backtop

Hardware NVIDIA Launches Tesla Supercomputer
Shane McGlaun (Blog) - November 19, 2008 12:30 PM
Print E-mail del.icio.us 63 comment(s) - last by spread.. on Nov 28 at 3:40 PM

NVIDIA Tesla Personal Supercomputer  (Source: NVIDIA)
Tesla Supercomputer uses up to 960 parallel processing cores

When talking about supercomputers, the images that come to mind are those of massive computer systems with hundreds or thousands of individual CPUs taking up huge rooms and requiring millions to build and operate. Those mental images are correct for most supercomputers.

NVIDIA is looking to change the way people thing about supercomputing with the introduction of its new Tesla Personal Supercomputer. NVIDIA claims that its GPU-based Tesla Supercomputer is capable of delivering the computing power of a cluster at 1/100th of the price.

Tesla Supercomputers can be purchased for under $10,000 and offer 250 times the processing power of normal desktop workstations. Despite the massive processing power offered by the Tesla system it is housed in a standard desktop form factor system.

Microsoft Technical Fellow, Burton smith said in a statement, "We’ve all heard ‘desktop supercomputer’ claims in the past, but this time it’s for real. NVIDIA and its partners will be delivering outstanding performance and broad applicability to the mainstream marketplace. Heterogeneous computing, where GPUs work in tandem with CPUs, is what makes such a breakthrough possible."

The Tesla Supercomputer is based on NVIDIA's CUDA architecture allowing the system to be programmed in the C language. Up to 960 parallel processing cores can be placed inside the system. NVIDIA claims that Tesla Supercomputers are in use in major research environments like MIT, Cambridge and others.

Antonio Julio from Dell said in a statement, "Dell has led the workstation category for almost a decade and GPU computing represents a massive leap forward in performance that will bring supercomputer power to the masses,” said Antonio Julio, director, Dell Product Group. “The Dell Precision R5400 and T7400 will allow the scientific community to harness the capabilities of the NVIDIA Tesla C1060 GPU with up to two teraflops of computational power.”

NVIDIA announced its Tesla Personal Supercomputer days after the DOE announced its Jaguar supercomputer had reached record setting levels of performance.

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
By Creig on 11/19/2008 2:27:29 PM , Rating: 3
This "supercomputer" is basically just a desktop PC with four Tesla C1060 computing processors (ie - GTX280) in quad-SLI.

http://www.dvhardware.net/article27987.html

The only difference between the Tesla C1060 Computing Processor and a regular GTX280 is that the C1060 has 4gb of memory while the GTX280 has 896mb.

Does this mean that a computer containing a pair of ATI 4870X2 in Crossfire should be called a "supercomputer" too simply because it has four RV770 cores that each does 1.4 TFLOPs? I mean, this would be an even faster "supercomputer" compared to Nvidia's four GT200 cores that put out 0.933 TFLOPs each.

I think the term "supercomputer" should be reserved for specialized high-end computers such as Crays & purpose-built weather analyzing computers and should not be used for desktop PCs.




By SlyNine on 11/19/2008 4:39:46 PM , Rating: 1
Yes but unlike the other supercomputers this one.. here it comes. Can play Crysis.


By Pavelyoung on 11/23/2008 8:26:40 PM , Rating: 2
No it can't, because once you configure the system to use the GPUs for graphical purposes, then its simply a desktop PC again.


By TennesseeTony on 11/19/2008 7:25:50 PM , Rating: 1
It was not until 1997 that the first 'supercomputer' was able to break the 1 teraFLOPS level of performance. It filled not only 2500 square feet of floor space, it covered it from the floor to the ceiling, with 110 cabinets/server racks.

Now that same level of power is available in a standard sized case. If it is NOT a supercomputer, than neither was the one described below, nor any of the Cray's before it.

I don't expect you to understand, because your juvenile statement reflects the mentality of one conceived and born AFTER the first teraflop machine was put into service. Now finish your homework, brush your teeth, and get ready for bed, 8pm is right around the corner, and you've got school tomorrow.

