Transcript
0:07 let me move now to GPU Computing we've always felt that heterogeneous Computing parallel Computing could really revolutionize Computing because of its characteristics you could use the CPU for single-threaded performance where single-threaded code is you could use a parallel processor for massively parallel complicated problems and so there's all kinds of applications for such a thing it would be higher performance the speed up would be phenomenal and it would be the most
0:39 efficient way the most energy efficient way of doing computation the challenge for us was how do we take a brand new Computing model you know we're familiar with using C compilers and highle programm constructs and we run it on basically the microprocessor architecture of today how do we take the brand new architecture and bring it to the world it's a bit of a chicken or egg problem on the one hand if there are no applications that
1:12 are heterogenous why would somebody buy a computer with a parallel processor inside and if there are no computers with parallel processors inside how do you create the install base to inspire software programmers to go create software for such an architecture and so that's the classic chicken or8 problem 10 years ago we came to a realization that maybe the GPU is the perfect platform the perfect vehicle to bring this new architecture to the marketplace
1:42 and the reason for that is several folds number one computer Graphics is inherently programmable it's inherently parallel if we could extend the architecture so that it could be more general purpose Not only would we be able to expand its ability to program and run applications from all different fields we could also push computer Graphics to the next level from programmable shading to now computational Graphics the two examples that I showed you this morning would not have been possible without
2:13 Cuda the second thing of course is that the GPU had a day job it didn't need any applications to go to market it already has a day job the day job is called computer graphics and so with that inspiration we decided to take Cuda to Market make every single one of our gpus Cuda compatible and look what happened GTC happened the impact to the industry is
2:45 really hard to even express and every year I give you guys a status report in 2008 we had 100 million gpus that are CED compatible 150,000 Cuda downloads 150,000 is a lot of different programmers downloading it and we had one supercomputer that was powered by Tesla 60 universities were teaching parallel programming courses using the GPU using Cuda and there were 4,000 academic papers this
3:16 year nearly half a billion processors Cuda Processor have been shipped you couldn't go anywhere around the world without getting hit by one if you wanted to program parallel if you wanted to program using Cuda you just had to run down to the local store 1.6 million times it's been downloaded the Cuda programming kit and now 50 supercomputers around the world
3:47 640 universities that's a lot of universities I really appreciate all of you evangelizing and teaching to Next Generation how to think in parallel how to program in parallel and of course the work that you guys do in research 37,000 papers the industry support has been phenomenal this is what it looked like at the important supercomputing conference in 2008 all the greens are
4:18 partners of ours or OEM or vendors who are somehow promoting parallel Computing somehow promoting Tesla this is what it looks like this year that was that's about a that's about 50 by the way this is what 2012 supercomputing looks like and it is almost completely completely penetrated it's very clear that we are close to the Tipping Point we're if we're not at the Tipping Point we're racing at it if we're not at a Tipping
4:49 Point we're racing at it and in fact you could kind of see from the top top 500 supercomputers these are early adopters of Technologies but these are informed adopters they're building these computers for scientific applications for scientific research they're building these computers for real work they're building these computers so that they can Inspire science that could save lives this represents about 20% of the
5:20 overall of the total top 500 Computing horsepower this year we represented 20% and so it's going to be really interesting to see what happens next year out out of that includes the world's most powerful the world's most advanced super computer the oakd Titan the oakd Titan has been in operation now for some time and I'm pleased to say that not only is it the highest theoretical performance supercomputer it also just
5:51 recently did the world's largest solids mechanical simulation at so stained of 10 pedop flops 40 million processors 40 million Cuda processors I can't even fathom that 40 million Cuda processors came together to solve a singular problem and deliver 10 pedop flops of sustained performance which is five times larger five times faster five times however you think more complex than anything achieved up to
6:24 that time this week the superc Computing Center announced that they are also going to adopt the Tesla GPU and build you Europe's fastest GPU supercomputer they're going to be using it for weather simulation so that they can predict weather very accurately a day to up to a week in advance getting people out of Harm's Way whole bunch of other research is going to be done on this supercomputer
6:55 called pisant pisan I think is the tallest mountain in Switzerland so now we have major supercomputing centers and large supercomputing installations in Japan in China in the United States in Europe and many others really really exciting to see that we are racing at the Tipping Point if not at the Tipping Point but nothing is more exciting really than frankly the
7:27 research that has been done you guys are all here to Showcase your work to learn from other people's work there's some just some amazing signs being done here I didn't know that the human genome wasn't just a double helix apparently that's only the short strands tomorrow's keynote is going to be really interesting apparently that our genetic code could be up to 2 meters long and somehow this 2 meter code
8:01 is somehow compressed and condensed in a three-dimensional way into a nucleus of a cell a few microns wide and somehow it doesn't knot up and somehow depending on the way that it's it's folded the same genetic code will store different information so that it could be a liver cell or it could be a something else cell
8:31 amazing Insight the scientists at Baylor and rice are going to talk to you about how they discovered this and how they used the GPU to decode their Discovery 50 gigapixel camera my camera is 4 megapixel 50 gigapixel camera made up of a whole bunch of offthe shelf sensors 50 g AP pixels so many pixels that they needed gpus in
9:03 the back end to do the image processing and the filtering and the operations on it but imagine the possibilities it's just a large array of lowcost cameras if we could do the image processing fast enough this super resolution image could be used for digital zooming refocusing it'll be taken in 3D of course look at the size of that camera
9:34 imagine the benefits in sports events liveaction events really fantastic work they needed a GPU to achieve what they achieved the work done at at Stanford folding at home understanding how proteins misfold they used the spare cycles of millions of Gamers around the world millions of Gamers around the world donate their spare Cycles up to Stamford
10:05 so that they could run this program to understand better how proteins misf this year they identified a canidate drug for Alzheimer's off of the spare cycles of millions and millions of Gamers around the world really fantastic stuff you guys are going to see some really great work I just want to I I want to name off a few of them one of these days we're going to make a a smaller stage this is really amazing
10:36 stuff well first of all the companies that are here Audi's here Apple's here ESPN Exxon Mobile Harley-Davidson Lego NASDAQ Nike SpaceX both of Elon musk's companies are here Elon SpaceX and Tesla Motors are are both here Walt Disney's here Pixar's here Netflix is here NASA's here iroot is here Goldman Sachs is here E Trade is here Chevron is here Amazon's here this is just a small list of all
11:08 the companies that are here all kinds of papers 400 different sessions I was going through the the the program guide last night it's just incredible this is Mecca this is Mecca of scientific discovery and this year we have manufacturing real well represented we have image processing all over the place I'm going to talk a little bit more about that later image processing is absolutely all over the
11:40 place there's some strange ones too I like this one this is a Cuda for a dating site they're using it for somehow matching the compatibility of whatever and the website you'll be happy to to learn is called fish.com I don't know why so somehow they have a cluster of 48 gpus they're
12:10 running 247 looking for a date there's a another paper that's really interesting GPU accelerated diamond cutting okay let's not waste any diamonds shall we so apparently it makes it brighter it's a more accurate cut and none goes to waste