DARPA enlists NVIDIA to build exascale supercomputer that's '1000x faster' than today's quickest
By Darren Murph 
posted August 11th 2010 5:38AM
posted August 11th 2010 5:38AM
At this point, it's pretty obvious that GPUs will soon be playing a huge role in modern day supercomputers -- a role that may just rival that of the tried-and-true CPU. Virginia Tech is gleefully accepting $2 million in order to build a GPU and CPU-enabled HokieSpeed supercomputer, and today DARPA is handing out $25 million to NVIDIA in order to develop "high-performance GPU computing systems." Specifically the Defense Department's research and development arm is aiming to address a so-called "crisis in computing," and if all goes well, the four-year project will eventually yield a "new class of exascale supercomputers which will be 1,000-times more powerful than today's fastest supercomputers." That's a pretty lofty goal, but NVIDIA will be aided by Cray, Oak Ridge National Laboratory and a half-dozen US universities along the way. And yeah, if ever anyone's ego was prepared to topple Moore's Law, it'd be this guy. NVIDIA-Led Team Receives $25 Million Contract From DARPA to Develop High-Performance GPU Computing Systems
NVIDIA Team Includes Cray, Oak Ridge National Labs, Six Top U.S. Universities
SANTA CLARA, CA -- (Marketwire) -- 08/09/2010 -- A team led by NVIDIA has been awarded a research grant of $25 million by the Defense Advanced Research Projects Agency (DARPA), the U.S. Defense Department's research and development arm, to address what the agency calls a "crisis in computing."
The four-year research contract, awarded under DARPA's Ubiquitous High Performance Computing (UHPC ) program, covers work to develop GPU technologies required to build the new class of exascale supercomputers which will be 1,000-times more powerful than today's fastest supercomputers.
The team -- which also includes Cray Inc., Oak Ridge National Laboratory and six top U.S. universities -- is being funded by DARPA to address the challenge that conventional computing architectures are reaching the practical limits of energy usage and will not meet the challenges of exascale computing. The research team plans to develop new software and hardware technology to dramatically increase computing performance, programmability and reliability.
"This recognizes NVIDIA's substantial investments in the field of parallel processing and highlights GPU Computing's position as one of the most promising paths to exascale computing," said Bill Dally, NVIDIA's chief scientist and senior vice president of research, and the team's principal investigator. "We look forward to collaborating to develop programmable, scalable systems that operate in tight power budgets and deliver increases in performances that are many orders of magnitude above today's systems."
"The DARPA UHPC program is attacking technical issues that are key to the future of high performance computing, from the embedded terascale to the exascale," said Steve Scott, Cray's senior vice president and CTO, and the Cray principal investigator on the team. "We are excited to be working with this team, and we believe the directions we are pursuing will lead to radical improvements to the state-of-the-art in the coming decade."
In addition to the NVIDIA-led team, DARPA awarded contracts to three other teams to study UHPC systems. Prototype systems are expected to be completed by 2018.
The names of those universities on the NVIDIA team will be available once details with them have been finalized. For more information on the DARPA UHPC program, please go here.
Wow
@finishhim27 Apple picked a bad time to move away from Nvidia. The future of HPC is with Nvidia at the helm. ATI is busy sleeping in the shed.
@MySchizoBuddy, If you mean a future in the power usage and the heat department then yes. Other than that Fusion and Sandy Bridge will lead the way in the future. Nvidia is pretty much screwed.
@finishhim27 Yea thats what I thought too. Because, when they wrote "crisis in computing" they surely meant to write Crysis in computing.
Which leads to the inevitable question:
Does it play Crysis?
@MySchizoBuddy when has apple ever been interested in performance? they are more interested in sleek lines and eye candy
@finishhim27
You'd think they would ask the company that has the most powerful graphics chips on the market. But ok
@Drybones5 They did.
when talking GPUs, isn't it spelt crysis?
Will this help in the war in Afghanistan?
@rgz175
if by helping you mean faster computing it's FAIL outcome, then yes, it will help.
@charlietheunicorn
*sigh* Theres one in every article.