From Wikipedia: (http://en.wikipedia.org/wiki/ASCI_Red)

"It was a mesh-based (38 X 32 X 2) MIMD massively-parallel processing machine initially consisting of 4,510 compute nodes, 1212 gigabytes of total distributed memory and 12.5 terabytes of disk storage. The original incarnation of this machine used Intel Pentium Pro processors, each clocked at 200 MHz. These were later upgraded to Pentium II OverDrive processors. The system was upgraded to a total of 9298 Pentium II OverDrive processors, each clocked at 333 MHz. It consisted of 104 cabinets, taking up about 2500 square feet (230 m²)."


By foolsgambit11 on 11/19/2008 8:04:23 PM , Rating: 3
That's nothing. Do you remember the movie "The Last Starfighter"? I remember watching a 'making of' special on that where they talk about all the incredible work they did with their Cray getting all of the graphics made for that movie. They had to wait hours to get short scenes back, find the errors, and re-run the programs. It took them forever to get stuff done that now can be done on-the-fly by my laptop's nVidia graphics chip. That's what's amazing. That was less than 25 years ago.

Ah, Wikipedia says it was a Cray X-MP - which peaks at about 400 MegaFlops. That's 5000 times slower than this 'supercomputer'. And it cost more than $10 million. How far we've come!


By Creig on 11/19/2008 9:22:28 PM , Rating: 5
Wow, is it your time of month or something? Loosen up a bit before you hurt yourself.

I've been in the IT industry for 17 years now and have been working with computers for even longer. The term "Supercomputer" is usually used to describe one of the fastest computers in the world AT THAT MOMENT . Yes, Nvidia's Tesla "Supercomputer" may be faster than the 1997 Intel TeraFLOP Supercomputer, but this isn't 1997, now is it?

http://www.answers.com/topic/supercomputer
quote:
Supercomputer - Any of a class of extremely powerful digital computers. The term is commonly applied to the fastest high-performance systems available at a given time; current personal computers are more powerful than the supercomputers of just a few years ago.

The term "Supercomputer" is relative to current computer technology, not obsolete systems.


By Storkme on 11/21/2008 7:09:44 AM , Rating: 2
Sorry gramps.


By Clauzii on 11/22/2008 12:17:58 PM , Rating: 2
ATI is on their way too:
http://ati.amd.com/technology/streamcomputing/prod...

Actually they are looking to be great contenders with about 240 GFlops Double Precision.

The nVidias are 'only' 78 GFlops DP:
http://www.nvidia.com/object/tesla_supercomputer_t...

The ATI card is set for 1Q, 2009.


supercomputer?
By Gul Westfale on 11/19/2008 1:07:46 PM , Rating: 1
really? when they say 960 cores they do mean cores, not chips. how would you fit 960 GPUs in a standard case?

so i'm guessing this is a simple desktop PC with a custom motherboard that allows for the use of quad-SLI... and that's about it.

here are two different pics directly from the nvidia website, showing machines built by partners. nvidia calls them "personal supercomputers":

http://www.nvidia.com/docs/IO/62722/TeslaPSC_image...
http://www.nvidia.com/docs/IO/62722/TeslaPSC_image...

velocity micro's website also confirms that these are just multiple-SLI PCs:
http://www.velocitymicro.com/wizard.php?iid=174




RE: supercomputer?
By FITCamaro on 11/19/2008 1:50:09 PM , Rating: 2
They're referring to the stream processors (shaders) in Nvidia GPUs.


RE: supercomputer?
By afkrotch on 11/19/2008 4:01:42 PM , Rating: 2
It's also possible that Nvidia might throw in specialized cards for it. Like a card with 6 GPUs on it and the ability to support 4-5 of these cards. Who knows, until someone gets their hands on this and opens it up for us.


RE: supercomputer?
By nafhan on 11/19/2008 2:20:49 PM , Rating: 5
Probably means it has 4 Tesla equivalents of the GTX280 since the GTX280 has 240 "stream processors".
240 x 4 = 960


RE: supercomputer?
By Gul Westfale on 11/19/2008 10:02:45 PM , Rating: 2
yes i know that, that is what i was pointing out. they are just normal computers.


Standard C or CUDA?
By dickeywang on 11/19/2008 1:02:34 PM , Rating: 3
I think there is a difference, isn't it?




