NVIDIA is known primarily for their PC graphics cards, but they’ve been in the news recently both for the expansion of their CUDA and PhysX initiatives as well as (allegedly) developing a CPU to challenge Intel and AMD. They’ve been proponents of parallel processing for quite a while, naturally, and have made some investments in companies like MotionDSP and Elemental Technologies, both of which are developing software that really leverages the GPU.
The success of these investments is difficult to measure (NVIDIA’s been hit as hard as the other semiconductor-related companies, losing 50% of its revenue), but it’s hard to argue with the fact that parallel computing is where practically the entire industry is heading. In light of this, NVIDIA has started what they’re calling the GPU Ventures Program, by which they hope to “identify, support and invest” in companies that are working on GPU-based computing applications. A ton of GPU-accelerated (or simply parallel-ready) applications can be found here at the Cuda website, including image manipulation, physics simulation, and advanced computation.
They’re hoping to draw some of the best startups and the minds within to NVIDIA as a sort of parallel processing father figure. A couple million here, maybe a peek at the new hardware there, and they’ve got themselves a pretty solid partner. I talked with Jeff Herbst, VP of business development there, who outlined the idea of it. It’s very much an investment thing and not some sort of “GPU App Store.” They’re hoping that VCs might like to get in early on what NVIDIA thinks is the next phase of computing. Personally, I was already convinced by the kind of interesting work being done by their partners already, but Mr. Herbst has been talking with many more companies and is planning on expanding their Emerging Companies Summit this fall.
The site is just launching today (any moment now…), and hopefully it’ll be populated with the kinds of hot young startups that tech investors love to just run their hands over.
Sometimes hard to tell if these corporate venture things are serious or just a bunch of marketing. I’ve got a serious parallel processing algorithm for multi-dimensional indexing/search running on Amazon EC2 cluster and is orders of magnitudes faster than existing quad-trees, no reason wouldn’t run on GPUs — crazy applicability to 3D — and would be perfect for these guys. Wonder how to get them to actually talk to me. We’ll see what the site says when it is up.
I use NVIDIA graphics card . i just love them. I wish success in new venture.
The market for GPU apps is very limited. The consumers have voted.
As a current student in a CUDA course taught by ex-NVIDIA professors and current NVIDIA engineers (special topics course at UIllinois), I can see both sides of this story.
On the one hand, you need a really substantial application requiring thousands of threads to have any shot at outperforming a CPU given the overhead required by going to the GPU for computation. Further, the programming model requires considerably more thought and skill to correctly leverage than a sequential one.
On the flip side, there are thousands of (non-mainstream) applications out there that could already benefit. And with a bit of a paradigm shift in the way people program, additional uses might be found in more common programming. At the very least, hopefully this initiative will enlighten people to the wonders massive parallelism can bring to huge problems (simulation, protein folding, etc.), and there’s nothing wrong with that…
i certainly would not mind being able to fold more proteins, I wish nvidia well in this endeavor
“At the very least, hopefully this initiative will enlighten people to the wonders massive parallelism can bring to huge problems”
I would love to see a cascading spin lock panic on millions of threads in a server side graphics system.
I would be laughing my ass off.
I wonder if that will happen at city space.
I wasn’t aware we’d had an election! Consumers are hardly even aware of the idea, much less able to pass judgment on its usefulness.
Where is the NVidia equivalent of this ???
http://www.google.com/search?q=fglrx+source
If you can’t optimize the GPU to render several frames per memory swap(back buffer), then how can you render say 1000 clients running a game on the server side?
OGL isn’t enough. You need the ko source to really get the best out of it. The drivers for these cards are made for client side rendering.
IBM cell may be an even more attractive solution for server side rendering then GPUs because you don’t have the bus or DMAs. Cell’s data busses are on chip == fast.
At any rate, I am knee deep in mobile. Good luck OTOY clones!!!
Plus all the frames have to be saved as images queued up into movie frames and encoded to .flv.
If I was going to do something like this, I would look at a Cell PU farm also as a possible solution for large scale processing.
i used nvidia graphic card too… but i wonder why if I play games like Warcraft and Special Force, there’s a time that the game will going to lag, the screen slowly moves… is there any reason why it’s gonna be?
I use NVIDIA cards too. My fan is really loud for some reason.. wait, what are those lines and blotches on the scre
The link to “GPU Ventures Program” is broken :(
it should be live now.
Great blog! It’s always so nice to be inspired
I have used Nvdia Graphics cards since the early days of the company. I still use to this day because of their great driver support. So i am looking forward to the new generation cards.
i am also going buliding a new 32 core 256-bit processor with 1 gb of case and 5000 mhz bus
in 2011 Join me
Nvidia GEforce 4-now unchanged architecture.
If they don’t do this then their going to have some big problems on their hands.