NVIDIA announces cost, energy-saving Tesla Personal Supercomputer
By Donald Melanson 
posted Nov 18th 2008 12:27PM
posted Nov 18th 2008 12:27PM
AMD has already outlined its plans to harness the power of its GPUs for some added computing muscle, and it looks like NVIDIA is now taking things one step further by announcing its new GPU-based Tesla Personal Supercomputer, which promises to deliver the power of a traditional supercomputer cluster at 1/100th of the price. That "personal supercomputer" is actually a platform based on NVIDIA's new Tesla C1060 GPU Computing Processor, which itself is based on NVIDIA's CUDA parallel computing architecture. The supercomputers themselves will come from a whole host of manufacturers that have already partnered with NVIDIA, including ASUS, Dell, Lenovo, and a number of more specialized computer makers. While complete details on those systems are still a bit light at the moment, they'll apparently be "priced like a conventional PC workstation," and the first few out of the gate should be available starting today.
At that cost, we should all get one and donate the downtime to folding@home .
Agreed.
Great minds think alike. This could be another revolutionary step for Folding like the PS3 client before it if the real thing lives up to the marketing.
Also like to donate some power to SETI@home http://setiathome.ssl.berkeley.edu/
Their gpu only does 1TFLOP, with the 4870 doing 1.4 TFLOPS. If you are just doing folding@home you might as well just get some 48xx ranging from 1/11the the price of the latest tesla for 1 TFLOP to 1/5th for 2.4 TFLOPS (more than 2x the performance using the 4870x2).
how long does it take to get to 0-60 TeraFLOPS?
3.6 seconds
Oh wait...
And I always thought that cost/energy-saving and Supercomputer don't really go together..
me either
Tesla... now that's a name that evokes a reputation of energy saving.
Considering he invented AC which was far more efficient than the DC it replaced in power lines, I would say, yes, he should.
RIP tesla
Did he "invent" AC or "discover" it?
...and the confusion ensues...
He discovered air conditioning?
... efficient over long runs, but not as efficient at actually doing things at the other end which is why everything transforms it to DC.
Tesla pioneered AC for power transmission, he didn't just "discover" it. The electricity found in nature, like lightning strikes and electric eels, is DC. The kind of regular, sine wave AC juice you get out of the wall has to be man-made. Tesla didn't invent the first instance of AC electricity, because he didn't make the first hand-cranked generator, but he refined it for use as a safe power source. It's also vastly more efficient to transfer AC over long distances.
I think you can say he "independently invented" it. Although you can have natural generators of what can be described as AC current, he had to build a generator to create an AC current of the specifications he wanted.. therefore he "invented" it --sort of.
For example, natural bridges have existed (tree falling across a river).. but are we going to say that the Brooklyn Bridge was discovered and not invented?
Technically, the bridge was designed and built... not invented. It was a modification of current designs, plans and experiences of other bridges. It would be like saying that Chevy invented the electric car because they built one that looks different than the rest.
I never knew the market for a super-cheap number crunching machine was so likely to be lucrative. I mean, are these machines good for all-around work or are they primarily marketed at those who will write custom software to model particular computationally intensive problems (for example optimization of the shape of a low-gain antenna for a satellite, etc)?
I would think any college worth their salt would want some of these, as opposed to their current space/energy consuming clusters.
These require custom software to take advantage of the hardware. But, for example, a lot of electronic design today is already being done on similar hardware. Your cell phone was probably designed on an NVIDIA gpu.
For running vast RPG worlds servers? Or, perhaps, a multi-user Transport Tycoon server? Where do I queue for a bunch of those?
So, if it requires custom software, is there any chance of a Linux port (for normal desktop usage)? Or can GPUs not be utilized for the same kinds of things as CPUs?
And to think I was gonna get the Thomas Edison-based Super Computer.
Great because everyone wants to recode all their files to work with CUDA. Make it able to run all my c based code including mpi calls as is and I'd buy several in a heartbeat. ( via grant $)
That should be mentioned in the article
If you could save millions of man hours, i think you would spend the hundreds of hours to recode.
Think of the energy savings alone.
@Joseph:
Sure and I'd consider it but I only want to go through that process once and I'd like to see who wins out CUDA or OpenCL before I take that plunge. (Plus not to sound like a dick but I can run my stuff on a Blue Gene right now so I'm in no particular rush)
@Joseph: yes, think of the millions of people who will have to calculate on paper for one hour :)
@ Joseph
You have the right idea, but in practice - even if you have the time to recode, which is a big if considering customer (sponsors for the universities) demands - recoding the types of things that run on current clusters is difficult, because they have been built up over years by numerous people who have scattered all over the place from where the project started. At work we have a Tesla "Deskside Supercomputer" breakout box that cost something like $5k that the IT guys claim is as powerful as one of our $100k+ AMD64-based 80+ core clusters. We write a lot of code in-house as well as modify third-party codes, so we thought we could tackle it pretty easily, but after setting the thing up it has done nothing but collect dust for 6 months. Until somebody comes up with compilers that that can take existing code and make good enough guesses at where to direct the CUDA calls that actually gain some performance, then it isn't economical to code for them. There are some stop-gaps out there - I'm personally looking at some situations for some of my code where a company has written an environment plugin that have unique variable type definitions for execution in parallel on the GPUs where a find & replace job might get me some significant gains - but I'm still too busy with things that work on the cluster that is already there to sit down and do it. Where these have the most utility is for new projects, which, for now anyway, makes it a niche market.
Im not expert at this but would this mean that all virus have to be rewritten to be able to run on this architecture? If so then sign me up! You are essentially killing 2 sticks with one stone. Am I right?
im pretty sure you dont have to recode stuff for CUDA. There was a lot of miscommunication with the writers of articles who claimed that CUDA required a modified version of C to run on their GPUs. However, Nvidia debunked that a long time ago saying CUDa is just a different c compiler and pretty much all c code written previously should work for it.
Sooo I can play games at a billion FPS with this?
Are you an idiot or something? The human eye can't detect above 999 million FPS.
Its a joke man sheesh. No wonder I don't post much on Engadget.
The reason you don't post on Engadget so much is that you lack the scientific knowledge to make a witty comment...
That's a little harsh.
Grow up guys, he was just joking.
no, that's over 9000.
Nothing goes over 9000.
Now that is Crysis- licious!
That doesn't even work.
did you watch this season of project runway?
I think I will bring sexy back w/:
What about a Beowulf cluster of these?
I was thinking of that right when CUDA first came out. Would be pretty epic I imagine.
They exist. And yes, they are shockingly fast. :-)
Also very hard to code for.
Will this be able to run Crysis 2?
Serious question, don't flame me, I really want to know...
Is this just a really powerful computer, capable of running intensive games and applications, or does everything need to be specially written for it?
Everything needs to be recoded in order to work with CUDA. I know, it sucks, I want a supercomputer too :(
Ah damn. I thought for a minute there it was gonna give everything a massive performance boost for the same cost.
Oh well, bit far fetched lol
Is anybody else scared about this? From hackers to terroists there are probably a lot people with not so good intentions who have always wanted an affordable super computer.
No. What are they gonna do with a faster computer that they can't do now?
@FireStar,
Uh I dunno, maybe decrypt harder encryptions, manage DDoS attacks better, run world-domination simulations faster, and overall have a stronger presence on the internet.
if hollywood has taught me anything - and i like to think it has - it is that computers blow shit up.
Available starting today??? And we still don't know anything about it?
Isn't is what Apple already... nah, just kidding :)
Am I the only one who saw this picture and thought of GlaDos saying something like "When you complete this chamber you will be dead, and then there will be cake."?
Well if it's a choice between cake and death, I'll take the cake.
http://www.youtube.com/watch?v=BNjcuZ-LiSY
yeah but you cant eat it too. Turns out it was poisonous.
Ok, so what the hell runs on it? What O/S? I know Windows or any basic *nix OS will run on it. Would be nice though if they did..
** will NOT run on it. Sorry,, typed too fast.
Windows and Linux are both supported. These machines are basically fast desktop workstations with oversized nvidia gpus that are optimized for computation rather than graphics.
+1 to the first person that makes a Tesla coil case mod for one of these.
What every computer needs, a static discharge rod next to delicate electronics.
And there I was going to ask if it could play Red Alert 3.
I'll get my coat.
ummm, anyone found/priced one of these yet?
What OS are they using?
nevermind, check this out:
http://en.community.dell.com/blogs/insideit/archive/2008/11/18/your-hidden-supercomputer.aspx
It's the applications that will need to be coded for the GPU's, not necessarily the OS' themselves...
I should have known that. duh.
Thanks for finding that, I was under the impression that CUDA was enabled through the drivers and that was it, I didn't know the application itself had to be coded for CUDA support.
Will it run XBMC or Crysis?
Can Engadget please prevent all posts with the word Crysis, Doom, blend and overlord in it? Apart from this one, obviously.
I would suggest to add some ..
e.g. : "copy of the iPhone's xxxxxxxxxx" or "some much like the iPhone's xxxxxx" or "the iPhone already does this"..
Especially when it comes from people who don't know what a MMS is and think photo sharing on mobile was invented by steve Jobs.
(I'm not totally sure you can actually track that but ask google, they know what color are my pants so ...)
Bah, in my humble opinion (as the success of netbooks demonstrated) there is already available all the computing power the average geek is ever gonna need; what's sorely lacking are the apps and programs to exploit and take advantage of that power.
Vista (for exemple) seems explicitely designed to waste as much computing power as possible.
This kind of machine could be useful only to heavy coders or for special professional needs unless some future boy-genius/soon-to-be-billionaire will manage to write TEH killer app for this toy.
Until then, betterment in citizens' computing experiences will keep marching in step with Moore's law inflexible rythm.
Quantum leaps are rarely seen in computing industry, and this does notseem to be one of those.
Well may you should start work on a super computer powered spell checking program. A grammar check probably wouldn't hurt while you're at it.
"Bah, in my humble opinion (as the success of netbooks demonstrated) there is already available all the computing power the average geek is ever gonna need"
And no one will ever need more than 640Kb RAM
@Kal326
Maybe you should need a supercomuter to figure out aproximately how idiot (and boring) you really are.
But probably such a task is byond the possibility even of the most advanced SC clusters.
Maybe now is the time to pitch the idea of an open source Dwarf Fortress server on one of these... mmm, massive kitten death towers.
You would think that after all the "will it play crysis?" issues you would realize that computing power usage will always be pushed by the human imagination. These computers can probably be used for things you cant fathom. In the present maybe most people cant. But its like cell phones. Did you ever imagine everything that cell phones would be capable of now? Were you asking "Why are we putting faster processors in phones if all we need to do is make phone calls?" back in the 90's... Or "Why put a camera on a phone if I already have a digital camera?" .
This is good news! Love it!
I expect this to be a MacPro Option in January. Nvidia and Apple are really good friends now. And don't forget Adobe and Nvidia.
Will it play Crysis on High Details? Or maybe Crysis 2 on Low Details?
So who wins?
Tesla, or Helmer ( http://helmer.sfe.se/ ) ?
Very interesting, can't wait to see full reviews of these.
What about using the GPUs as a renderfarm? Would the software have to be ported to this CUDA language first?