RE: Standard C or CUDA?
By gramboh on 11/20/2008 12:12:00 PM , Rating: 2
Yep, you'd have to develop everything in CUDA, which from what I understand is not easy.


ATi
By FishTankX on 11/19/2008 5:13:22 PM , Rating: 2
Does ATi have a CUDA equivalent? I could see some really interesting setups with a bunch of 4870x2's all banded together with hypertransport links...




RE: ATi
By dsx724 on 11/19/2008 6:29:30 PM , Rating: 2
AMD Stream SDK? Brook+? CUDA isn't new or revolutionary. Brook+ came out long before CUDA. AMD hardware has slightly less performance on anything other than MADD though.


The Incredibles
By Subzero0000 on 11/20/2008 4:43:05 AM , Rating: 3
"I'll sell my inventions so that everyone can have powers. Everyone can be super! And when everyone's super... no one will be."




Competition with IBM?
By TheHarvester on 11/20/2008 1:14:16 AM , Rating: 2
I was just wondering if anybody has any idea what this means, if anything, for IBM and its line of Blue Gene computers. I'm not incredibly technical, but my understanding is that those computers do use cluster technology. Is it possible to carry over this non-cluster technology into some of the big-time supercomputers IBM puts out? Or is that entirely different technology?




Progress
By Jeff7181 on 11/20/2008 12:02:40 PM , Rating: 2
I can't believe anyone who reads the stories on this site would question the usefulness of a machine like this. True Tflop computing in what's basically a desktop form factor is a HUGE leap forward.

The question is not what can I do with this today , it's what can I do with this tomorrow ?




Glorified PC
By goz314 on 11/19/08, Rating: 0
Yeah but...
By FaceMaster on 11/19/08, Rating: -1
RE: Yeah but...
By jonmcc33 on 11/19/2008 12:36:20 PM , Rating: 5
Better yet, it runs Crysis, chews it up and spits it out as Pong. Then it yells in all known languages, "How do you like them apples?"


RE: Yeah but...
By Kyanzes on 11/19/2008 12:37:29 PM , Rating: 2
Yes. And you know what? Even through software renderer.


RE: Yeah but...
By FaceMaster on 11/19/08, Rating: -1
RE: Yeah but...
By Brazos on 11/19/2008 1:57:13 PM , Rating: 2
what about Folding @ Home?


RE: Yeah but...
By FITCamaro on 11/19/2008 1:18:40 PM , Rating: 5
I'd be more interested to know if it can achieve sentience, build a body for itself, go to your house, and rip off your hands so you can never type that again.

If it can, Nvidia you have a new order.


RE: Yeah but...
By pxavierperez on 11/19/2008 1:37:22 PM , Rating: 3
i concur.


RE: Yeah but...
By FaceMaster on 11/19/08, Rating: -1
RE: Yeah but...
By FITCamaro on 11/19/2008 4:24:48 PM , Rating: 2
I'm going to program it to rip off your hands after its done with him.


RE: Yeah but...
By FaceMaster on 11/19/2008 5:16:43 PM , Rating: 2
I wouldn't if I were you- don't know where my hands have been. Your Mum does, though


RE: Yeah but...
By Tsuwamono on 11/19/08, Rating: -1
RE: Yeah but...
By IWS on 11/20/2008 6:40:53 AM , Rating: 2
Ah, the old "everybody should speak American English" argument!


RE: Yeah but...
By FaceMaster on 11/20/2008 7:25:21 AM , Rating: 2
Next they'll be spelling colour with out a u.

Don't even get me started on the Z's.

Just because Firefox tells you something, IT DOESN'T MEAN IT'S RIGHT... though its depictions about yer Mam were fairly accurate.


RE: Yeah but...
By NubWobble on 11/20/2008 8:29:39 AM , Rating: 1
Correst spelling:
mUm
coloUr

The Brits invented the language, remember that.
Hereby ends the spelling lesson for today


RE: Yeah but...
By IWS on 11/20/2008 8:35:25 AM , Rating: 4
quote:
Correst spelling:


LMAO!!!


