to be "table top" material (assuming you don't get a singularity effect from the first one... then you're looking at ~36 months to effectively infinite computing power)

the sum of (18/n) N from 1 to infinity = 36, right? if my math isn't too bad...

Re:Now we just need to ask it tough questions!

[Anonymous Coward]

But those arn't Significant figures, they are insignificant figures. Trailing zeros...

Re:Now we just need to ask it tough questions!

[UniverseIsADoughnut]

no, they are just as significant as if they were 1s or 2s.

There are only insignificant if it was a case of 4.0 x 3 =12.0 in that case the 0 isn't significant. Since the answer is actually 12

Re:Now we just need to ask it tough questions!

[jericho4.0]

Comparing MFLOPS to intelligence comes from the same school that promised that true AI is "just around the corner" for 30 years.

The mind might indeed be a Turing machine, but it's a very different architecture and OS than the ones we know about.

Re:Now we just need to ask it tough questions!

[Creepy Crawler]

41.999987448205

Intel Inside (the warning, NOT the logo!~)

Re:Now we just need to ask it tough questions!

[LnxAddct]

Read up on neural networks :) The brain can be measured more accurately in complexity of its connections. When we can start simulating neural networks with trillions of connections all running in parallel and sending signals a few times a second, then we'll be there. Not only would the computer be aware like we are, but itd think faster too, i.e. it would realize something is happening or the proper action to take faster then we currently can.
Regards
Steve

Re:Now we just need to ask it tough questions!

[guile*fr]

I'd say it's the other way around. the computer is just an idiot that can count really really fast.

Re:Now we just need to ask it tough questions!

[admactanium]

So, your computer should be more powerful than your brain by 2030.

with the handy process of "aging" i think i can actually meet that goal earlier if the computer and i just agree to meet in the middle.

Re:Now we just need to ask it tough questions!

[nietsch]

knowing The question and The answer are both mutually exclusive. If the two ever meet in the same person, their will anhilate eachother. The resulting energy will destroy the universe, but only to have it replaced by something more inexplicable. Others say this has already happened.
(rough quote from the tertiary phase)

Re:Now we just need to ask it tough questions!

[TopSpin]

I've actually heard the complexity bruited about as more like 100PFLOPS, given that it's not just the number of cells, but the number of axons, dendrites, receptors, and neurotransmitters.

Numbers like that arouse my skepticism. I wonder if there is any credible way to extrapolate computational capacity from power consumption; can 3lbs of grey matter powered by sugar possibly approach 100PFLOPS without vaporizing? Just how many FLOPS must be performed per synapse to do that?

Please, put away the pitchfork; I'm not making any claims. IANAB, and my credibility with regard to biology approaches zero. However, allow me to cite a paper that presumably has some credibility.

The most interesting part of that work, IMO:

There are considerations other than sheer scale. At 1 MIPS the best results come from finely hand-crafted programs that distill sensor data with utmost efficiency. 100-MIPS processes weigh their inputs against a wide range of hypotheses, with so many parameters, that learning programs adjust better than the overburdened programmers

Nature has had billions of years to write code. If you gave a large number of researchers several thousand years to squeeze maximum performance from 100-MIPS, what would you have?

To what degree do our highly discreet, digital methods limit our ability to emulate the mind? Brains are not discreet, precise mechanisms. Claiming to be able to quantify the capacity of a system we do not understand is highly suspect.

Yet, consider our progress. Today, we market mobile phones with speech recognition as incentives to sign up for service. Once the algorithm was understood we reduced it to a trivial "feature" using poor quality sensors and a few milli-watts of power. Hidden Markov Models make this possible, and I'm still astonished by it.

Simple introspection tells me a great deal about "memory." Our minds have amazing spatial memory; you can probably remember the layout of every place you have ever lived and most places you have visited. I can remember the significant aspects of every Quake map I have spent more than an hour running around "inside." How? Compression, probably. Our brains have managed to reduce spatial data to a tiny fraction of what we consider necessary. The algorithms involved are far beyond our understanding.

I believe, in the end, we will find our minds are actually low capacity, computationally inefficient systems that run profoundly complex, highly lossy algorithms. This bodes well if you wish to achieve sufficiently powerful computers; we probably already have them. The unknown, obviously, is how much time is necessary to "figure out" the methods of the mind.

Re:Now we just need to ask it tough questions!

[mrselfdestrukt]

So, your computer should be more powerful than your brain by 2030.

Dude,
By 2030 my potplant will be smarter than my brain.
Oh, the tragedy of alcohol abuse and growing old...

Re:Now we just need to ask it tough questions!

[Eivind]

The mind is not only a turing-machine, it can't be, a turing-machine has no inputs and no outputs.

A turing-machine is a useful mathemathical idea for classifying problems into those that are computable and those that aren't. It's also usefulf inthat it prooves that any machine (or language) that is turing-complete (that is, capable of simulating a turing-machine) can solve the same set of problems.

Re:Now we just need to ask it tough questions!

[sentientbeing]

My brain only thinks about 2 things. Either food or sex.

It must run a Reduced Instruction Set.

Re:Now we just need to ask it tough questions!

[kyojin the clown]

mod parent up

Re:Now we just need to ask it tough questions!

[xAXISx]

Geez, my PDA only gets 10 Flops, but I can write hello upsidedown on it by writing ".1134"

Re:Now we just need to ask it tough questions!

[vertinox]

Maybe we'll get better answers than "42".

Yeah! You'd think we would since we keep plugging in "WHAT DO YOU GET IF YOU MULTIPLY SIX BY NINE?"

Re:Now we just need to ask it tough questions!

[Viol8]

"Not only would the computer be aware like we are"

Not necessarily , since we don't know how aware ness arises in the first place. Conciousness isn't necessarily an emergent property of complexity itself, more likely it emerges on when a certain type of complexity starts "processing" in a certain way. After all, the most complex supercomputers already have the complexity of invertabrates , but have all the awareness of a rock.

Re:Now we just need to ask it tough questions!

[Anonymous Coward]

The mind is not only a turing-machine, it can't be, a turing-machine has no inputs and no outputs.

Hey buddy, your ignorance is showing.

Of course Turing machines have inputs and outputs. The tape state before the TM begins is called an input. If the machine halts, the resulting tape state is the output.

Re:Now we just need to ask it tough questions!

[Skjellifetti]

After all, the most complex supercomputers already have the complexity of invertabrates , but have all the awareness of a rock.

Most invertabrates probably also have all the awareness of a rock.

Re:Now we just need to ask it tough questions!

[Anonymous Coward]

Want irony? The Cell prototype, as described on IBM's research page, doesn't even implement IEEE compliant single precision floating point operations.

Re:Now we just need to ask it tough questions!

[Viol8]

That'll be why Bees can communicate to other bees the location of certain flowers I suppose.

Re:Now we just need to ask it tough questions!

[Skjellifetti]

OK, awareness should be amended to mean self-awareness. It is nearly possible now to build a little robot that emulates a bee's behavior w/out requiring a supercomputer to drive it. So a supercomputer is therefore no less aware than a bee.

Re:Now we just need to ask it tough questions!

[fyngyrz]

There are other issues as well. Intelligence is almost certainly not memory. Yet a very large amount of our "grey matter" is consumed with issues directly related to memory. 2d memory, 3d memory, smell and other non-visual memory... an interesting question is, how much of the remaining brain matter is involved in "processing"?

We know how to store memories; we've got multiple paradigms for just about anything you can think of. Some of them probably have no analog in the human brain. We know how to sense things -- we've got everything from cameras to sonar to magnetic field sensors -- so finding input is not a problem. We're capable of building huge storage systems, though no one has really sat down and given that a shot (lots of expense, here, though less all the time.) It's certainly in no way outside our technological capability.

Now, consider... for a while, it was deemed a "very, very difficult problem" to get a robot chassis to walk up stairs. All manner of research teams were working on it. It was eventually solved by a one guy with a z80, which isn't exactly the peak of modern computing ability. Turns out that the key (as usual) was an adequate methodology, not huge amounts of computing power trying to solve calculus on the fly.

This last bit offers quite a bit of hope for AI as an emergent quality of a better approach. There is no validity at all to any line of thought that says every possible approach to emergent AI has been tried. In reality, only a few methods have been given a serious try.

I think there is a huge flaw in the idea that "X teraflops" will be required before we can have AI; that's a complete and utter mis-statement of the problem, and I think I can show you why to just about anyone's satisfaction:

Intelligence isn't about speed; it is about results. Given enough memory to store and retrieve information, and enough sensors to obtain that information, and a software methodology to process that information, if it takes said software a week, or a month, or a year, to formulate and present an answer that we might come up with for any general question -- even though we humans might arrive at the answer it in seconds -- then I think what you have is in every way is an intelligence that is working on a different time scale. It's not less intelligent, it's just not as fast. That will change its utility to us, certainly, but it won't change the fact of its intelligence. And the day someone comes up with the software methodology is the day prior to the day that engineers begin to create hardware optimized for that methodology, and software engineers begin to optimize the methodology itself. The fact that a slower intelligence could outlast us in terms of years available to think is interesting as well; an interstellar space probe might make very good use of "slow" intelligence, for instance, given that as far as we can tell at this point, interstellar means tens of years just to get started.

My instincts and my money are on the position above; I don't think AI will be nearly as complex or resource-intensive as academics and pundits are making it out to be, and furthermore, once we figure it out, we can expect to see it explode in a way that makes the current technological revolution look like molasses in January. Making a new AI will be doable in the time it takes to copy bits from here to there. Seconds, or a fraction of a second. If the most advanced researcher in the world comes up with one, there is no particular reason that the lamest hacker can't have one five minutes later.

It's always good to keep in mind that folks like Minsky at MIT have been completely, utterly wrong about the simplest, most peripheral AI issues, for instance neural nets, wrong to a degree that has actually retarded progress, and have been found out later (though you don't hear much about it.) The conventional wisdom, even amongst the "AI community", is only as valid as it is, not as valid as the level of respect or reputat

Re:Now we just need to ask it tough questions!

[inode_buddha]

Yeah, but can it get horny?

Re:Now we just need to ask it tough questions!

[TopSpin]

...for a while, it was deemed a "very, very difficult problem" to get a robot chassis to walk up stairs. All manner of research teams were working on it. It was eventually solved by a one guy with a z80...

There are many examples of this. Once the secret behind one of our abilities is revealed, we find that it's very resource efficient. How could it be otherwise? There is no magic; the brain is slow, low power system.

OCR and, more specifically, handwriting recognition were once "computationally infeasible." Today, Post Office machines routinely route messages based on handwriting and any given PDA can handle this input.

Piloting all sorts of vehicles through novel terrain is in practical use in many places and quickly becoming feasible for others. I am certain I will one day own a car that can drive itself on ordinary streets and highways using a machine tucked under the dashboard just behind the firewall.

I've already mentioned speech recognition; high quality implementations are now commodities. I'll be retired 20 years before this ceases to astonish me.

Someone clued into the fact that faces are easily distinguished through a handful of biometrics, combined this with machine vision and walla, highly efficient, accurate facial recognition.

Naive Bayesian algorithms are hounding spammers. This is not trivial; research has shown these methods to be better than human recipients at analyzing messages.

Why are we all not astonished by this? This was the stuff of science fiction 30 years ago. Yet, we manage to solve these discrete problems one after the other and, after a headline or two, it's considered trivial. What does that tell us about ourselves?

My instincts and my money are on the position above; I don't think AI will be nearly as complex or resource-intensive as academics and pundits are making it out to be, and furthermore, once we figure it out, we can expect to see it explode in a way that makes the current technological revolution look like molasses in January.

I agree. Eventually, we'll understand that we are not profoundly intelligent beings endowed with a mysterious cognitive magic. We will have discovered that awareness is at once both subtle and efficient. The "programming" done today be on par with a knotted rope relative to what nature will teach us of systems. We'll have to face the consequences of this. Imagine meeting a self-aware machine that, for whatever reason, doesn't like you...

It's always good to keep in mind that folks like Minsky at MIT have been completely, utterly wrong about the simplest, most peripheral AI issues, for instance neural nets, wrong to a degree that has actually retarded progress

I have a large reservoir of forgiveness on the matter. It wasn't until the 18th century that science began to accept that humans were not distinct from animals. Couple this inflated view with the astonishing degree that simple mathematics does correctly describe much of physical universe, and I can see how academia misled itself. Trying to shoehorn the slush that is cognition into predicate logic and simple neural networks takes a lot of math.

We still have a long way to go. Nature, at the scale of humans, has had nothing as precise at a straight line to work with. The unambiguous nature of a bit is unknown to evolution. Yet, this is how we think. We're going to have to figure out new ways of thinking before we'll understand thought. We're in over our proverbial heads using our existing tools to understand the product of billions of years of refinement.

When someone says "the brain is X teraflop equivalent if you look at it this way, so you'll need X teraflops in a computer to make a brain" you should definitely snicker a little and go on with your day.

I do. In part because, at some point, I lost any profound respect I once had for my own mind in terms of capacity. Most of what is going on in th

Re:Now we just need to ask it tough questions!

[Anonymous Coward]

... the same school that promised that true AI is "just around the corner" for 30 years.

Reminds me of a funny comment from one of my computer science profs in college (while taking an AI class with him). He said he had a friend who told him "AI is the dream of the future. Always has been, always will be."

Re:Now we just need to ask it tough questions!

[Viol8]

"It is nearly possible now to build a little robot that emulates a bee's behavior "

BS. We're not even close.

IBM should seed Cell Processors to Linux clubs...

[Nova Express]

...all across the country so they could all develop software autonomously.

They could call the program "Cellular Automata."

I don't get it

[Anonymous Coward]

The Cell is just a PPC with 8 little miniprocessors tacked on. The miniprocessors have explicit control over and direct access to the contents of their own cache, but can only access data in awkward ways; and are super-optimized for vector/SIMD instructions and floating point operations, but are not so good at algorithmic or complex flow operations.

The Cell's bonus processors are absolutely great for DSP and multimedia apps, such as that we see in the Cell.

But, they are going to be at a strict disadvantage in data retrieval and pushing operations-- which is, incidentally, exactly what most servers, such as a file, web or database server, need to be best at!

What kind of servers *ARE* these??

Re:I don't get it

[child_of_mercy]

Achitecturally they're meant to be the business at shifting jobs to other processors, not just those on the same die.

makes you wonder why sony wanted these for standalone consumer devices but the on-the-fly clustering opportunities should be very attractvie to a lot of IT shops.

Re:I don't get it

[chill]

Unless you buy into the whole "utility computing" paradigm, like IBM does. In that case, servers are going to be doing more than just handing up files and indexing databases.

Using a two-tier or three-tier approach to client/server architecture, with something like a full-duplex GigE connection to fat, diskless clients and you have some real potential.

A fat client (512+ Mb RAM, 1 CELL processor) that can use the backend for the more heavy-lifting tasks would be a fantastic setup for a lot of businesses.

-Charles

Re:I don't get it

[ArbitraryConstant]

"What kind of servers *ARE* these??"

I think Pixar may want one or two.

Re:I don't get it

[YU Nicks NE Way]

Sorry, but it really wouldn't. Think about it.

If you're running a rendering farm, then the Cell is a great tool. (And, if you think about it, a game console is essentially a low end rendering farm.) If you're running a word processor, however, SIMD instructions are useless. If you're performing a standard query against a database, SIMD instructions are useless. If you're sending an electronic mail, hey, guess what? SIMD instructions are useless.

I think that IBM Microelectronics is trying to Cell their new processor in the hopes of Celling their bosses on the (dubious) proposition that they can recoup the losses they've seen on their contract with Sony. They've packaged up the right buzzwords, and they're creating a lot of fog. I sort of doubt that it will work.

