Slashdot Mirror
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."
Now we'll have to put up with people's web servers ringing in movie theaters.
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.
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??
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!
"We demonstrated the prototype to show that Cell continues to mature..."
...I thought Gohan killed Cell?
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.
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.
Snowden and Manning are heroes.
"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.
In the beginning the universe was created. This made a lot of people very angry and is widely considered as a bad move.
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->
..what next?
.5? Any supercomputer geeks care to postulate?
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)
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
42.0
Got to use the floating point power.
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!
Try to hack my 31337 firewall!
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...
It is exciting...but (not to sound trollish)...I'll believe it when I see it.
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.
---
"Of course, that's just my opinion. I could be wrong." --Dennis Miller
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.
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?
Illegal? Samir, This is America.
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.
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.
One of these days I'm moving to Theory - everything works there
Maybe it's late, but am I the only one who thought he was saying that IBM had "fat, dickless, clients"?
My other car is a 1984 Nark Avenger.
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?
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?
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.
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.
IBM designed the Cell
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.
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.
Why would anyone engrave "Elbereth"?
The mind might indeed be a Turing machine, but it's a very different architecture and OS than the ones we know about.
"A language that doesn't affect the way you think about programming, is not worth knowing" - Alan Perlis
41.999987448205
Intel Inside (the warning, NOT the logo!~)
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
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.
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.)
well, let's be nitpicky too, then.
oh well. Anyway, I could go on, but I'll stop here. You get the idea.
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...
"I used to have that really cool,funny sig
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.
This space is intentionally staring blankly at you
They are ruining our "Yeah, but can it run Linux?" jokes by going right ahead and using it in their first demos!
My brain only thinks about 2 things. Either food or sex.
It must run a Reduced Instruction Set.
------
beware he who would deny you access to information, for in his mind he dreams himself your master
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".
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.
I am TheRaven on Soylent News
IBM doesn't tend to release code to the public until it's been through a long approval process ;-)