RE: Yeah but...
By SlyNine on 11/19/2008 4:41:58 PM , Rating: 4
America, Unreal, Crysis, Violence. All good things.


RE: Yeah but...
By bribud on 11/19/08, Rating: -1
RE: Yeah but...
By tdawg on 11/19/2008 1:50:12 PM , Rating: 1
Careful with that joke, it's an antique!


RE: Yeah but...
By FaceMaster on 11/20/2008 7:28:33 AM , Rating: 1
So's your Mum but I'm not being careful with it.


RE: Yeah but...
By FITCamaro on 11/20/2008 7:36:12 AM , Rating: 2
Seriously? "Your mom" jokes? Are you like 12?


RE: Yeah but...
By FaceMaster on 11/20/2008 8:23:01 AM , Rating: 2
At least I'm not fantasising about computers coming to life and ripping people apart.


How good is it?
By spread on 11/19/08, Rating: -1
RE: How good is it?
By monitorjbl on 11/19/2008 1:04:34 PM , Rating: 5
I think you were a touch misinformed. Traditional GPUs are not necessarily any better or worse at math than a traditional CPU. However, GPUs are designed to do a LOT (like, hundreds) of things at the same time while CPUs are (generally) constructed to do a handful of things really fast. If you ever seen some of the diagrams of the datapaths of a GPU and a CPU, you'd probably see what I'm talking about.

They're optimized very differently, but since super computers do a lot of work each clock cycle, GPUs are better suited for most tasks in that field.


RE: How good is it?
By Mitch101 on 11/19/2008 2:07:38 PM , Rating: 3
Unless its a classroom of Little Johnny's. That kid is trouble.


RE: How good is it?
By geddarkstorm on 11/19/2008 1:12:23 PM , Rating: 5
That analogy isn't talking about accuracy, but speed.

A math professor could whip through a single problem rapidly and with ease. A single student would most likely go through a complex problem slower. But a whole classroom of students could go through many complex problems at once and arrive at many solutions before a single professor would have time to go through them all.


RE: How good is it?
By Goty on 11/19/2008 1:20:11 PM , Rating: 4
GPUs aren't any less accurate than CPUs (they both can both do FP24), but I think that comment was referring to the fact that a CPU can handle a couple of large, highly complex problems at a time whereas a GPU can handle hundreds of simple calculations simultaneously.


RE: How good is it?
By trisct on 11/19/2008 1:35:23 PM , Rating: 2
GPUs are usually optimized for single-precision 32 bit math, while CPUs can do IEEE standard 64 bit, and with vector instructions, up to 128 bits at a time for specialized calculations. The GPU does have processing units that handle double precision numbers, but not nearly as many as the less precise units. So, your analogy has merit, but note that there are one or two professors in the room too.
The overall effect is that the 2 Teraflops is talking about single-precision 32 bit arithmetic, which fits a large swath of simulation problems where overall behavior rather than fine accuracy is the goal. For problems that require more detailed calculation, you would have to lower the Tflop rating of the box to account for necessary simulation of higher precision arithmetic in software, to address the GPU precision limits. It would still be faster than most general purpose CPUs though.


RE: How good is it?
By Oregonian2 on 11/19/2008 6:50:36 PM , Rating: 2
AFAIK most supercomputers revolve around floating point processing. If the GPU is integer math (32 or 64 bit) the super processing power may be limited in application more than one might hope for.


RE: How good is it?
By finalfan on 11/19/08, Rating: 0
RE: How good is it?
By spread on 11/19/2008 4:53:31 PM , Rating: 2
This makes more sense now.

But how useful would this be with supercomputers? What if you run a complex problem. Wouldn' there be a performance hit from the 'dumb' GPUs?


RE: How good is it?
By SlyNine on 11/19/2008 5:01:09 PM , Rating: 3
I know the people at Stanford had a few pages on this for their Folding@home client, they have been using GPU's for scientific research for some time.

But in short, yes some things are better on a CPU cluster right now.


RE: How good is it?
By SlyNine on 11/19/2008 4:54:34 PM , Rating: 3
To think. C'mon, You're making it sound way more dramatic then what it is, Yes we all know CPUs can do more complex work in one cycle then GPU's. But since more scientific can be divided up in to many smaller tasks the GPU whoops the CPUs today.