Re:I don't get it

[stevesliva]

You're probably right-- Dell's not worried, but Cray, NEC, and Sun might be.

Re:I don't get it

[TheKidWho]

If you're running a word processor, a 3.2ghz processor is a bit useless too don't you think?

Re:I don't get it

[chill]

Yeah, but most "normal" apps like e-mail, word processing, web browsing, etc. are more than handled by the base processor. No, you don't get any help by the SIMD but you don't need it.

When you move to things like editing audio & video, print rasterizing, hi-res photo deforms and filters, then it'll kick in.

Think of what this sort of processing will do to GIMP/Photoshop filter speed... Or DVD/music ripping/encoding/decoding... Or audio mixing...

Re:I don't get it

[NitsujTPU]

What are any of those companies worried about? The best parallel platform at the moment is owned by IBM as well. It's called Blue Gene.

If I were any of those companies, I'd be a lot more worried about that.

Re:I don't get it

[chill]

Cost. Blue Gene costs a fortune.

The PS3 using the Cell is going to drive the price of the base components down. While IBM blades may stay a bit pricey, you can bet there will be others at a fraction of the cost of big iron.

Re:I don't get it

[NitsujTPU]

Ok. I know that Sun makes entry level machines that do this sort of business, but Cray? I thought that supercomputing was sort of their only business.

Re:I don't get it

[chill]

True. However, last I heard they were almost out of business. A last minute gov't contract saved them from getting dissolved; class action lawsuit about the stock; share price at a 52-week low; etc.

I also can't remember who owns them now... ...

Ah, bought by SGI. Spun to separate business unit. Sold to Tera. Renamed back to Cray.

Rumors were they got the gov't contract because too much "supercomputing" was turning towards clusters and the gov't wanted to make sure at least one HPC company survived in the U.S.

-Charles

Re:I don't get it

[Anonymous Coward]

Well, unless the said word processor has something like Clippy.

Re:I don't get it

[wik]

Be careful. Do you know the difference between the words "cost" and "price"?

Re:I don't get it

[SnowZero]

Hey, I frequently run a word processor on my rack of blade servers you insensitive clod! I'm hoping clippy will be self aware soon...

Re:I don't get it

[YOU LIKEWISE FAIL IT]

I seem to recall Sony suggesting that in the future, all of your "standalone consumer devices" will be networked via ip over powerloop or something. The idea being that the Cell in your kettle and fridge can kick processing power to the PS3 in the living room.

Re:I don't get it

[Anonymous Coward]

Renderfarm under your desk

Re:I don't get it

[SnowZero]

TCP offload, stream encryption, and compression are all things you could run on a cell SPE. Someone else mentioned building a database kernel to run on SPEs for simple queries. Sure, it's still more suited for numerical computation, but a lot of servers do some crunching on the side.

Re:I don't get it

[Maserati]

That reminds me, I need to order a fire axe with an insulated handle for the server room.

Re:I don't get it

[adam31]

but are not so good at algorithmic or complex flow operations .... they are going to be at a strict disadvantage in data retrieval and pushing operations

Data retrieval? You have to remember that there's a 25.6 GByte/s bandwidth to main mem! And they each have 256k of single-cycle access memory (i.e. take your 32k L1 and shove it).

It's not like the SPEs are only capable of SIMD vector operations... they are real CPUs with the memory limitation and single-thread limitation. It's just that they do have SIMD capability, and they'll tend to excel most when there is a high computation/(data consumption + branching) ratio.

So what kind of servers are these? They even have a 76.8 GByte/sec bus between SPEs... sounds like they're really freaking high memory bandwidth servers!

Re:I don't get it

[Anonymous Coward]

Usually, the extensions for the processor are used to improve performance in the kernel libraries where ever possible - thus speeding up absolutely everything. Take a look at memcpy in the Linux kernel sometime - it uses MMX.

Re:I don't get it

[jobsagoodun]

And how about offloading SSL computation to SPEs too. For a web server these could actually be pretty neat.

Re:I don't get it

[AuMatar]

Utter bs. A game machine needs to do all its calculations in real time, or the game lags. The speed it would take to send it off to another processor over the net and get the return would be waaaay to slow. Its interesting for clustering, but not gaming or consumer electronics.

Re:I don't get it

[drmerope]

Nono. They are just fine for algorithmic or complex flow--as good as a powerpc is, but this is a poor utilization of the hardware.

The real limitation of the Cell processor would be context switches. There is a hugh register file to take-care of in that case.

Sadly, this processor will only be useful in graphics applications. It's double-precision floating-point performance isn't very good (about 1/10 of the speed)--all us science types really value double precision performance...

as for accessing data in awkward ways... it isn't so bad... it's more like you have to manually drive the caching strategy. Some might see this as a good idea when the normal sequential heuristic used in caches isn't good enough.

Anyways, I was at this demo. The graphics themselves weren't too impressive but the frame-rate on the ray-tracing application was very good--about 10x faster than their version for the normal powerpc.

Re:I don't get it

[bani]

Cells would be perfect for accelerating eg SSH and SSL -- it should be great for crypto in general. Also things like compression - gzip/bzip2/etc. would benefit.

For desktops Cells should be awesome. But servers should still find some good use for them.

Programmers are creative. I'm sure many great uses will be found to utilize Cells in purely server environments.

Re:I don't get it

[Andy_R]

"What kind of servers *ARE* these??"

Cheap ones.... at least as far as IBM is concerned. A large chunk of the money customers are paying is no longer going to be poured straight into the bank account of their competitor, Intel. They can either make huge mark-ups, or more likely bump up the specs to amazing levels to add to the buzz around Cell.

IBM's long-term strategy always has been to position the Power architecture as the successor to x86, so this is a logical move, after their success in ensuring that every console game programmer in the world will be writing for their chips. IBM wants your next computer to be Power or Cell based, running either OSX or Linux. Just like PC makers always do, they are putting their top-of-the-line chip into servers first, then when production ramps up, it will trickle down the range.

What I'm wondering is if they would have better throughput with the triple-core chip that's going into the Xbox360, which seems at first glance to be a better general-purpose device.

Re:I don't get it

[aliquis]

On the other hand i think they talked about "Cell storage" in a recent story, a place where you store your music, movies and whatever content, and the cell "improves the quality" of it. I can understand that for a rendered game, however not for a movie. Sure you can interpolate the whole thing, but I don't know if I would call that an improvement.

Re:I don't get it

[tempmpi]

Fine for algorithmic or complex flow without branch prediction?

And about managing the cache by yourself: that sounds like almost the same stupid mistake Intel made when they designed the Itanium, wrongly assumed hardware stuff could be done just as well by the compiler or the programmer. What makes it even worse is that programmers nearly always believe the hype of strategies like that. After all they are good programmers, so they should be able to do that little cache management by themself, can't be that hard right?

Re:I don't get it

[drmerope]

Actually IBM has a C and Fortran compiler they made just for that reason--whether it really works well, that's hard to tell at this point.

And I think you grossly misunderstand what's wrong with the Itanium. It had *nothing* to do per se with permitting the compiler to organize the fine-grained parallelism. It had to do with throwing in the kitchen sink so to speak of every clever idea people have had--big surprise they don't all work out well together.

But the major problem was placing so much additional load on the memory interface by using such a large register set and using such large instructions.

Re:I don't get it

[seweso]

IANAP but couldn't these processors be used for genetic fuzzy algoritm's?

Re:I don't get it

[slashflood]

But, they are going to be at a strict disadvantage in data retrieval and pushing operations-- which is, incidentally, exactly what most servers, such as a file, web or database server, need to be best at! What kind of servers *ARE* these??

Servers you can find in the render farms of ILM. One of the demonstrations was a realtime ray tracing of a landscape. The resulting jpegs were streamed to an Apple G5, because the Cell-based blade server had no high end graphic board.

There are thousands of other applications for such a kind of server. The Earth Simulator is also not a file, web or database server.

On the other hand, even a web server can profit from a Cell server. Look at all the computations a PHP server is doing nowadays. A content management system like relies heavily on ImageMagick to generate the images on-the-fly. Look at all the content servers, like video and audio (mp3) download sites. Some of them are rendering thumbnails or converting uploaded content on the webserver itself.

Database servers are not only looking up entries in an index, they are also doing heavy calculations.

SIMD (like SSE) helps a lot in different areas. A file server could do real-time encryption...

Re:I don't get it

[Lemming Mark]

How about using the extra power to optimise crypto, checksumming for software raid, etc? These should be useful on a large subset of servers (and existing vector extensions are already used for these tasks).

Re:I don't get it

[milimetric]

Dude, you haven't even written one line of assembly for these things and you're saying that there's no way this could help data retrieval and pushing operations?

Why not? If the damn things are so much better at vector instructions, why is it so hard to immagine a server's software being re-written to use vectors. For example Matlab. It is obviously based on linear algebra principles and it is better at vector and matrix computations than a program written in C to do the same thing without fancy algorithms. Yes, I realize Matlab is written in C, it's the interesting nature of diagonalized and sparse matrices that makes Matlab faster. Matlab works as a web server and a database client. Just imagine if we pump Matlab with these Cell beauties what amazing performance we would see... from Servers!

Don't hate on progress just because you have no imagination.

And you Microsoft haters... there's also no reason to believe that Microsoft won't take advantage of this... they're not idiots you know.

Re:I don't get it

[cybergrue]

Anyways, I was at this demo
You were!, any chance that you could write up a review of what was said and presented? Inquiring minds want to know.

The real limitation of the Cell processor would be context switches
Why would this matter? The PPC core would handle the interrupts, and as I understand it, the SPUs would be treated like vector processors, which would be assigned a work unit and would process that till completion. Was something else said durring the presentation?

It's double-precision floating-point performance isn't very good (about 1/10 of the speed)--all us science types really value double precision performance.
Yes, but even at that slower speed, the cell is still among the fastest processors for doing double precision calculations, and if Sony et all actually make as many cell processors as they claim, it will probably be among the cheapest as well ($ per GFlop).

Re:I don't get it

[Anne Thwacks]

This is exactly the architecture I specified for a hardware database engine I designed 20 years ago.

Naturally the implementation details are considerably different.

You have to think about the problems from a totally different perspective to get the best from this architecture, and re-compiling existing C programs is not the way to do it. You need languages like Algol68 or Occam to get benefits. Ie you need a language that allows you to specify what cannot be done in parallel, and assumes that the rest can - but SQL is also good! There is any amount of existing knowledge on how to use parallelism for database work, it is inherent parallel anyway - you perform database operations on sets!

Re:I don't get it

[glitchvern]

The Cell is just a PPC with 8 little miniprocessors tacked on. The miniprocessors have explicit control over and direct access to the contents of their own cache, but can only access data in awkward ways; and are super-optimized for vector/SIMD instructions and floating point operations, but are not so good at algorithmic or complex flow operations.

And the floating point operations are only single not double precision. While that may be fine for a game console, I imagine it will limit the Cell's usefulness in certain applications in which having vector/SIMD instructions would be extremely useful.

Re:I don't get it

[timeOday]

Sadly, this processor will only be useful in graphics applications. It's double-precision floating-point performance isn't very good (about 1/10 of the speed)--all us science types really value double precision performance...

I disagree that double precision should be assumed a requirement for all scientific apps. Most of scientific computing is simulation, and I would argue unrealistic models are almost always a bigger problem than numerical precision!

Re:I don't get it

[LWATCDR]

Rendering farms and Simulation are to uses. Think Cray not Netware.
How about running SETI at home on one of these bad boys :)

Re:I don't get it

[stevesliva]

Cray indeed. Their bread an butter was vector supercomputers, if there's one thing a Cell will emulate better than any other consumer architecture, it's a vector processor.

Re:I don't get it

[Anonymous Coward]

How about using it for running grammar checks on slashdot posts...TWO, not to.

Also, given the steep sticker prices most servers carry, why one earth would someone buy one(or any server that wasn't just collecting dust otherwise) to let it chew away at completely unremarkable data sets using questionable analysation methds???? But YOUR server is gonna be the one to find the aliens. Right...
I'm all for supporting the cause, but I'd think you'd have to be foolish beyond comprehension with your money and priorities if you'd honestly spend a few grand to improve your SETI@HOME scores.
Or maybe you've just got so much money it's falling out of your pockets when you walk down the street. Either way, the money would be better spent giving it to the SETI folks directly.

Re:I don't get it

[Troy Baer]

I disagree that double precision should be assumed a requirement for all scientific apps. Most of scientific computing is simulation, and I would argue unrealistic models are almost always a bigger problem than numerical precision!

The scientists developing the simulations typically understand the limitations of their models, though. If you ask a physicist or chemist their opinion, they will not be satisfied with the 6-8 digits of precision available from 32-bit floating point. Also, when validating an old code on a new platform, you're generally comparing results with something that did exclusively 64-bit floating point (like a DEC Alpha or an older Cray vector system).

The fact of the matter is that folks who do scientific computing have had machines that do 64-bit floating point well for so long that a processor that doesn't do 64-bit FP well isn't going to be taken seriously. (This is why most scientific computing types didn't take Macs serious for clusters before the G5; the much-hyped AltiVec unit was useless to them, because it didn't do 64-bit FP.)

--Troy

Re:I don't get it

[benjcurry]

I agree with alot of this. It's possible that the Cell Processor won't give much of a performance boost to the Office apps of today.

But...

The current trend (See Longhorn, OS X and various Linux tools) is too create visual environments that are more and more "real" and three dimensional. In three years, I could see the value of the Cell Processor being to help speed up this new generation of GUIs while still keeping the backends running at incredible speeds.

I get the feeling that computer UIs are getting to the point where another quantum leap like the jump from CLI to GUI is set to occur. The processing power is there. There are oodles of projects exploring new ideas about how to incorporate the use of real three dimensional space into the computing experience. I don't pretend to know what direction this will take or whether or not the Cell Processor is going to succeed (although Sony's vote of confidence is somewhat reassuring), but I feel like the ground is shifting and it's going to be fun to be there to see what happens in the next 5 years or so.

Re:I don't get it

[LWATCDR]

It was a joke you poor bastard.

Re:I don't get it

[fitten]

The real limitation of the Cell processor would be context switches
Why would this matter? The PPC core would handle the interrupts, and as I understand it, the SPUs would be treated like vector processors, which would be assigned a work unit and would process that till completion. Was something else said durring the presentation?

Well, on a context switch, the PPC has to save all its registers out for the next context to load. The fact that it has 128 VMX/Altivec registers (all pretty large) is a big load to push every context switch. Similarly, what are you going to do about the SPEs on a context switch? In a console, there are no/few context switches, especially not the general purpose OS kind to load another process. What do you do with the SPEs? Do you let them keep going doing their thing? In that case, one user can lock them all down and no one else can use them. The other side of the coin is that you have to swap them all out as well... so now your context switch has gone from ~64 64-bit registers and 32 VMX registers and a few other things... less than 1 KILOBYTE (and this is considered a lot in the realm of context switches)... to 8x256 KILOBYTES (2 MEGABYTES) just for the SPEs and the additional 64 VMX/Altivec registers are just noise in this. So, you want to context swap out 2 MByte every time the OS wants to run another task (60 times a second?) Can you say "slow"?

Re:I don't get it

[MyHair]

If you're running a word processor, a 3.2ghz processor is a bit useless too don't you think?

Yeah, you'll probably want at least 3.8ghz minimum for the next version of MS Word.

Re:I don't get it

[Dolda2000]

I wouldn't be so fast to deem these out in the web server arena.

It seems that the extra processors could be great for accelerating things like SSL encryption.

Re:I don't get it

[apoc06]

precisely what i was thinking. php consumes alot of cycles on any given webserver. look at the direction that current server side technology is going and you wind up needing that extra power once your server has any substantial load.

Re:I don't get it

[Flaming Death]

Maybe you dont get it, because its a different way to handle data and code. Try reading a bit about the Cell before making all sorts of incorrect assertions. Heres a link to help:
http://www.research.scea.com/research/html/CellGDC 05/index.html

Make certain you read about the _various_ uses of the SPE's and the huge number or programming models you can use to drive the Cell..

To answer you question - these are insanely fast data crunching servers.. geee.. thats almost any server.. nugget..

Re:I don't get it

[the_lesser_gatsby]

Once somebody writes the software (or compilers really)...

Media encoding, cryptography, simulations, and, I guess, anything for which you could use a vector coding paradigm.

Re:I don't get it

[Flashpot]

They need someplace to run all that middleware!

Re:I don't get it

[the_lesser_gatsby]

What he's saying is that, in a context switch, you treat the SPES like hardware io and wait for them to finish the assigned task. The thread can simply block on the SPE and give up control to another thread.

Re:I don't get it

[Wdomburg]

As per IBM: "Single precision floating point computation is geared for throughput of media and 3D graphics objects."

The problem isn't that single precision isn't adequate for a lot of tasks. The problem is that single percision mode sacrifices IEEE compliance in order to maximize throughput for the targeted application.

I wouldn't be surprised if they go on to release a different impelementation with some or all SPUs geared toward more general purpose tasks, but it won't have the raw throughput of the current chip.

Re:I don't get it

[Anonymous Coward]

one or two? Probably at least one or two racks.

Now, if we can only get hold of the distro...

[Sensible Clod]

PS3Linux awaits!

Re:Now, if we can only get hold of the distro...

[mrchaotica]

I'm more looking forward to Xbox360 Linux or Nintendo Revolution Linux, with Mac-on-Linux running on top...

Re:Now, if we can only get hold of the distro...

[CableModemSniper]

touche good sir, touche.

Re:Now, if we can only get hold of the distro...

[AuMatar]

All true geeks will go for 360 linux- remember that MS takes a loss on every sale!

Re:Now, if we can only get hold of the distro...

[Anonymous Coward]

Christ you're thick.

The more Microsoft sell, the quicker they break even, the faster they establish a potential customer base, the more attractive the XBox looks to developers.

More units == more games. Go figure. Don't like Microsoft, don't buy an XBox.

Re:Now, if we can only get hold of the distro...

[Anonymous Coward]

No.

If Microsoft makes a loss on XBoxes such that each XBox costs more to build, that is, the component and manual labour associated with each XBox is more than the share of retail revenue it receives on each sale, then Microsoft will always make a loss on the machines themselves, regardless of how many they sell.

If what you were saying was true, Microsoft wouldn't have a problem with the idea of people porting GNU/Linux or NetBSD or whatever to the XBox. More sales = closer to breaking even and convincing game developers, right?

As for the last comment, there are only a handful of game publishers in the world right now. It's hard to actually break into the business. So I can assure you that if the XBox had more owners but Revolution had more people buying games, the developers would stick with the Revolution. The publishers would go for the larger market, not the larger installed base, and the developers would be forced to tag along with them.

The Cell concept is really cool

[Dancin_Santa]

These things work fine alone, but when connected together they really shine. Built-in clustering hardware interfaces makes this a nerd's wet dream.

Putting them together into a rackable case looks to be very cool and finally putting a nail in the Windows coffin will be a delicious treat for IBM (the Cell ain't x86).

I can't wait to get my hands on my PS3 and see what I can do.

In the meantime, I just wish IBM had Cell samples available for a reasonable price. I just can't afford one for hacking yet!

Re:The Cell concept is really cool

[child_of_mercy]

that'll be your ps3 + sony linux kit...

Re:The Cell concept is really cool

[V 0 ! D]

Yeah. . . I have wash my pants.

Re:The Cell concept is really cool

[kilraid]

Finally putting a nail in the Windows coffin will be a delicious treat for IBM (the Cell ain't x86).

What coffin...?

Re:The Cell concept is really cool

[Gabrill]

Don't worry. Anything as big as Windows has plenty of competitors preparing coffins and waiting for their eventual need.

Re:The Cell concept is really cool

[mrselfdestrukt]

Bloody hell!
That's a very cool idea!
Just imagine if Sony made the PS3 (or PS4 after they've ironed out all the bugs and did some further development) so versatile that it evolved from being only a games machine to something that companies can buy and run linux clusters on.
Imagine it not only having enough *oomph* to draw games graphics really well, but in cluster formation it can outshine other hardware configurations for number crunching or data pushing. That would be really sweet, but I can't figure out if it will push the price up or drive it down.And IT staff might sneak 1 or 2 out of the server room to play games on at home.

Re:The Cell concept is really cool

[gl4ss]

*I can't wait to get my hands on my PS3 and see what I can do.*

play some games? that's what you'll be able to do.

Re:The Cell concept is really cool

[Anonymous Coward]

Sadly, that's true.

I'm just a slashposer. I even run Windows XP.

Re:The Cell concept is really cool

[Anonymous Coward]

I really want DRM on my processors too. Blah

Re:The Cell concept is really cool

[zoney_ie]

Longhorn looks like being far from a definitive release, *as well* as being delayed. Just about every milestone technology supposed to be in it won't be fully there from the start, and it will end up being as patched up as an old bicycle inner tube.

It's not unlikely that many will realise this, and stick with XP until the next OS release *after* longhorn (and that's some time away). Don't forget, if people want the new tech patched in, why buy Longhorn? WinFS for a start will be available for XP.

MS may be fine - but they've left an awfully long period open to attack from others.

BTW - it may have been very calculated that IBM sold off their PC division. They may not end up capitalising fully on the Cell architecture, but potentially they could start afresh in the desktop arena (they'd have to go for an OS they control to some degree, rather than allow an MS-like co. to dictate terms). Ideally they should ensure a PC replacement where they sell the processors and software - and only leave the actual machine assembly/sale (which becomes generic again) to other Dell-like companies.

Re:The Cell concept is really cool

[Anonymous Coward]

Hell yeah!

Imagine a cluster of these being used to model WMD for totalitarian regimes. And still have enough power left over for some GT5 goodness.

Re:The Cell concept is really cool

[LurkerXXX]

Nail in Window's coffin? I don't think they are anywhere near ready for a coffin yet.

Besides which, you do realize Windows used to run on the PPC platform right? Windows NT4 ran on that MIPS and Alpha machines. And guess what? The new XBox, which is a cut down-modified version of windows 2000, is going to run on PPC chips. MS could quite easily move Windows the the Cell chips of they chose to. This is competition for AMD and Intel, not MS.

That's odd...

[Anonymous Coward]

"We demonstrated the prototype to show that Cell continues to mature..."

...I thought Gohan killed Cell?

Re:That's odd...

[stratjakt]

No he decided to blow himself up and take earth with it, luckily Goku used his instant transmission to transport him to King Kai's world and blew it up instead. And himself. What a selfless act. *sniff* That Goku, what a guy.

Re:That's odd...

[binkzz]

That didn't kill Cell though; Gohan had to finish him off.

Re:That's odd...

[fyrewulff]

What you didn't see was the "IBM saga" where a new IBM engineer finds the remains of Cell, and brings him back to the design room.. Next time, on IBM Z!

Very promising technology= investment opportunity?

[guyfromindia]

Guess it is time to invest in Sony and IBM! This technology really looks promising, especially when you read this article --> http://www.blachford.info/computer/Cells/Cell5.htm l
The first Cell based desktop computer will be the fastest desktop computer in the industry by a very large margin. Even high end multi-core x86s will not get close. Companies who produce microprocessors or DSPs are going to have a very hard time fighting the power a Cell will deliver. We have never seen a leap in performance like this before and I don't expect we'll ever see one again, It'll send shock-waves through the entire industry and we'll see big changes as a result.

correct me if i'm wrong....

[Lehk228]

Wast the benefit with Cell supposed to be that the programmable DSP's worked somewhat like pixel shaders except useful for all kinds of complex serial data so that operations on serial data could be massively improved, which does not seem to me like it would be a major help in a server, unless it is running a specialized app that just happens to be on a server for data access rather than using the Cell to speed up web servers etc.

Re:correct me if i'm wrong....

[NovaX]

I think its because web servers have thread pools, so a cell processor could handle many of these light-weight threads simultaniously. This makes it perfect for a blade server.

Sun's Niagara is aimed at this market, where the work is of great quanitity, not huge number crunching. This could mean searching, web page serving, and streaming media. So if you need to handle thousands of requests, this type of processor is ideal. Of course we won't truly know until one of these massively multicore beasts is out in the wild and can be tested in a realistic scenario.

Re:correct me if i'm wrong....

[simcop2387]

a straight webserver i doubt would be much help, however i bet a database or something similar doing lots of sorting/searching could probably be greatly improved by the design of the cell architecture. they tend to deal alot with organized and comparing serialized data.

Re:correct me if i'm wrong....

[MikeFM]

Wouldn't this increase the speed of things such as encryption?

Also I'd be curious as how well this could handle lots of small simple problems. A good bit of program logic is working through the same little loops doing simple mathematical operations. Is there any reason that these kinds of processes couldn't be farmed out to the SPE's?

The truth though is that most servers aren't lacking in power. My experience is that a single server can handle the web serving, dns serving, and email for dozens or hundreds of small to midsize companies. For tasks such as fileserving I think fast, hardware-based, network cards and responsive disks make more of a difference than the CPU speed. Just serving files doesn't take much in the way of CPU resources.

Of course I have other apps that are so resource hungry that the single app spreads across many powerful servers. I'm interested to see what the Cell can do for those problems. If the cell has built-in clustering does that mean that we can upgrade our computers just by tossing more CPUs into the box or even by chaining multiple boxes together?

If so I foresee some really cool PC/servers. They could embed a Cell into a lego-like interface that lets them literally just be snapped together. Whenever you need more power just plug new bricks on. The design could literally be as easy as playing with a childs toy if it was designed right. That'd just be staggering if it could be made to work.

Interesting quote from the article

[wyldeone]

"If operated at 3 GHz, Cell's theoretical performance reaches about 200 GFLOPS, which works out to about 400 GFLOPS per board"

From TFA. Interesting, considering that they're claiming that the PS3 will run 5-10 faster than this.

Re:Interesting quote from the article

[Rallion]

Oh, well Sony has the special cell. It has bigger numbers on it, because it's Sony.

Re:Interesting quote from the article

[theid0]

considering that they're claiming that the PS3 will run 5-10 faster than this

The numbers include the operations from all chips on the motherboard, which includes the massive video chips. Besides, these are all preliminary numbers.

Re:Interesting quote from the article

[damiam]

Sony's 2TFlops number for the PS3 includes the NVidia graphics chip, which has an insanely high FLOPS count but isn't really useful for general-purpose computation.

Re:Interesting quote from the article

[Anonymous Coward]

I thought the NVidia graphics chip didn't have finalized specs yet?

Re:Interesting quote from the article

[Anonymous Coward]

The Broadband Engine is around 220Gflops.

The 360 chips is way down around 80Gflops.

It's too bad MS's marketing department had to try to confuse consumers by including throwing the GPU into the system floating point rating. Now we are stuck with it.

Thanks for the hype MS!

Re:Interesting quote from the article

[Namarrgon]

"Broadband Engine"? Which marketing dept came up with that one?

Re:Interesting quote from the article

[Anonymous Coward]

"isn't really useful for general-purpose computation"

I thought modern pixel-shader and vertex-shader languages were turing complete? Thus leading one to suppose that they are capable of all the general-purpose calculations one could ever need (of course with some wrangling since it's not _built_ for computing pi or primes or what have you)

Re:Interesting quote from the article

[gl4ss]

you know, they just count the gflops from the chips together and smack that as the number. kind of like earlier you would call a console 32 bit if you had somewhere in there two 16bit cpu's, and all other shady pr like that.

hell, they don't even have the gfx chip fabbed yet apparently(they used some other nvidia chip to demo it).

however-- if they showed the thing.. wtf do we need theory for, couldn't we have some real benches and usage? or did they just show a plastic shell and not have fabbed the chip yet at all and/or at those speeds it's supposed to run at?(most likely).

Re:Interesting quote from the article

[cuddles]

Wow, don't tell the good folks at GPGPU that...they'd have to find new stuff to research.

Re:Interesting quote from the article

[PhoenxHwk]

Two slight clarifications:

1) The 2 TFLOPS figure is commonly thought to be a big lie (or mistake), as a GeForce 6800 Ultra is capable of somewhere around 50-75 GFLOPS. This would mean a 50x performance increase between two generations, which seems very unlikely.

2) See www.gpgpu.org and you'll find that graphics cards can do some serious general-purpose number crunching.

I'm just curious

[SixDimensionalArray]

Ok, so the way I see it, we have invented a lot of ways to increase our MIPS and our processing power.. something along the lines of this->

1) Single CPU
2) Multiple CPU
3) Multiple Machines in a grid with single CPUs
4) Multiple Machines in a grid with multiple CPUs
5) Multiple grids with many machines
6) Multiple cores in a single CPU
7) Multiple cores in multiple CPUs
7) Multiple cores in multiple CPUs in a grid
8) ..what next?

We also went from 8-bit to 16-bit to 32-bit to now 64-bit and beyond. 64-bit words.. nice! Of course, more parallelism means more threads for more simultaneous processes, and 64-bit means twice as much "word" space than 32-bit, but what next?

It's truly mind boggling, and it's a great time to be in IS/IT!

What I want to know is, how much further? How can we increase the multiples more? For example, what happened to quantum processing and multiple states for a bit instead of 0 and 1? When can I count my bits 0, 1 and .5? Any supercomputer geeks care to postulate?

Re:I'm just curious

[BiAthlon]

Ok, I've got five mod points and you make me post something to this story instead of mod.

What would you do with a 'bit' that was "pretty close to 1" or "just a bit over 0"? You no longer have any exact state of data which every language I've ever used has depended on.

I like my 1's and 0's just fine thanks ;)

Re:I'm just curious

[krautcanman]

7) Multiple cores in multiple CPUs
7) Multiple cores in multiple CPUs in a grid
8) ..what next?


Profit!!!!!!!
Actually, you have two sevens, 8) should read: 9) Profit!!!!!!!

Re:I'm just curious

[Anonymous Coward]

I wonder how hard it would be to dual or triple core the PPC part of a cell once IBM gets multiple-core-power stuff down pat. For general purpose computing it'd be a big step up, as it seems like that (as opposed to vector and numerical computing) is the only weak point the cell has vis a vis intel or even vanilla power.

Re:I'm just curious

[birge]

What I want to know is, how much further? How can we increase the multiples more? For example, what happened to quantum processing and multiple states for a bit instead of 0 and 1? When can I count my bits 0, 1 and .5? Any supercomputer geeks care to postulate?

Don't worry about quantum computing. It's only going to help the NSA as there are only a limited number of algorithms which will be worth it, namely factoring prime numbers. The power requirements are going to be huge, and by the time they figure out how to keep more than 128 qubits coherent long enough to do a computation, you'll be long dead.

Quantum computing is just a clever way for physicists to get money out of the government to study the kinds of stuff they really want to study. They just mention that their project could eventually be used to build a quantum computer (which covers about 90% of physics research) and the Feds throw money at them like it was cookies.

Physicists aren't dumb, you know, but the people working for funding agencies are. So just because you hear the latest buzzwords, be it, "quantum computing" or "nano-blah-blah-blah" just remember that it's probably just scientists gaming the funding system. The research changes very little, but the hype is always moving.

Re:I'm just curious

[Michael Hunt]

'factoring prime numbers'?

You mean deriving the factors of products of primes, right?

Re:I'm just curious

[Anonymous Coward]

My guess (as somone who works in the field) is 5-10 years for a working quantum computer capable of solving non-trivial problems. 10-20 years for a quantum computer capable of solving problems out of the realm of their contemporary classical computers.

But, they will do no good for you, at least not for 30-40 years. There are only a few known interesting quantum algorithms. For instance, it is strongly believed (though not proved last I heard) that quantum computers cannot solve NP complete any faster than classial computers. (of course, one explanation for this could be P=NP...). I suspect that for quantum computers to be useful for a random guy that we will need to discover a new problem that everyone needs to solve.

Re:I'm just curious

[calambrac]

Let's say you had a computer that stored decimal digits. The number of digits necessary to hold the value 9 is 1, compared to 4 bits (1001); the number of digits needed to store the value 256 is 3, compared to 9 bits (100000000)... so on and so forth. If a digit (or a trit, or a hexit, or whatever) could be implemented in hardware in the same way as bits, we would drastically improve storage capacity.

There's also absolutely no reason that a digit can't be treated as boolean... if it's zero, it's false, if it's anything else, it's true. This is the way booleans are abstracted with integers in C. Greater-than-binary digits have the added advantage of being able to represent everybody's favorite logical state, "unknown".

AFAIK, the reason contemporary computers are binary is because that's the easiest way to do it from an electronics standpoint, but I'm not into EE, so I could be wrong on that. There have been ternary computers; see the Setun for an old example from Russia...

Re:I'm just curious

[lnjasdpppun]

The number of digits necessary to hold the value 9 is 1

And how (at a transistor level) is the computer going to store and manipulate this digit?

This is like saying a computer can store Hex numbers more efficiently than binary 'cos it takes 16 binary numbers to show the same information as a hex number!

Re:I'm just curious

[SixDimensionalArray]

Here's one possible example (from a patent):

"Trinary signal apparatus and method

Extended trinary signal apparatus includes window comparator logic having first and second inputs for first and second trinary input signals, wherein each the trinary input signal can be a high, low or mid state, and an output for outputting signals dependent on the states of the first and second trinary input signals. A switch, which is connected to one of the first and second inputs, can be selectively activated in one phase to set the one of the first and second inputs to a state other than the mid state and can be inactive in another phase. Control logic is responsive to output signals from the window comparator output during the one and the other phase to provide extended trinary decoding of the trinary input signals. In this manner ninth and tenth input combinations can be identified by detecting whether two inputs which show a mid state are electrically connected to each other or not, this being achieved by selectively pulling one of the inputs to a predetermined state and determining whether the other input follows or not. Trinary encoding can thus be extended to provide ten, rather than the conventional nine states from two inputs."

From http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PT O2&Sect2=HITOFF&p=1&u=/netahtml/search-bool.html&r =1&f=G&l=50&co1=AND&d=ptxt&s1=trinary.TTL.&OS=TTL/ trinary&RS=TTL/trinary

Re:I'm just curious

[ShieldW0lf]

And how (at a transistor level) is the computer going to store and manipulate this digit?

If it was going to use transistors, it could be done the same way as with binary digits, by varying the voltage. Instead of having a range between 0 and 0.8 volts = 0 and a range between 2 and 5 volts = 1, you make the chip recognize (for example) 0-0.8 = 0, 1.5-3 = 1, 3.6-5 = 2. Wow, a trinary computer.

Bringing it from the transistor level to the practical level, a bit harder....

Re:I'm just curious

[grammar fascist]

'factoring prime numbers'?

You mean deriving the factors of products of primes, right?

What in the crap is this, Slashdot? Moderators on...something much worse than crack?

So the grandparent states his honest opinion of quantum computing and gets modded down as flamebait, and the pedantic creep parent poster gets modded up as informative for pointing out a simple syntax error?

It's an easy one to make, too. I've said it, and that was after being immersed in the field for six months!

What the grandparent says is mostly correct. There aren't any good standalone quantum algorithms besides Shor's algorithm. (You cite Grover's O(sqrt(n)) search? I dare you to find a practical application of it. Who knows? You might turn up something good in ten years or so.) It's hard to keep a quantum system coherent long enough to make it useful. Quantum computing does get improperly used to squeeze money from the feds. It's a tough, esoteric subject with few practitioners and a sexy name, and if you can wedge the word "quantum" into your research somehow, you magnify your chances of getting published or getting funding fourfold.

I don't share the grandparent's cynicism, because I do think it has a chance of working. His position isn't entirely unreasonable, though, just looking at the facts.

Re:I'm just curious

[diabolus-ex-machina]

7) Multiple cores in multiple CPUs in a grid
8) ..what next?

A Beowulf cluster of those, of course !

Re:I'm just curious

[Anonymous Coward]

64-bit means twice as much "word" space than 32-bit

eh, no? 33 bits would be twice as much space as 32 bits. we're talking 2^32 times as many possible adresses!

Re:I'm just curious

[calambrac]

it's not like saying that, it is saying that. if there were a transistor that could represent 16 different values, it would store numbers far more efficiently than a binary transistor. as far as the hardware implementation goes, that's not my problem. i just use what they give me. but it isn't some ridiculous idea; it would just be a matter of interpreting more values from more bins in the range of possible voltages. it's not like transistors are little inherent 1s and 0s; all information is just representation and interpretation, including the electronics of a modern binary computer.

Re:I'm just curious

[The Creator]

I'd much prefer Natalie Portman...

Re:I'm just curious

[birge]

Well, actually I meant prime factoring, i.e. factoring any integer into its constituent primes. You are right that what I wrote is technically trivial.

Re:I'm just curious

[birge]

What in the crap is this, Slashdot? Moderators on...something much worse than crack?

Thanks for pointing it out, it is a little frustrating. Some of my best posts have been modded up, and then modded down by people with agendas. I think we need to keep moderation, but quit making it anonymous. You should be able to know your accuser. I don't think my post was flamebait, and it certainly wasn't meant to be.

I don't share the grandparent's cynicism, because I do think it has a chance of working. His position isn't entirely unreasonable, though, just looking at the facts.

For the record, I think it will work, too. I just don't think it will happen in our lifetimes. Looking at the rate of progress to date at getting coherently entangled systems, I think that's warranted. The NMR guys are tapped out, and right now I don't think the atom guys have more than a few bits working. I'm not being critical of those in the field; they are all smarter than me and working on an unbelievably hard problem. I was just trying to make a statement about the challenge, as well as a statement about the way scientists game funding agencies. And we do, all the time. I figured people would appreciate somebody being honest about the fact that public money is often awarded based on who can shift with the hype the fastest. (Look at how many electrical engineers have suddenly become "bioengineers," and how quickly materials scientists doing the same research they've always done are now doing "nanomaterials," as if crystals suddenly changed their dimensions in the mid-nineties.)

Re:I'm just curious

[speedbump]

Innovation is not just happening in throwing more processors at a problem, but in HOW we solve problems.

I've been in the computing business for over twenty years, and I am amazed at the rate of acceleration of technical innovation.

For instance: I work for a client with multiple web projects going, using a variety of operating systems and development environments. A couple weeks ago, one of the other teams gave a presentation to the rest of us about how they have switched from using a relational database for data management to Lucene (http://lucene.apache.org/java/docs/). I was skeptical that a Java-based XML indexer back-end could retrieve information faster than MySQL, but these guys showed that it was much more efficient to do so. Lucene is not a perfect solution to every problem, but for primarily static word searches, it utterly trumps an RDBMS in performance.

The moral of the story is that there are more than one one to skin a cat, and some ways are faster than others!

Re:I'm just curious

[maraist]

Instead of having a range between 0 and 0.8 volts = 0 and a range between 2 and 5 volts = 1, you make the chip recognize (for example) 0-0.8 = 0, 1.5-3 = 1, 3.6-5 = 2. Wow, a trinary computer.

Bringing it from the transistor level to the practical level, a bit harder....

As an EE: Consider this analogy. Fiber-optics verses copper. Which is faster? fiber you say. Ok why? Because the medium allows for higher bandwidth capabilities? Ok, sure.. But here's an interestin factoid, fiber is usually base-band, while copper often is voltage-varied. Consider the phone line which uses frequency to transmit info (more recently it has effectively evolved to voltage sampling as you reference), but fiber is still 0 or 1; presense or absense of light.

Next, consider parallel v.s. serial BUSes. It is intuitive that parallel buses provide greater bandwidth for a comparable medium, so why did we shift to serial? It wasn't just because it takes up less space / clogs less air (though that's a bonus).

While fiber is a common denominator in both examples, you might ask yourself, if it's possible to send full-spectrum signals in fiber and it's possible to put fiber in parallel like a BUS, why don't we do this with with BUSes and phone lines? Well, the obvoius answer is that the jump in speed was enough to satisfy current demands; there simply aren't 1GBps hard drives yet. But there is another, more practical answer.

Because it's cheaper to use base-band, serial technology.. Computer busses use to be serial on low-end computers in the 70's. It was cheaper to write software on both transmitting and recieving end to serialize/decerialize the material. Plus the computers were FASTER than the medium's bandwidth.

Likewise, modems were analog devices; frequencies were measurable by capacitors; you'd get voltage on one of several tuned sensors, representing bit-patterns. This was originally faster than performing a fast-fourier-transform on the signal and extracting signal-noise compensation, yadda-yadda (which modern home-modems do). The sender, obviously can just digitally generate a voltage-pattern representing error-corrective code-patterns, but due to the phone-line standard, they must ultimately represent sine-waves (unlike say ethernet or direct serial buses).

Today fiber is faster than the computational power, so encoding/decoding signals isn't practical. And the issue with parallel bus synchronization is likely as expensive in fiber sysems as in copper. Thus it is only when the bandwith is really hurting are we going to see parallel buses.. Fiber external RAID hard drives have parallel fiber BUT one in each direction (which is the cheapest way to "double" the bandwidth).

The main advantate of digital v.s. analog was error rejection. When you have a signal that if beyond a threshold is definatively 1 and else 0, then you can amplify it with little fear of INTRODUCING error. With analog amplification, the amplifier ALWAYS introduces some non-linear distortion or power-generation flutuations, etc. So even if you had a perfectly clean input signal (which is nearly impossible), you are garunteed to distort the output signal. Having several amplification hops means noise in inherently introduced into the signal. Then take into account external noise (that isn't compensated for by cabling techniques). Trans-continental phone services were a joke. Introduce bi-state logic. Here you NEVER let voltages get near the quasi-zone, you jack them up signficantly higher than the threshold transition voltage. Now so long as noise voltage doesn't exceed 50% of your voltage, you will never have an in-advertant transition from 0 to 1. Cabling noise-cancellation techniques do wonders to minimize the voltage of noise. So now you can repeat a signal an infinite number of times and not introduce new noise into the system. (Though there is always a statistical chance that it will be intoduced somewhere).

Why does amplification matter? Consider your CPU BUS.. your

Re:I'm just curious

[maraist]

eh, no? 33 bits would be twice as much space as 32 bits. we're talking 2^32 times as many possible adresses!

No, he said "word" with quotes. Meaning the physical space it takes up. 64bit computers take up twice as much data cache. It implicit that the concents of the word have greater value, though it is not specified.

Just being picky about you being picky. :)

Re:I'm just curious

[Anonymous Coward]

It's only going to help the NSA as there are only a limited number of algorithms which will be worth it, namely factoring prime numbers.

SHH! The prime factoring algorithm is top-secret!!

Re:I'm just curious

[cintyram]

i dont know the real answer,
but usually on /. such lists end in
9. PROFIT!!!

Re:I'm just curious

[Locke2005]

No, he's paraphrasing Bill Gates:``The obvious mathematical breakthrough would be development of an easy way to factor large prime numbers.'' -Bill Gates, The Road Ahead, pg. 265

Re:IBM should seed Cell Processors to Linux clubs.

[Anonymous Coward]

'They could call the program "Cellular Automata."'

Marvin: Life; Don't talk to me about life.

Have your heard of numerical computation ?

[gorim]

I think people who crunch numbers are getting woodies just now...

Re:Have your heard of numerical computation ?

[TheKidWho]

It's like Viagra for Physicists =)

Re:Have your heard of numerical computation ?

[DSP_Geek]

[Ed McMahon]
YOU, SIR, ARE CORRECT!
[/Ed McMahon]

Seriously, compute farms will never be the same. Even if they use double precision we're still talking an order of magnitude more processing per CPU, and the single precision numbers are just ridiculous.

What IBM really needs to do is build some libraries like Intel's excellent Math Kernel and Integrated Performance Primitives, which cover a lot of the operations heavy math users are likely to need. At the end of the day, getting a one or two order of magnitude speedup usually is worth recoding CPU-intensive apps.

Francois.

Re:Have your heard of numerical computation ?

[avidday]

Except for the slight problem of a lack of proper hardware IEEE 754 compliance in the Cell. That will cause a number of people (computational geometers in particular) a bit of heartburn. IBM/Sony/Toshiba have elected not to implement all the nuances of the IEEE 754 spec (especially tie breaking and special case values), because they are not particularly useful in a console and require more transistors. The PS2 emotion engine FPU design was the same.

Have you ever tried to determine the true sign of the determinant of a near singular double precision matrix? Even Intel's extended 80 bit IEEE FPU design in the x86 makes this harder than it should be. On the cell it will be even harder again.

Re:Have your heard of numerical computation ?

[Anonymous Coward]

Nope, can't say that I have. If anything approaches being singular, we use a little technique called "preconditioning" to take care of that....

Re:Have your heard of numerical computation ?

[avidday]

A lot of very useful geometric predicates and tests deduce their result purely from the sign of the determinant of the equation matrix. The actual system of equations themselves are never solved. For example, a simple 2-D orientation test tells you whether the test point lies to the left, right, or on a vector lying on the plane of interest, according to whether the determinant of the matrix is negative, positive or zero. Even tiny floating point approximation errors can mean obtaining a "left" instead of a "right" when the point is very close to lying on the vector. Preconditioning doesn't help you in that case.

OS X on Cell?

[Mr. Flibble]

Imagine OS X on cell... with the collusion between Apple and IBM, and OS X running on open hardware... This could be the killer OS that supplants windows.

Linux wont do it (not in the desktop arena, it does kick ass in the server area though) but OS X could very well.

That would be something to see, and I would bet, that much software that was OS X capable on Cell would ALSO be Linux capable (perhaps a recompile by the vendor? maybe native... not certan here.)

Would be nice to have a stable easy to use OS as the dominant platform. Of course, the irony would be that if this did become the case, then I suppose that Apple would eventually become as lazy and as dominant as Microsoft.

*sigh*

Still, nice to dream!

Re:OS X on Cell?

[Enrique1218]

I feel this may find itself in a Apple appliance. Maybe an advanced multimedia device that fetures an embedded version of OSX and iLife that can connect to HDTV. In two years, the desktop may be replaced with application specific devices and laptops. Only professional systems may remain.

Re:OS X on Cell?

[nokiator]

There are several contexts which can lead to Cell processor being used in future Apple platforms:

• As a media co-processor in next generation PowerMacs, and potentially even high end iMacs, similar to "AV" badged Macs from a few years back. Cell can work as a pretty good general purpose media co-processor to offload video encode/decode operations from the main processor(s). Even the current high end dual processor PowerMacs are being challenged when decoding HD H.264. A co-processor that can enable real-time H.264 encoding would make a big impact on the user perception.
• As a physics modelling co-processor for Macs to accelerate animation and games. This is really what the Cell processor is designed for in the first place, and there is likely to be plenty of libraries/engines written for PS3. This will go a long way to eliminate the existing perception that Macs are inferior game machines. The same capabilities can be used by professionals for 3D animation work.
• As the core of a home media center that can encode/decode/store/stream video/audio. If the Cell can fit the thermal and cost constraints of a game console, it would also be a good fit for a next generation media center.

Re:OS X on Cell?

[gsfprez]

anyone here who's ever worked on Final Cut, After Effects, Motion, Logic, Shake, or Maya or any number of hundreds of applications that often require seconds to HOURS of rendering can imagine Cell processors in their Macs, you collective morons that say stupid things like "why do you need a DSP processor to make file serving go faster"?

Some of us in content creation could use a little help here...

That some people are stumped by the utility of using Cell processors in Apple-built blade servers to take the place of XServes, or for what purpose would there be in 1-4 Cells in the Mac of the average Joe makes me really doubt the usefullness of our public schools.

Power users being able to add 15 animation effects with translucency and kenetics in Motion to a video with 8 layers of HD video and then watch it automatically copy straight to a DVD-R - without rendering time - makes us wet with anticipation...

and wondering when the hell hard drives are going to be able to catch up.

I would easily give up my right nut for a Apple-based blade server now that the Pro apps are starting to use XGrid for co-processing the heavy lifting portions of our work. My DVD projects of 2 hours still take long over an hour to render 2-pass MPEG2 on a high end DP G5. Multiples faster than realtime, could the Cell do.

But the average Joe? Why does Safari have to use a 8 core DSP? its doesn't, dumbass. But that's not all people do with their macs...

That iDVD render? What render? The lag is now your DVD burner - 100%.

Encoding settings for your iPod, vs. encoding settings for your files. iTunes could EASILY convert - on the fly - Apple Lossless encoded files to some kind of smaller, lossy codec as it filled your iPod. No waiting except for your iPod's slow-ass hard drive.

all this - while the Cell is still using a basic G5 at its core... so, no, Word isn't going to get any punch - but if Cell processors are as cheap as G5's, then what the hell is the issue here?

I'm damn ready for radical leaps in DSP... i'm fscking sick and tired of watching progress bars, DAMNIT! and if the Cell can do everything IBM says it can - hell, yeah, bring it on.

Server guys - try to think beyond your damn file services.

Re:OS X on Cell?

[Chaset]

I'm probably not the first one to think of this, but the said "libraries/engines" could very well be coreimage/corevideo/coreaudio. Part of the problem of using a specialized processor is that people have to code for it. If apple abstracts the hardware so that the developer doesn't have to care whether it's running on the CPU, GPU, or Cell would make the adoption so much faster.
In fact, the Cell may be the REAL reason Apple started developing Core* tech.

Re:OS X on Cell?

[Anonymous Coward]

"while the Cell is still using a basic G5 at its core"

The Cell power core is not like G5 at all, it in generally slower than the current G5s. It has fairly high frequency but it is an in-order CPU.

Re:OS X on Cell?

[Anonymous Coward]

Waiting for hard drives to catch up to the Pro Apps? One word - Xsan. You can now stream hundreds of megabytes (not bits) per second to every G5 on your network. Why even use a hard drive when you have a SAN?

Re:OS X on Cell?

[Anonymous Coward]

Not unless apple starts selling cells. OSX is proprietary software that's used to promote the sales of apple hardware. So either 'no cell' or 'no OSX'.

Re:OS X on Cell?

[Stephan Seidt]

AFAIK in the Cell they came up with a new PPC5 core BUT they stay with the same instruction set.

Re:OS X on Cell?

[Anonymous Coward]

I hope it will speed up iHype. Woohooo!

Re:OS X on Cell?

[slashflood]

Server guys - try to think beyond your damn file services.

I hear you. I am a server guy, but I also know, that file, mail and web servers are doing much more than only delivering files, mails and static web pages. See my previous post. Those guys, not realizing, that the Cell will improve even traditional server applications, are not really up-to-date or well informed.

PHP is the best example. Maybe PHP itself doesn't benefit from SIMD (SSE or Cells SPU), but all its dependencies (see your phpinfo [random example]): xml, ssl, gd, zlib, mcrypt, ...

The same is true for a mail server. Sure, its main purpose is to receive mails and store them into mailboxes. But what about ssl encryption? What about the virus scanner, that has to decompress attachments?

SIMD can help a lot in those areas.

Re:OS X on Cell?

[TheRaven64]

I would imagine that Apple would integrate the Cell with Core{Audio,Image,Video} so that only the people writing the filters would need to know about how it worked, but allow general purpose application writers to use it. Even simple things like image rendering could potentially be accelerated, making it possible for things like Safari to benefit - decompress the image on the SPU, render using Quartz 2D on the video card and the CPU is free to handle the page layout.

Re:OS X on Cell?

[derniers]

I agree, h.264 is where Apple would use cell, the dual G5s just barely decode it (1080p), and slowly encode it, my guess is that the Apple media center will be a client that encodes/decodes h.264 and connects wirelessly to the host, the current mini won't do I don't think that Apple cares that much about gamers

Re:OS X on Cell?

[99BottlesOfBeerInMyF]

making it possible for things like Safari to benefit - decompress the image on the SPU, render using Quartz 2D on the video card and the CPU is free to handle the page layout.

I'm pretty sure all the speed issues with Safari are network related. It loads every local file I have instantly. I don't really think a faster processor is going to help. Aside form that, your point is well taken. It would be useful for any number of applications (like speeding up the generation of PDFs from InDesign please).

Re:OS X on Cell?

[Anonymous Coward]

geez, go rub one out and calm down, pissface.

! Graphics only

[theid0]

I've been trying to ignore everybody's outspoken assumptions about the Cell being a graphics chip which can't do general processing for a desktop computer. The fact is that it's rightly a multi-core chip with loads of vector processing capacity. It might not be as fast on a single-threaded task, but the software world is going to adapt quickly for this type of setup because it's where the hardware is going. No semiconductor lab can (cost) effectively compete in a megahertz race anymore, so more power = more transistors (more cores).

Server programs are ahead of the curve at this point because they've had multiple CPUs in abundance for a long time. However, even today it doesn't make sense for games like Doom III to avoid taking advantage of this hardware when possible (for instance, the G4/G5 systems have had dual processors for YEARS but Id won't use them properly). For petessake, calculate audio on one processor and AI on the other...

Re:! Graphics only

[Mr. Flibble]

My guess as to why id does not support this is manpower vs. usage. IIRC, you could use dual procs with Quake 3 only on NT/2k/XP... This is likely because Carmack realized that the most common userbase for dual proc and gaming would be users on a windows machine. With signifigantly fewer users on Mac and Linux, and even fewer of THOSE running Dual Procs, it probably did not make much sense for Carmack to change the code for these OS's to add dual processor capability.

Carmack is one smart cookie. I suspect that it would take much more than a simple recompile for Linux and Mac OS to support dual procs on Doom and Quake...

Re:! Graphics only

[theid0]

It is already known that the Doom III code simply wasn't designed to be multi-threaded. However, it should be relatively easy to split well-defined functions such as the audio and the AI because they probably aren't all muxed in the same thread.

I just think it's disappointing that they spent years working on a game but didn't consider speed like they used to with previous games. Quake 3 was a popular benchmarking application, and it became really great on the Mac when the porting company (don't remember who) added vector optimizations (imagine that!) and it ran 3x as fast. It may be true that id was just trying to save money by not developing those features in Doom III, but they didn't remember one of the basic coding rules: reuse good code whenever possible. Reuse knowledge, at the very least!

Anyway, my previous point was meant to be that software developers ought to start thinking about dual cores. They've been caught up in various vector units and instruction set optimizations for some time, and the focus is about to shift.

Re:! Graphics only

[Anonymous Coward]

that it's rightly a multi-core chip with loads of vector processing capacity. It might not be as fast on a single-threaded task, but the software world is going to adapt quickly for this type of setup because it's where the hardware is going.

If so, it'll be the very first time in the history of computing that has ever happened.

Hardware runs software. Not the other way around.

(for instance, the G4/G5 systems have had dual processors for YEARS but Id won't use them properly).

Yeah, because Carmack is such a howling dumbass and all.

Re:! Graphics only

[Anonymous Coward]

"Yeah, because Carmack is such a howling dumbass and all."

Wow. That really is an amazing comment.

You really have the assumption that a guy who has spent pretty much his entire career writing indoor rendering code for x86 processors is somehow supposed to be automatically good at writing multi-threaded PPC code. That is amazing fanboyism.

What is telling about Id is John Carmack's comments about how he was 'troubled' by the move to concurrent architectures. Not sure why he felt anyone would care about his opinion on the matter.

Re:! Graphics only

[spoonboy42]

Actually, as far as I know, Doom3 is a multithreaded engine. I have an opteron-based system, running 64-bit linux (but a 32-bit linux binary, as id won't release a 64-bit one). I went from 1 to 2 processors and saw a 25% increase in framerate. AND keep in mind that I run at a resolution (1280x1024x32 with all effects enabled) where graphics tend to be more GPU bound (in my case a stock GeForce 6800) than CPU bound.

Anyway, the cell will hopefully drive forward the adoption of multithreaded game engines (for that matter, the XBox 360's 3-core processor will help in this regard as well) just as multicore CPUs start to arrive on the PC market in earnest. Considering that major next-generation engines (Unreal Engine 3, for example, which had functional demos on X360 in the form of Gears of War and PS3 with UT2007 at E3) are shooting to cover all 3 of these platforms, it's probably safe to bet that a fair amount of development time is going to be invested into making game engines highly parallelized in order to realize maximum performance.

Re:! Graphics only

[Anonymous Coward]

It is funny that you are so peecee myopic to think that game engines need to 'be driven forward' to multithreaded designs.

The gaming world does not revolve around the x86 peecee. The x86 game market is the technological ghetto. Tiny and best avoided.

Re:! Graphics only

[Anonymous Coward]

What is telling about Id is John Carmack's comments about how he was 'troubled' by the move to concurrent architectures. Not sure why he felt anyone would care about his opinion on the matter.

There are forums of which you are unaware, where these matters are discussed by those who bear the burden of actually using all of this wonderfully-parallel hardware.

Clue time: nobody knows wtf to do with these things. Not Carmack, not Sweeney, and not fucking Gandalf the White, for that matter.

I'm pretty sure the semiconductor industry is on the brink of its biggest mistake since Itanium. They have, quite literally, forgotten who they work for.

Re:! Graphics only

[Anonymous Coward]

"Clue time: nobody knows wtf to do with these things. Not Carmack, not Sweeney, and not fucking Gandalf the White, for that matter."

HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA !!!!!!!!!!!!!!!!!!!!!!

Oh yes!!! It's been a long night of coding and a magnificent retarded gem of a comment gives a much needed laugh.

Did you actually just give your range of people to listen to as one x86 peecee fps programmer all the way to the other well known x86 peecee fps programmer????

HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA HAHA !!!!!!!!!!!!!!!!!!!!!!

It's so fucking hard to believe people like you really exist out there.

Sorry you fucking dunce, Cell/PS3 is exactly what all of us console developers knew and wanted from Sony. And it is why we all are and will be for the foreseeable future on board with Sony. And yes we know EXACTLY what to do with Cell.

Go back to teamxbox/flipcode/aceshardware/and all the other x86 gamer fanboy sites. The console world doesn't give a shit if your fps 'geniuses' can handle Cell or not.

Re:! Graphics only

[Anonymous Coward]

Damn dude, what crawled up your ass and died?

Re:! Graphics only

[Anonymous Coward]

Did you actually just give your range of people to listen to as one x86 peecee fps programmer all the way to the other well known x86 peecee fps programmer????

Call it a wild-assed hunch, but I'll bet the people I mentioned have shipped a wee bit more code than you have. :)

Re:! Graphics only

[Anonymous Coward]

"Call it a wild-assed hunch, but I'll bet the people I mentioned have shipped a wee bit more code than you have. "

Uh, no. Combined, probably, yes.

Re:! Graphics only

[SnowZero]

You sir, are an idiot. GP said:

Anyway, the cell will hopefully drive forward the adoption of multithreaded game engines

i.e. the Cell, the processor for the PS3 and with roots in IBM's Power/PowerPC processors, is what is driving game engines forward. The XBox360 is also Power/PowerPC based. There is no x86 bias whatsoever, but rather a console bias. Rightly so, because those *do* drive the industry whether you like it or not. Macs and Linux certainly don't. Go back and crawl in your hole.

Re:! Graphics only

[Anonymous Coward]

Ah. In that case, I, um, bow in your general direction.

Re:! Graphics only

[beerits]

A couple of points:

1. Quake 3 used dual processors under Mac OS X and it made a large diffrence in FPS.

2. Probably a larger percentage of Mac users have computers with multiple processors than windows or even linux users.

Re:! Graphics only

[Anonymous Coward]

The Cell only has one general purpose core, which is the only one an operating system can directly schedule normal threads on.

The SPEs are entirely different from the general purpose core, have to have their "threads" written specifically for them, and are only good at very limited kinds of tasks.

Using the SPEs is non-trivial and extremely non-portable. Certain types of software would undoubtedly use them (games, 3d, high-volume audio/video processing etc.), but most software would simply suffer from degraded performance compared to equivalent general purpose processors.

Doing AI on a SPE also makes absolutely no sense, since they're good at processing streams of numbers (3d, audio, video), but not branching around logic routines.

Multi-core is definitely the way to go, but for most uses, a multi-core general purpose CPU is going to be far better than a single Cell.

However, presumably the Cell is going to be cheap, so alternately, you might have general purpose computers with 4 or more Cells, assuming that motherboards for such beasts could be made affordable.

Re:! Graphics only

[mrseigen]

Quake3 used SMP on Mac OS X and Linux (there's a quake3-smp binary that comes with the latest point release for Linux). I don't know about Windows.

Re:! Graphics only

[Tape_Werm]

I call bullshit on that. Furthermore, you're nothing more than a bandwagon jumping troll. Seriously, when is slashdot going to give me the power to electrocute people like this?

Re:! Graphics only

[Anonymous Coward]

Yes it is clearly a crazy claim that there are many people out there who have either been in the games industry longer or who have shipped more games than either of the two guys mentioned...

Everyone knows that Carmack is, like, a god!!!!!

Re:! Graphics only

[IntergalacticWalrus]

Have you ever done any game programming work? I guess not, because if you did you'd probably knew by then that games can't be magically multithreaded. Of course it's so easy to whine about lack of multithreading in a game when you're only playing it. But that's okay, I'll explain it to you.

First of all, multithreading a game more trouble than it's worth, because a video game depends on having everything synchronized together. The second you offload a given task to another thread, it becomes the jurisdiction of the operating system's thread scheduler to decide when that thread gets to work on whatever it needs to do. That's not good, because you end up having every thread wait on each other to finish its required work in a given timeframe before going on to the next processing timeframe. If the OS decides to give less time than necessary to a single one of those threads, you end up with a loss.

Also, add into the fact that multithreading is the univeral way of not only making harder to debug code and create nasty deadlocks (any programmer eventually comes to this) and it suddenly becomes even more unattractive. A widely used rule of thumb in software engineering is "use threads only if it's really necessary for your program flow". It is necessary for a server to process multiple queries at a time, and it is necessary for a GUI app to avoid locking up the interface while there is a processing task running in background, but in the case of a game everything must be live and synchronized. There is no immediate need for threads.

The only thing that makes sense putting on another thread is the sound processing, because it's essentially a strictly background task, that gets called on-demand everytime a sound or music needs to be played, and doesn't need strict synchronization with the gameplay. And yes, every modern mainstream engine already does this, including DOOM III.

So finally after all this trouble, what advantages do you get? Not much. Most multi-CPU systems are only dual, and we have 1) the main thread, 2) the sound thread, and 3) the operating system's kernel. That's at least three tasks on two processors. Granted the kernel and sound thread does not work as much as the main thread, but nevertheless you're already overloaded, and therefore those other threads you want to create will only add minor performance increases by using what's left of the sound/kernel processor.

So in short, multithreading a game only adds marginal performance increases at a way too significant development (and stability) cost.

Re:! Graphics only

[Anonymous Coward]

"So in short, multithreading a game only adds marginal performance increases at a way too significant development (and stability) cost."

What the fuck?

No wonder x86 game programmers are shitting their panties over Cell if that is what they really think.

Re:! Graphics only

[Anonymous Coward]

Yes it is clearly a crazy claim that there are many people out there who have either been in the games industry longer or who have shipped more games than either of the two guys mentioned

No, it's not crazy; there obviously are plenty of people like that. But they don't show up on Slashdot with the attitude of an 8-year-old with Asperger's and a technical perspective that is shared by virtually no one in the industry.

Re:Very promising technology= investment opportuni

[TimeForGuinness]

It is exciting...but (not to sound trollish)...I'll believe it when I see it.

IBM Blade Server Management

[AntiFreeze]

This won't go anywhere if IBM doesn't clean up its blade management console.

I've been doing extensive research on blade servers recently for my company, and when it comes down to it, IBM's centralized management for blade servers is hands down the worst in the industry. RLX used to be the best, but they're out of the business now. HP was #2, now they're the leader. Egenera is doing some really cool things, but their setup is just way too expensive (almost 5 times the price of the other leading blade systems).

So, even if these cell blades were to be the coolest thing ever, if IBM doesn't make an investment into improving their management software, no one's going to buy these things unless they already have a large investment in IBM hardware or are just downright masochistic.

Basically, what it comes down to is, someone needs to buy the RLX software, it's on the market now. If I were IBM, I'd buy this and retool it for IBM blades. What I'm scared of is Dell buying the RLX software. Dell blades suck, but with the RLX console, even I would consider buying Dell blades, that RLX management software is just that good.

In short, if I were IBM, I'd buy RLX in a second, and catapult myself to being the industry leader in blade servers.

Re:IBM Blade Server Management

[ciroknight]

I'll one up your bid; IBM should either 1) Hold a $$million dollar bounty on Open Source Blade Management software, or 2) Buy RLX and open it to us!

It's more likely that the first will happen, as IBM has had problems enough acquiring other's software and using it for any purpose (take a look at the SCO case). And with their investment so deeply in Linux right now, I'd say this is just a bit over the horizion for Big Blue.

Re:IBM Blade Server Management

[Anonymous Coward]

I recently attended a presentation from BladeFusion (http://www.bladefusion.com/ that maybe is the answer to your prayers. Is a tiny software layer above IBM's BladeCenter hardware, that gives management felxibility to the platform. The presentation was held in the local IBM headquarters, and IBM representatives fully endorse this vendor for the management layer for BladeCenters.

Currently it only supports Linux (several predefined RedHat versions, as well as customized ones as supplied by third party ISV), but the y promise Windows support in the coming months (post-summer). From this management software you can assign specific blades to specific applications, load a blade with a predefined (and customizable) operating system image, define load balancing and high availability, with fail over mechanisms in case of high CPU load, blade or service down, etc. All from a very simple Active-X or Java (don't know for sure) web interface.

It looks simple and powerful enough, don't know the pricing but if you are after BladeCenter I suppose money is not the biggest problem. Hope it helps.

Re:IBM Blade Server Management

[turgid]

IBM has a vested interest in making things difficult and complicated for its customers. After all, it makes its money from support.

Re:IBM Blade Server Management

[Anonymous Coward]

I've been trying to use one of their BladeCenter type cabinets with some Power-based blades running SuSE for a while now, and I must say the thing is truly awful at the moment. Especially the firmware and web manglement consoles seem half-assed at best. There may be advantages to the platform and internal architecture, but they sure are hidden well under heaps of Issues from a sysadmin point of view.

Maybe if it matured a bit, but they've been out for a good while now and I don't see any truckloads of blades being hauled into any data centers, so for now it seems the only advantage to the blades is the low floorspace per node ratio.

Re:IBM Blade Server Management

[Anonymous Coward]

and catapult myself to being the industry leader in blade servers.

Errrrr, there are the industry leader in blade severs.... http://news.com.com/Dell%2C+HP+gain+on+IBM+in+serv er+market/2100-1010_3-5721535.html?tag=cd.top ... in terms of revenue.

The future for Apple as well?

[Biggerveggies]

I am curious to see how this will work out, especially since the Apple+Intel article came out in the Wall Street Journal.

(Think Secret's take: http://www.thinksecret.com/news/0505itunes49.html)

I think this is a better indication for Apple's future processors, as opposed to the Intel rumours.

Re:The future for Apple as well?

[child_of_mercy]

they'll need to work on the heat issues in a big way if it's going to fit into any apple form factor.

Deep thought...

[kernel_dan]

If IBM has ported the Linux kernel to the Cell processor, does that mean that they have to release the source code as a derivative work of the GPL if they ever sell a Cell-Blade with Linux?

Re:Deep thought...

[Dancin_Santa]

Since they have obviously already released the binary to Sony for testing on the server, that counts as a distribution.

Under Section 3b of the GPL, the source must be given to any third party that requests it.

Are you a third party? :-)

Re:Deep thought...

[kieronb]

But under section 3a they only have to give the source to the second party, in this case Sony.
OR they can offer to give it to any third party.

But anyway, I doubt there'd be many changes to release - I imagine the kernel would just be dealing with the PowerPC master core, the stuff to operate on the vector cores would probably happen in userspace.

Re:Deep thought...

[chris_eineke]

There's a party down the street and one across the street. I'm a party guy, so I guess I have to say:

Yes, I am a third party. :)

Re:Deep thought...

[Anonymous Coward]

Gosh, I wish someone had told the FSF. They wrote the license and apparently their interpretation is wrong according to you.

Re:Deep thought...

[bersl2]

I saw some code flying around on a mailing list somewhere. This looks good enough:

http://seclists.org/lists/linux-kernel/2005/May/26 57.html

Re:Deep thought...

[megan_of_wutai]

They already have.
One of the more interesting posts: http://lkml.org/lkml/2005/5/13/218
Arnd Bergmann works for IBM, btw.

Re:Deep thought...

[Rakishi]

*sigh*, such lack of reading skills. If a third party gets the source code they can legally use it... the key word being "if they get it." IBM may give it to any third party but they don't have to.

Re:Deep thought...

[SnowZero]

Did you read the FAQ entry?
...this does not require you to *do* anything physically for them.

We have a license to the source, but they don't have to give it to us. What this means is that anyone who has it can give it to us and IBM can't stop them. The GPL is very clear that you only have to give source to those who get the binary, but the license has a guaranteed sublicense to whoever you give it to, so they can give it to anyone.

This is all moot anyway, as IBM has posted the sources to LKML already. Search for "BPA" on any list archive.

Re:Deep thought...

[ctid]

Under Section 3b of the GPL, the source must be given to any third party that requests it.

This is not correct. They must offer the source code to anyone they distribute the binary to. They are not obliged to offer the source to "third parties". They can't prevent anyone that they do give the source to from giving the source to other people.

Re:Deep thought...

[petermgreen]

this isn't correct either

they must either distribute the source and the binary together OR include a written offer with the binaries to give the source to anyone.

Re:No.

[Lord Bitman]

you spelled frist psot wrong.

Re:Very promising technology= investment opportuni

[soricine]

After you've read Blatchford's write-up, read this for a reality check:

http://arstechnica.com/news.ars/post/20050124-4551 .html

It uses such terms as 'hogwash' and 'wild-eyed and completely unsubstantiated claims'. Ouch.

You don't need to be the NSA to factor primes

[Anonymous Coward]

In fact, it's a faster algorithm than most anything else.

let x be the prime you want to factor.

factors: x, 1

Someone call the RSA! I think I've found a flaw in their encryption!

Troll

[Anonymous Coward]

IBM is the biggest supporter of open source. Of course they are more than happy to release the source code.

Obligatory

[threaded]

Imagine a beowulf cluster of ...

x86 emulation?

[Timbotronic]

Just wondering, could one or more of the supplementary cores be used for translating x86 instructions to RISC (and back) for the Cells main processor? I'm not really familiar with the Cell's architecture but it'd be interesting to see what companies like VMWare could do if this was the case.

Re:x86 emulation?

[Anonymous Coward]

As long as a cell is Turing complete, which I assume they are, yes.

Re:x86 emulation?

[Anonymous Coward]

Assuming the extra cores have suffient programability (pretty sure they do), yes they can. Now I'm not sure that it would be that usefull, as the performance you would have would be a little less then if you just had 2 ppc cores doing the emulation.

The problem is that there is no easy way to convert x86 code into ppc code optimially in real time. You can convert each x86 instruction into a couple of ppc instructions, but then you are getting a percentage of the performance.

what would be interesting is if someone built a hot spot compiler (similer to java) where commonly used code was actually recompiled for ppc as the code was run. so that extremly common code could run at near real time.

But considering the most common ppc x86 emulator is owned by microsoft, why would they do something that stupid

Not A Troll

[Anonymous Coward]

Why is he a troll? He asked a relevant question.

What you sayin'?

[HiggsBison]

Using a two-tier or three-tier approach to client/server architecture, with something like a full-duplex GigE connection to fat, diskless clients and you have some real potential.

Maybe it's late, but am I the only one who thought he was saying that IBM had "fat, dickless, clients"?

Low enough heat...

[Anonymous Coward]

Well,

If the Cell has low enough heat to be fitted in a blade, perhaps a future version could be cooled well enough to find its way into a PowerBook?

Would *that* shut up the "Apple has to switch to Intel to have faster cooler laptop chips!!! or they're D000000Med!!!!! " crowd? Maybe? Perhaps?

Re:Low enough heat...

[Anonymous Coward]

did you see the SIZE of that thing?

Re:Low enough heat...

[atcurtis]

Have you ever seen... or even listened to... the kind of fans IBM has in their BladeCenter?

Two large drum fans - the entire thing sounds like a jet engine when it powers up. It takes it a while before it settles down to a much much slower speed - but I suppose it is surplus cooling capacity which could prove handy to ship a blade which has high heat dissipation needs.

Re:Low enough heat...

[ciroknight]

Sorry, but that's extremely not likely to happen.

The Cell's max heat production should be almost spot on with the G5, plus seven times that of a typical SPU (which could be fairly accurately estimated by relating the thermal production per transistor for the G5 in the unit with your unknown over the transistor count for the SPUs). This is likely to require liquid cooling once you get into prolonged usage, and it's highly likely the blades IBM looks to build with it will be a Blade-enclosed heatpiping system.

What *needs* to happen for the continued production of laptops in Apple's camp is a low-power equivalent of the G5; do like the Intel engineers did. Look at the archetecture. What there is good, and what is bad. What can you take from the G4 that was good and apply it to the G5. Things like the Velocity Engine could be scraped if it's discovered it's thermal production per performance isn't up to par (as I hypothesize it does, along with SSEn). Any last step you can do to reduce the transistor count, but keep the performance up.

The Cell is clearly built _not_ to be portable. It's built for use inside of things that will stay relatively static, and require high amounts of processing power. Things like High Def LCD/Plasma televisions (Picture in Picture, perhaps?), things like Sony's PS3, things like your fancy new Streaming Media Server, or your super fancy new, multi-channel encoding TiVo.

Re:Low enough heat...

[argent]

What *needs* to happen for the continued production of laptops in Apple's camp is a low-power equivalent of the G5;

There's nothing wrong with the G4 that a faster memory bus won't fix. The G4 is clocking up to 2 GHz even with only a 7-stage pipeline, and the only thing holding it back is the 166 MHz memory bus. It really reminds me of the P-III versus P-4 mess, where the P-III was faster clock-for-clock than the P-4 but was held back by its slower FSB (133 vs 400 MHz, as I recall).

Which is why Intel's newer mobile chips are based on the P-III core. Oh, Intel describes them as a mix of the P-4 and the P-III, but if so it's a mix that's almost all P-III. "what intel's engineers did" wasn't "take what's good in the P-III and apply it to the P-4", they took the best part of the P-4, the faster external bus, and applied it to the P-III. And that's EXACTLY what Apple needs. Not a cut-down G5, but a liberated G4.

And that's what Freescale's been working on. The Freescale e600 is basically new G4 with a MUCH faster bus. I don't know if the Freescale e600 will actually end up faster clock-for-clock than the G5, but it's certainly possible... at least for code that generates a lot of pipeline flushes. But it will certainly be faster watt-for-watt.

Apple's already using the standard G4 chips from Freescale, so going to their new single- and dual- core chips seems like a MUCH more logical next step than trying to build a "Power-PC-Lite" G5.

Heat sinks

[CBob]

You'd that that with all the time & $ invested, they'd at least show 'em off with active cooling a bit more advanced than the BIG sink/BIG fan combo.

An alpha teaser I wonder, or a bit of intended misdirection?

Uhhh

[ad0gg]

Cell is a normal PPC chip with 8 SIMDs. Last time I checked, MMX didn't revolutionize PC chips. Being MMX is Simd instructions for pentiums. Cell is nothing new, they just call the simds,a Synergistic Processor Unit. Gotta love marketing speak.

Re:Uhhh

[gorim]

The equivalent for Mac / PPC - altivec, velocity engine, or vmx (whatever you want to call it) certainly revolutionized that platform.

The fact that on the x86 platform there was little revolution, or one little seen, may be more a reflection of the platform itself.

Honestly, people who can't see the value of making true and powerful use of SIMD are missing the boat. That is what the future is all about.

You look at your cellphone, mp3 player, mp4 codecs, digital tvs and radios, it is SIMD that makes all that happen (through DSP).

More visible to you, look at your GPU's powering your favorite games, specialized SIMD.

The main CPU is truly just the conductor and SIMD is the orchestra, as Sony puts it.

Re:Uhhh

[plalonde2]

That's because MMX was so badly designed. Overloading it on the floating point registers was just plain brain dead.

Go write some VMX code and compare it to MMX - especially if you have any streaming/number crunching to do.

Re:Uhhh

[chthon]

The only reason MMX instructions exist, is because at the time (about 1996), there where some large companies which had initiatives for building powerful multi-media chips for PC add-on cards.

MMX was introduced by Intel to kill off those initiatives.

Re:Uhhh

[Anonymous Coward]

Fixed using the SSE extensions.

Re:Uhhh

[TheRaven64]

The reason SIMD did not really revolutionise anything on x86 is that MMX sucked, and sucked badly. It added extra registers, which could not actually be used 90% of the time (just adding half a dozen GPRs to the x86 ISA would have given more of a speed boost than MMX in many cases, as x86-64 has shown). It required a context switch (very expensive on x86) to use, and it could not be used at the same time as the regular FPU (as I recall). Oh, and I seem to remember that it only worked with very small vectors, and only integer instructions.

In short, MMX was a lame duck, but was hyped to a huge extent by Intel. The shortcomings of MMX have (at least partially) been addressed in subsequent x86 vector units, but by that time it was too late. If you are doing scientific computing, and can target your code to a particular CPU, this is fine, but who[1] has the effort to optimise code for MMX, SSE, SSE2, SSE3, 3dNow!, etc?

On the Mac, AltiVec was the first vector unit to be released, and was of a very high, proven, quality. Subsequent chips have retained compatibility with exactly the same vector ISA, meaning that everything written for a G4's vector unit will gain the same benefit on a G5[2].

[1] Microsoft, in DirectX, and a few other people, but not very many.

[2] Clock-for-clock, I believe the G5's vector unit performs slightly worse than the G4's, however G5s usually come with higher clock speeds, so it makes little real-world difference.

Re:Uhhh

[maraist]

MMX was introduced by Intel to kill off those initiatives.

Except that the path to the GPU was still formed and is essentially winning the specialized device wars over Intel. So while you still do find occasional systems with primative graphics devices thereby using the CPU for transformation purposes, the original point of MMX has largely been lost (with the exception of a 515% boost in performance here or there in applications that bother using it's legacy (SSE3)). I don't know about the on-board audio chips.. I know that 3D audio was previously off-loaded from the CPU, but the on-board audio chips are considered just as good these days, and I don't know if they rely on the CPU to do most of their work. So Intel might be considered a winner in this department.

Re:Uhhh

[Anonymous Coward]

The only reason Mac users have a hardon for AltiVec is because the G4 CPU stagnated for 2 years, so AltiVec was the only performance improvement they saw in that period.

Re:Uhhh

[MojoStan]

In short, MMX was a lame duck, but was hyped to a huge extent by Intel. The shortcomings of MMX have (at least partially) been addressed in subsequent x86 vector units, but by that time it was too late. If you are doing scientific computing, and can target your code to a particular CPU, this is fine, but who[1] has the effort to optimise code for MMX, SSE, SSE2, SSE3, 3dNow!, etc?

On the Mac, AltiVec was the first vector unit to be released, and was of a very high, proven, quality. Subsequent chips have retained compatibility with exactly the same vector ISA, meaning that everything written for a G4's vector unit will gain the same benefit on a G5[2].

Intel's SSE3 is a superset of SSE2 which is a superset of SSE which is a superset of MMX. Everything written for MMX will gain the same benefit on SSE, and everything written for SSE... we know the logic. I don't see the difficulty here.

AMD's 3dNow!, on the other hand, was decent (and came before SSE and Altivec) but not used much because of AMD's weak presense at the time (K6-2). AMD has since adopted Intel's SIMD instructions, so very few people target 3dNow!.

I'm not saying SSE3 is any better or worse than Altivec (or that MMX was not overhyped), but I don't understand why SSE/SSE2/SSE3 were "too late" or difficult to target your code for.

Re:Uhhh

[petermgreen]

iirc the reason they overloaded it on the floating point registers was to allow it to work without support from the OS.

adding extra registers would have meant that the feature would have to be explicitly enabled by a supporting OS that new how to context switch the new registers and it would have been unusuable without such support.

Re:Uhhh

[plalonde2]

True, but still sad. It made it nasty enough that it didn't get used enough, and did a lot to slow adoption of SIMD units :-(

Imagine

[Araxen]

A beowolf cluster of these...

I wonder

[bersl2]

Was it really an engineer who said these things?

If so, did he say them of his own accord, or was he instructed to say certain things? And even if that is so, it is still refreshing to hear somebody besides a marketing or management bot speak to the press.

It's not Sony's hype

[Anonymous Coward]

IBM designed the Cell

Re:It's not Sony's hype

[TyrionEagle]

And another person with a damages sarcasm centre in thair brain!

and here I thought...

[Anonymous Coward]

they were talking about biological computers.

oh well, maybe next time!

Apple is hardly any better than Microsoft

[leoc]

I'll take Linux on Cell over OS X on anything any day.

Apple is already as arrogant and obnoxious as Microsoft. For example, despite the fact that OS X in its current form would not exist without the efforts of the Open Source community, Apple is still actively working behind the scenes in Europe to destroy the ability of the open source community to work with their proprietary formats.

Maybe you don't get it

[Anonymous Coward]

Remember this?

Streaming a Database in Real Time
http://developers.slashdot.org/article.pl?sid=05/0 1/21/2223234&tid=221

So was your questions why would someone want a processor that is good at streaming on a server?

And for things that don't needed streaming operations, you do still have the PPC to do regular stuff.

Looking at

IBM Plans to Open the Cell Processor
http://developers.slashdot.org/article.pl?sid=05/0 5/24/1744205&tid=136&tid=8&tid=10&tid=137

I think there's a presentation linked to in the article on different coding models.

It looks like you could get the PPCs to do a lot of stuff without getting in the way of the stream processing units too much.

Carmack has tried...

[Namarrgon]

Carmack explored threading the Quake3 engine pretty thoroughly - and concluded that it really didn't help much, due to the nature of the problem - high-bandwidth communication between threads.

Some types of computing problems (e.g the compositing app I work on) multithread very well, and some just don't.

It's possible Q3A might thread better on a Cell, due to high bandwidth between SPEs - but then again, he was using a the second thread for vertex processing, which is done by the GPU these days anyway.

Re:Very promising technology= investment opportuni

[jericho4.0]

Keep your money in Sony. Chip fab doesn't make a lot of IBM's money, IIRC. The PS3, OTOH, is a huge part of Sony's future.

I don't get it-Trip the processor fantastic.

[Anonymous Coward]

"What kind of servers *ARE* these??"

Welcome to the WWMW (World Wide Media Web).

Graphics only-"Freedom" processors.

[Anonymous Coward]

"No semiconductor lab can (cost) effectively compete in a megahertz race anymore, so more power = more transistors (more cores)."

asynchronous logic

DRM!

[McGiraf]

From page 2 of this article:

http://www.blachford.info/computer/Cells/Cell2.htm l

"Some will no doubt be turned off by the fact that DRM is built into the Cell hardware. Sony is a media company and like the rest of the industry that arm of the company are no doubt pushing for DRM type solutions. It must also be noted that the Cell is destined for HDTV and BluRay / HD-DVD systems, any high definition recorded content is going to be very strictly controlled by DRM so Sony have to add this capability otherwise they would be effectively locking themselves out of a large chunk of their target market. Hardware DRM is no magic bullet however, hardware systems have been broken before - including Set Top Boxes and even IBM's crypto hardware for their mainframes."

Cell-based Server Blade Demonstrated

[ndansmith]

First we give criminals cable TV and now web servers? What's next?

NEWSFLASH!

[gorim]

An "xbox" related website says Sony is all hype!

Additionally, the cream of the crop of that website know more about hardware than the people designing the chip (IBM and Sony).

Yeah, hype is why I ignored anything but the raw specs and architectural papers on the cell.

Cool!

[kamukwam]

A computer made from living Cells! Technology advances fast these days!

Re:Gulliable IBM Falling For Sony Hype?

[Anonymous Coward]

"Gulliable" ./ reader Falling For xbox fanboy site Hype?

Re:Very promising technology= investment opportuni

[Anonymous Coward]

The first Cell based desktop computer will be the fastest desktop computer in the industry by a very large margin.

Fucking hell, some people on the internet really do spew a lot of hyped garbage.

That John Carmack

[Anonymous Coward]

What a lazy bastard.

show us the numbers

[cahiha]

With Cell, IBM keeps talking about "theoretical GFLOPS". I don't care about theoretical numbers. What I care about is how fast the thing runs when I run normal code compiled with a normal compiler and (possibly hand-optimized) numerical libraries.

So, what kind of SPECfp numbers does the thing get? What kind of BLAS performance does it get?

They have 2.6.11 running on it, so compiling the benchmarks should be trivial. If they haven't published anything yet (I haven't seen it), we have to believe that the numbers are less than impressive.

(Another company used to make inflated claims about the performance of their processors by computing theoretical maximums for a few SIMD instructions, unachievable in most real code. When people actually did some real benchmarks and published them against the wishes of the company, they found that their processor was no faster MHz for MHz than Pentium on real code with real compilers.)

Re:show us the numbers

[Jules Labrie]

The Cell is basically a Vector Processor. So it's not gonna be really fast for compiling or such things like that. IBM just took the opportunity of the PS3 to develop the perfect processor for super-computing (whose task are often matrix-based). Server with cell ? No advantage, or so few. Games ? Becomes interesting, but that's all. Supercomputing ? Here you are.

I have to say, this Cell is really a great marketing coup ! Everyone is speaking of this processor, even in the biggest newspapers of the mainstream press... Not really seen that for a few years !

Re:show us the numbers

[Anonymous Coward]

when they finally make it into the clustering lab, we'll let you know ;)

Re:show us the numbers

[argent]

What I care about is how fast the thing runs when I run normal code compiled with a normal compiler and (possibly hand-optimized) numerical libraries.

It'll run them exactly as fast as any other PPC 970 core. As far as I can see from the information that's been released so far, to use the coprocessors at all you'll need to redesign your application around an asymmetric coarse-grained parallel processing model, with explicit memory management to feed data to the shared RAM the SPUs have access to.

Re:show us the numbers

[cahiha]

It'll run them exactly as fast as any other PPC 970 core. As far as I can see from the information that's been released so far, to use the coprocessors at all you'll need to redesign your application around an asymmetric coarse-grained parallel processing model

Applications are already written for that model: they use APIs like BLAS and LINPACK that hardware vendors can reimplement to take maximum advantage of their parallel capabilities, and they use intrinsically parallelizable language primitives (like Fortran's parallel array primitives). Furthermore, many applications have been parallelized for either multithreaded execution or MPI. In different words, today's high-end numerical applications already contain the hooks and declarations for extensive SIMD and MIMD parallelism.

If the Cell architecture can't provide high performance for those kinds of existing parallel codes, then there is something wrong with Cell and/or its compilers.

Re:show us the numbers

[cahiha]

Supercomputing ? Here you are.

Well, and we have benchmarks for that. That's why I'm asking what its actual performance on real numerical code is, not some theoretical maximums.

Re:show us the numbers

[Jules Labrie]

As far as I know, the mainboards and all the stuff that still has to come with the Cell processor are still in development phase ; so don't expect to have facts right now from IBM. We will have to wait that the hardware comes to the market. I'm very curious too to finally know what this processor is really able to do.

Re:show us the numbers

[benjcurry]

I agree, to a certain extent, but the Cell Processor is going to require some changes to the way programs are written to get the full benefits. It's understandable that they don't want to put numbers out there before there is some software out there to take advantage of it. I would begin to worry if we don't see some numbers in the next...well...before the year is out. Sony's vote of confidence is reassuring, though. They wouldn't ruin the PS3 with a crappy processor would they?

Re:show us the numbers

[Anonymous Coward]

People don't design processors without detailed simulations and analysis using real software systems.

Furthermore, they have Linux 2.6 running on one of those boards, and Sony also must have working devices.

They know pretty well what they can realistically get out of of the Cell on real applications, they just aren't telling us. One thing that seems clear: it's not 200 GFLOPS per processor.

Re:show us the numbers

[Locke2005]

Another company used to make inflated claims about the performance of their processors by computing theoretical maximums for a few SIMD instructions, unachievable in most real code.

Uh, EVERY CPU manufacturer makes inflated claims about the performance of their processors! E.g. the FLOPS numbers are based on pure single-precision fused multiply-add instruction, neglecting data load and store times, and ignoring the fact that single precision floating point is useless for most applications due to rounding error. In practice, I'd be suprised to see them get more than 100GFlops out of this even in a hand-optimized HPL benchmark.

Also, this board is obviously using the preliminary 90nm chips; the final 60nm SOI chips should disipate a lot less power, so theoretically the monster heat sinks shouldn't be necessary. Of course, I'll beleive it when I see it, and I'd love to see power figures and BLAS benchmarks on the final chip too. Yes, compiling BLAS and MPI is trivial, but optimizing for a new architecture is a long trial-and-error process... I wouldn't want to publish the numbers from a default compile either, they are guaranteed to suck.

Re:show us the numbers

[argent]

Applications are already written for that model: they use APIs like BLAS and LINPACK that hardware vendors can reimplement to take maximum advantage of their parallel capabilities, and they use intrinsically parallelizable language primitives (like Fortran's parallel array primitives).

That doesn't make a bit of difference: unless they have actually implemented these libraries on the Cell processor, running the benchmarks YOU want will tell YOU nothing about the performance of the Cell using those libraries, because they'll be running that code on the PPC. It's not just a matter of running that benchmark under Linux as the OP implied.

If the Cell architecture can't provide high performance for those kinds of existing parallel codes, then there is something wrong with Cell and/or its compilers.

Given that the only people working on developing software for the Cell right now aren't doing it in Fortran or using the BLAS or LINPACK libraries, but more likely in C and C++ and using OpenGL and Renderman and NovodeX, then it's quite possible they don't even HAVE a Fortran compiler for it yet... so I suppose that counts as "something wrong with its compilers". :)

Re:show us the numbers

[Anonymous Coward]

Uh, EVERY CPU manufacturer makes inflated claims about the performance of their processors!

Uh, hence the question about SPECmarks, dude.

and ignoring the fact that single precision floating point is useless for most applications due to rounding error

You can safely ignore that "fact" because it's not a "fact".

I'd be suprised to see them get more than 100GFlops out of this even in a hand-optimized HPL benchmark.

I'd be surprised if the thing is faster than an Athlon, actually.

The ultimate render farm?

[ArronP]

Surely the logical use for a dense cell-based server would be as a monster render farm. Think of the capability of a rack of these things - Weta digital would be able to render LoTR in real-time!

The other use I can think of would be as a game server farm. IBM / Sony hae always said that the cell architecture was designed to off load workload to other cells on the network, surely there's an opportunity for ASP's to provide the heavy lifting behind a cell client device.

Also try this one at EETimes

[ahfoo]

Here's a few of the key tidbits from a two page article about Open Source and the Cell.

Regarding portability. . .

The team is actually running Cell at 3.2 GHz. Although the developers talked about the chip's going into handhelds such as next-generation Playstation Portables, that's not currently practical.

"The architecture is too power hungry for a handheld device," said Krewell. "They would have to do a significant re-design to get this into handhelds, and that's where a lot of the interesting business is these days, for things like portable MPEG-4 video players."

As for servers:

The high-end supercomputing option that Cell members have discussed may face hurdles as well. As currently architected, only two Cell processors can be directly attached to each other. A separate switch is needed to link more processors into the kinds of large arrays of CPUs used in supercomputers like IBM's BlueGene/L.

. . .

One thing the architecture clearly lacks is any of the peripherals, such as Gigabit Ethernet media-access controllers and packet-processing blocks, that would be of use in networking scenario.

Here's the article.

PS3 = 1% of Human Brain

[mattcoz]

According to Sony that is, and they would never lie about their specs.

Re:Very promising technology= investment opportuni

[hahn]

You don't really think that IBM plans on making money selling Cell processors just through Sony? This topic indicates where they plan on making the real money - through server sales, and the associated software and services that they sell along with it. In my opinion, they are simply using the PS3 as a beta test of sorts and to showcase/market the power of Cell, as a MEANS to promote server sales. IBM didn't get to be the company they are today by making foolish business investments.

Re:Very promising technology= investment opportuni

[Slashcrap]

Guess it is time to invest in Sony and IBM! This technology really looks promising, especially when you read this article --> http://www.blachford.info/computer/Cells/Cell5.htm l

I've said it before and I'll say it again - Blachford is a complete kook. He thinks he's invented a means of faster than light travel and if that isn't obvious kooksign, I don't know what is.

Oh, and he writes for OSNews. Two strikes and you're out kookboy.

Unfortunately he wrote a piece stating that Apple would adopt the Cell for Macs, making them the ultimate gaming platform. So he has plenty of people willing to defend him, realism not being one of the premier traits of the Mac community.

Please don't disagree until you've explored his site a bit or you will make yourself look foolish.

And as another poster said - check the rebuttal to his wide eyed claims on Ars Technica. You don't mess with Hannibal, beeyatch.

Re:Very promising technology= investment opportuni

[jsn13]

reality check, hmm. so, where exactly is `reality' there? the article doesn't say a single word about the Cell processor architecture. author's just being nitpicky about Blachford's analysis.

well, let's be nitpicky too, then.

• 1st paragraph summary: "Blachford sucks at understanding of basic computer architecture". ok, let's see.
• 2nd p. summary: "estimated benchmarks are all speculative (and, thus, it's hogwash)". oh great, mr. Hannibal probably knows how to get real (non-speculative) benchmarks for Cell. and nevermind Blachford says: "This is something of a "calculated guess", again based on the theoretical maximum computing power being achieved (in 4 Cells). It makes a lot of assumptions which may be in error so I will not be the least bit surprised if this figure is miles out...".
• 3rd p. summary: "Blachford talks about `no cache' and `local memory', but actually `local memory' == `cache'". someone obviously missed the detailed explaination in Blachford's article about exactly how local memory in Cell is different from cache (with terminology issues explicitly covered).
• 4th p. summary: "Blachford also declares that the longstanding problems inherent in code parallelism and multithreaded programming are now solved, because the Cell will just miraculously do all this stuff for you via fancy compiler and process scheduling tricks.". from what i can see, Blachford mentions the word `compiler' in his article exactly once: "Cells will have compilers just like everything else". and Blachford also writes:

  The Cell Compiler Will Magically Make The Code Parallel This is not true and I didn't say this. You still have to break up problems into software Cells. I still cannot figure out why people think I said this.

oh well. Anyway, I could go on, but I'll stop here. You get the idea.

PS2 "Graphis Synthesizer"

[distantbody]

They did something exactly like this with the PS2s GS (CPU in sony speak), except i think it was billed as a CG movie renderer.It was called the GScube,and i remember that one of its demo scenes had thousands of ants from the 'Bugs Life' movie rendered in realtime. It was a nice bit of tech, but as far as i know no one in the industry bothered with it. I dare say that this Cell-based server is just another GScube, only there to generate hype for the Cell/PS3, just as the GScube did for the GS/PS2. I have no problem with it if that is the case tho.

Re:PS2 "Graphis Synthesizer"

[iapetus]

Um. The GS wasn't the PS2's CPU - it was its GPU. The CPU was the 'Emotion Engine'.

probably only running on the central powercore

[nietsch]

Though it is very nice to see that IBM ported linux this quickly, I think they cut some corners. The cell has a central powerpc core, and 8 (or more) accesory processing units. The processing power lies in these APU's, not in the central power core. The APU are also very specialised, so you will ot only have to allow acces to the cell from the OS(and manage those), but you also have to write the userland programs that take advantage of the APU's strong points.
That applies to every program you want to use the apus, so the chance that this happens overnight/soon is pretty slim. Heck, they might even need to rewrite the benchmark programs for it.

Because they have not released any real benchmarks and only talk about theoretical numbers, i think they have not finished the porting fully (or have very disappointing benchmark numbers).

Giving early acces to LUGs would be nice for the street creds, but will not speed the code development of the mostly proprietary code that needs to run on it. Giving it to Gimp/Blender/other developers might work, if it comes with a crash course cell programming.

Re:probably only running on the central powercore

[Anonymous Coward]

Giving it to Gimp/Blender/other developers might work, if it comes with a crash course cell programming.

Yeah, I bet IBM really care about Gimp/Blender/GET OUT OF YOUR NICHE LITTLE MAN

Re:probably only running on the central powercore

[Anonymous Coward]

You heard it here first, LUGs are where the street cred starts.

Re:probably only running on the central powercore

[timeOday]

The APU are also very specialised, so you will ot only have to allow acces to the cell from the OS(and manage those), but you also have to write the userland programs that take advantage of the APU's strong points. That applies to every program you want to use the apus, so the chance that this happens overnight/soon is pretty slim.

Humbug! Code up an optimized Blas for Lapack, and you will have a vast number of scientific apps ready to burn rubber.

I'm pretty excited about this story, because it means IBM has the intent to make a blade server from the cell. The current state of the product isn't that important. 2.8 TFLOPS from a 7-blade rack sounds awfully good, even if that's just the theoretical max.

Re:probably only running on the central powercore

[cduffy]

Yeah, I bet IBM really care about Gimp/Blender

Why shouldn't they? It's not like they need to be under the deulusion that these tools are more important or popular than their commercial or popular -- but within their niche they area, and IBM has demonstrated a willingness to exchange money for street cred within this niche community (which, I think, has contributed handsomely to their revenues).

IBM sucks!

[IntergalacticWalrus]

They are ruining our "Yeah, but can it run Linux?" jokes by going right ahead and using it in their first demos!

Re:IBM sucks!

[iapetus]

Even worse - with so many SPEs, they're even making 'Imagine a Beowulf cluster...' comments redundant. And because it has two Cells, two sets of RAM etc, they've made snide remarks about dupes redunant too.

The only consolation is that with this new processor, Dragonball Z 'Perfect Cell' jokes will never get old.

Oh, wait...

Why do we need...

[el_womble]

I'm really excited by the Cell. I've seen a few posts so far, but they seem to be missing the point. This isn't about the move from single core to multicore, this is something quite different. I see it as the next logical step in this progression:

1. Integer Processors
2. Floating Point Processors
3. SIMD
4. Stream Processors

We're just adding a new basic functionality to the repertoire of genreal purpose CPUs. Its no more dual core than adding FPU, Altivec or BPU.

OK, not every program is going to benefit from having 8 DSPs in the core, just as most programs don't benefit from having 2 FPUs. But there will be core applications that take serious advantage, freeing up the integer unit for percieved performance enhancements.

I really hope OS X gets this. The Core APIs and Quartz Engine could be modified to take advantage, with end developers not even having to know or care.

Re:Why do we need...

[cnettel]

Well, most CPUs will schedule the instructions to the floating point units on their own. That means it's not a very hard problem to write a compiler that makes decent use of it. The normal memory management system is also used, with virtual memory, automatic transfer of data to and from cache lines, and so on. The cell vector units seem to operate on their own, with their own cache which has to be handled manually.

It's non-trivial enough to auto-SIMD code. I would say that this is somewhere between using SIMD properly and offload more stuff to the GPU, if you try to estimate the inherent difficulty. For example, is Quartz employing vertex shaders for text rendering? It could be done, if done properly it could even be faster than current text rendering, but I doubt that it's been done (yet).

Re:Why do we need...

[microbrew_nj]

More Buzzwords: Reconfigurable computing. System on a chip.

May I suggest a number five to your list? A single chip with microprocessor, DSPs and FPGA.

Not entirely hogwash

[Per Abrahamsen]

I do see it as a promising sign that IBM is actually basing a compute server on the chip.

Cell-based ideal compute servers...

[FellowConspirator]

I don't know why people pan these things as servers. Are people not aware that there's more to contemporary computing that HTTP daemons and database transactions?

I work in the biotech industry and we use computer farms and grids for all sorts of computationally intensive tasks: biopolymer sequence alignments, docking simulations, protein modeling, high-throughout 3D mass spectral analysis, etc.

A server with cell-blades and some minor tweaks to our software would generate a tremendous "bang-for-the-buck".

TEAMXBOX = LOL

[Anonymous Coward]

Please don't make me laugh, bub.

Hollywood CG studios

[galdur]

Combine Linux and Cell rack clusters and I can see some awesome possibilities here for animated and CG movie making.

The Cell's units may not be the best at serving up db data and files but applying it to frame processing could yield some pretty good performance.

Re:Hollywood CG studios

[Timberwolf0122]

Would that not depend on the DB architecture, a distributed DB taking advantage of multiple multi-core cell could extract data at a fantastic speed if the data was evenly split accross the CPU 'domanins' (ie 10 processors each with 10% of the data and a hard drive).
Each CELL 'work unit' could return party of the solution befor finaly combining the result on which ever unit is the least busy.

Re:Hollywood CG studios

[galdur]

My point is that with an equivalent die size a multicore CPU would perform better. Each of the eight SPE is a vector processor, not a fully-fledged processor, so that's why I think graphics-orientated distributed processing would gain more from the Cell than many other uses.

I'd be curious though to know whether distributed genome sequencing projects for example could put those SPE's to good use.

Security?

[argent]

I really hope OS X gets this. The Core APIs and Quartz Engine could be modified to take advantage, with end developers not even having to know or care.

I could see these units being used through abstract APIs like OpenGL, or used by WindowServer for compositing and rendering, but much of the advantage of the Cell would be wasted if it wasn't possible for applications to set up and use Streams... but the SPUs don't operate through VM and have no MMU, so if you let applications use these processors directly, how would you prevent an application from programming them to hijack or modify data that another processor is operating on?

Wait a minute

[Alkonaut]

The prototype, called the Cell Processor Based Blade Server...

So they made a cell processor based blade server prototype called the cell processor based blade server? The department of reduncancy department may want to comment on this.

Freescale...

[argent]

Would *that* shut up the "Apple has to switch to Intel to have faster cooler laptop chips!!! or they're D000000Med!!!!! " crowd?

I'd be happy with an e600 dual-core G4 with a 766 MHZ memory bus. What's taking them so long?

Re:Freescale...

[Anonymous Coward]

>I'd be happy with an e600 dual-core G4 with a 766 MHZ memory bus. What's taking them so long?

Should be right long the corner. Why? Because I bought a 12" PowerBook last month, that's why...

Obligatory Comment..

[Anonymous Coward]

I want one.. no! I want two! Hell, give me a bunch of 'em!

Ob. Simpsons

[Anonymous Coward]

Cells - is there anything they can't do?

Re:Very promising technology= investment opportuni

[Anonymous Coward]

well, to add nitpicky upon nitpicky, Blatchford does say:

Parallel programming is usually complex but in this case the OS will look at the resources it has and distribute tasks accordingly, this process does not involve re-programming.

Re:Very promising technology= investment opportuni

[MrVictor]

It's only fair to look at Blachford's rebuttal to Hannibals critique. http://www.blachford.info/computer/Cells/Rebuttal. html

Old News

[modular_forms_boy]

The patches were being sent to the Linux Kernel Mailing List a month ago for integrating kernel support for the Cell Processor. Just check the LKML Post from IBM-Deutschland employee Arnd Bergmann.

IBM doesn't tend to release code to the public until it's been through a long approval process ;-)

Re:Very promising technology= investment opportuni

[Slashcrap]

It's only fair to look at Blachford's rebuttal to Hannibals critique. http://www.blachford.info/computer/Cells/Rebuttal. html

Then it's only fair to keep in mind while reading his rebuttal that Hannibal has been writing CPU Tech articles for the last 5 years and is extremely well respected.

Blachford is an unemployed attention seeking non-entity with kooky ideas and the only respect he gets is from people equally eager to swallow the Cell hype.

DATED IN JANUARY!

[Anonymous Coward]

This was long before the Cell details were completely known. I'm sure once Hannibal actually sees what it can do, he'll change his tune.

Yawn..

[ad0gg]

How did altivec revolutionize people's everydays task(web browse, word processing)? I bet I can ask 90% of all apple users and intel users what altivec and mmx is respectively and they wouldn't know. 3dnow from AMD is 100x times better than mmx(ssex) instrustion sets, its benchmarks the same as altivec yet its not something gamers, regular people and system administrators ever consider when making a purchase. Also this article is about servers, please tell me how a simd is going to help my database? Help my Web server? I'd rather have 3 cores than 8 simds on my servers anyday of the week.

Re:Yawn..

[gorim]

Altivec was actually used heavily in both the user interface/windowing system of the Mac as well is in most of the multi-media apps that came WITH the OS. Its usage was heavily promoted and supported by Apple to developers, thus the usage is heavily used in 3rd party multi-media apps.

You have an extremely limited concept of what a "server" is. I have worked at and seen places with thousands of server hosts; databases and webservers comprised extremely small percentages of those servers.

Have you heard of:
- Real time multimedia production ? Or batch processing of multimedia for that matter...
- Numerical computation / simulation ? (for engineering / sciences)

MIPS are better than MFLOPS

[rbanffy]

Dunno about you, but I couldn't care less about the vector FP performance of my servers...

Putting Cell on them seems dumb for the kind of things we do on them. The 3 core PPC inside the XBox 360 seems a lot smarter choice.

For rendering farms or numeric modeling tough, they look like a smart choice.

Other blatchford weirdness

[dmh20002]

also check out blatchford section on distributed processing. I have seen several media level articles blabbering on about massively distributed processing with the implication that the Cell processor automatically does this. I don't see anything in his diagrams that can't be done by any computer with a network.

• Cells can distribute data to other processors (can't any computer do that?)
• the local network is used, Cell doesn't care which one (ok, so it isn't the Cell that is doing the distributing after all)
• OS should take care of it without user intervention (so its the software, not the processor)

I guess if all the processors are the same, then they could carry the code along with the data to get things done. But still, any computer can do this. Since he mentions SETI@home, isn't that what it does and it doesn't care about what the nodes are? I suppose you could imagine some finer grained computation scheme based on his diagram, but pretty quickly the network overhead would wreck your throughput.

Re:powerpc is not jsut for games

[Anonymous Coward]

No indeed. PowerPC is for:
Datacom/Telecom equipment
Automobile engine and chassis management
Current server equipment
Embedded industrial control

Re:No.

[Lord Bitman]

well he did...

imagine...

[namekuseijin]

a bewoulf cluster of PS3s!!

Summary of Blanchford Article

[coopex]

The Cell will not just be the most powerful desktop by far, it also saves you money on car insurance, improves gas milage, refinances your mortgage, never needs sharpening, cures cancer, creates world peace, and if you call within the next 20 minutes, we'll throw in the slashdot insta-girlfriend accessory kit*, free of charge.

*kit contains: 1 blow up doll (natalie portman naked and petrified), 1 bottle lube.

What about the X server?

[Rob Y.]

Would it be possible to implement a fast 3D X server using the extra processors? In software, so that 2D graphics cards could be used for 3D with a standard framebuffer-based driver that does all the 3D stuff in these cell thingys?

Re:Very promising technology= investment opportuni

[jdb8167]

The Cells are essentially identical so once the code has been parallelized the tasks that the code runs can be dispatched to any available Cell as it becomes available. The OS can do this automatically. This has nothing to do with how the code becomes parallelized but just how it is run.

Natural language queries

[fbonnet]

The Cell combines a general purpose PPC-based RISC core with a powerful parallel architecture. Both are complementary : the PPC is well suited and is a well-known solution for DBMS, but can't beat a parallel computer at scientific applications such as speech recognition, natural language, fuzzy logic or any AI. On the other hand, a parallel computer is less suitable for general purpose computing than a RISC CPU such as the PPC.

The killer app for a Cell-based server would be a DBMS with natural language queries, or image queries and the like. I remember IBM developed a query system for images some while ago, maybe they have a grand scheme for world domination.

YOU ALL HAVE WAY 2 FUCKING MUCH TIME ON YOUR HANDS

[Anonymous Coward]

I liked the part when Gohan and Goku were training in the time-dilation room; they ate lots of hot grits and after getting dirty they would give a Yaoi-ohhh performance in the pool.

Re:Very promising technology= investment opportuni

[loose_cannon_gamer]

Except it doesn't run x86 programs. So nobody will buy one. This is the latest in a very long line of superior architecture chips that nobody will buy for desktop processing. So relax a bit.