@Special Agent Steve
Well excuse me if I want to see my tax dollars put to better use of the general public. Much though I enjoy DARPA's creations, I want to see the USA's debt to decrease and the money to be spent on.... I dont know... creating jobs?
@rgz175 and of course, commenting on engadget will help the government hear your noble views.
@rgz175
nVidia doesn't employ people?
@rgz175 i agree. we should ban all R&D in the US for the foreseeable future so we can concentrate on putting our men and women in jobs.
We are spending millions of dollars in scholarship for higher education. What good is that for when there are no new jobs. We should immediately stop the education system untill we can guarantee new graduates a job.
We should also stop import of these rubbish gizmos that engadget reader like to buy. We don't want to send our Dollars to foreign countries like China and increase our debts. If china doesn't help use in creating new jobs then to hell with their products. we don't want them.
hmm lets see what else we stop. any takers
/s
@rgz175
Yes, it will play a lot of tic-tac-toe and somehow relate it to our current situation.
Moore's law can still be observed if it takes more than ~10 years :)
It's still nice to remember and laugh at Larrabee.
@Kurian
haha yeah
It still won't be as powerful as the WOPR.
;)
@Meekermoloko How about a nice game of Chess?
"will be 1,000-times more powerful than today's fastest supercomputers."
DARPA == Skynet
with current records in the dozens of teraflops -range, shouldn't 1000-times faster be petascale, not exascale?
or do they mean something other than flops?
@mrqs
Today's supercomputers are already in the petaflops range.
http://www.top500.org/list/2010/06/100
@zomgbbqftw
oh - my peek at wikipedia seems to have been too hasty
What is the need in this? Yes, of course it will be made 'because we can', but what actually needs that much processing power?
I doubt counting up Prime Numbers is why they need this.
Serious question.
@Alex R:
I'm thinking the war of the future is fought in a computer-simulated virtual world that mimics the real world. So you can simulate the risk of waging war and the computer returns the probability of success (or failure) all without ever firing a bullet.
@Alex R
ever heard of boinc or any of the @Home projects?
try googling it
@Alex R Well, some kind of away to analyze all data that flows through key places in the internet infrastructure perhaps? there are lots of applications that could be accomplished with more computer power, like trying to crack encryped data, or just processing data in an intelligent manner.
Other than that, anything from analyzing information from space to look for intelligent life, research genes or help predict the weather...
They should just connect it to f@h, complete everything they will ever need to fold, and then do seti@home, and then keep moving down the line.
@Sibuna
Neither of those projects will ever be "finished", no matter how fast you do it.
It's a great relief to know that men like this, through their sound logic and profound wisdom, will lead us all into the... future... and... stuff.
Countdown to nuclear meltdown starting now.
Doubt it will be of much use for anything other then a select set of computations. Most major scientific simulations don't scale much faster with GPU computing.
Today's simulations are no longer about just inverting matrices. There is quite a bit of logic and branch prediction involved. Unfortunately this means that stream computing paradigms don't apply all that well and while there may be small speedups, the money used on a $4000 graphic card is better spent on memory or a faster cpu.
Ah, so that's the can of whoop-ass Nvidia's CEO was talking about!
1000 times faster at 1000000 times the power requirements...
^_^
@Stormstrike
GPUs use less energy per FLOP than CPUs
@nate345
Not nVidia GPU's:P
There's never a shortage of money to find better ways to kill people and control the masses.
GPU Supercomputer eh?
A GPU is a processor specialized and dedicated in graphics processing. When it stops exclusively processing graphics and becomes the heart of a computer, it's okay to call it a CPU again.
@MrGreen
I imagine these GPUs would still be getting direction from CPUs. In this case, the CPU would be the lazy construction manager eating a burrito in the air conditioned truck telling the grunts to work harder.
Note to Darpa. save yourselves the grief, remember not to give it the launch codes!
@awlh
And how long would it take to hack into a defense system and unencrypt them?
@nate345
Get someone to memorise it.
1 guess for what they want to do with this! Crack encrypted information faster......Its the only real thing they could want it for!
@guigsy
Well, in the Navy at least, they just changed it so that leave requests can only be done on computer, and you don't want to hold up that process!
In heat and power per gigaflop, Nvidia has the best computing platform around.