As far as thinking. Well the most powerful thinking machine is the human brain, its important to note the brain is closer to a GPU then a CPU. I doubt anyone would give any software on the planet the credit of being able to "think".

As far as it not being accurate enough. Folding@home is using GPU's for actual research now. Since the GPU2 client came out they've seen it as a viable research method. I look too see the complexity of the models increase a lot.

Still there are definitely things better suited to a cluster of CPU's.


RE: How good is it?
By spread on 11/28/2008 3:40:33 PM , Rating: 1
-1.

Wow. Can't ask questions here otherwise you get modded down.


Ummm...what?
By Motoman on 11/19/08, Rating: -1
RE: Ummm...what?
By bandstand124 on 11/19/2008 1:05:28 PM , Rating: 5
R&D, Bioinformatics, Medical Imaging, Ray Tracing, Military, Intelligence, Meterology, Artificial Intelligence, Computer Aided Design, Astronomy, Physics, Mathematics ... the list is endless.

Any problems that may involve adding two numbers together essentially.

I don't think you are supposed to do anything with it since you are apparently incapable of any thought that isn't egocentric.


RE: Ummm...what?
By Motoman on 11/19/08, Rating: -1
RE: Ummm...what?
By Etsp on 11/19/2008 2:26:18 PM , Rating: 2
I don't think Video Rendering (ray tracing) counts as R&D.


RE: Ummm...what?
By Suntan on 11/19/2008 4:03:19 PM , Rating: 2
Don’t be a dink.

I would love to have access to a specialized computer at work that could crunch through complex CFD and FEA analysis faster than the current workstation I have now.

We don’t have a budget of a DOD or NOAA, but being able to setup and then compute an FEA run in 2 minutes, instead of 20 minutes, would significantly improve our ability to come up with better designs much faster.

Spending an entire work day to run through just 6 or 7 interation ideas means that there are still a lot of areas where a design is left significantly short of “fully optimized” when the design has to be frozen and preparation for production has to begin.

If it is cost competitive to have 4 or 5 of these available for a company consisting of 200 engineers, that is a significant market compared to building a true “supercomputer” that only 7 or 8 entities in the entire country are willing to pay for computing access time on.

Mainstream means “large quantity of customers.” It doesn’t mean “directly applicable to some turd still living with his parents and wanting to rip Blu Ray discs faster.”

-Suntan


RE: Ummm...what?
By Zirconium on 11/19/2008 1:06:57 PM , Rating: 3
quote:
...just exactly what are we supposed to do with this thing? Why does Nvidia think there is a need for anything like this outside of R&D kinds of applications?
Did you RTFA?
quote:
“The Dell Precision R5400 and T7400 will allow the scientific community to harness the capabilities of the NVIDIA Tesla C1060 GPU with up to two teraflops of computational power.”
This is targeted to industries which require tons of processing power.


RE: Ummm...what?
By Motoman on 11/19/2008 1:58:59 PM , Rating: 1
...did you RTFA?

quote:
broad applicability to the mainstream marketplace.


I acknowledged the scientific community implicitly in the reference to R&D. Now explain what applicability this has to the "mainstream marketplace."


RE: Ummm...what?
By geddarkstorm on 11/19/2008 1:08:45 PM , Rating: 2
Think broader.

Scientists, animation studios, and 3D rendering to name a few, would all benefit greatly from a computer with such raw computational power but a PC form factor. Seriously, doing protein structure calculations in our lab is an intensive process, and a computer like that would blow through it like it was nothing - greatly speeding the process and freeing us up for more projects. This is one awesome piece of hardware!


RE: Ummm...what?
By murphyslabrat on 11/19/2008 1:09:40 PM , Rating: 2
That's actually part of what makes it so amazing. That this is possible using primarily PC-compliant hardware, and in such a small (relatively speaking) form-factor, is ground-breaking.

That age-old question, "Can it play Crysis?" is actually valid...though no less annoying. *_^


"Let's face it, we're not changing the world. We're building a product that helps people buy more crap - and watch porn." -- Seagate CEO Bill Watkins











botimage
Copyright 2009 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki