Of NVIDIA and hybrid computing
Making hay until Larrabee shines?
At SC09, it was hard to travel the floor without running into hybrid computing in some way, shape, or form. In fact, you couldn’t swing a cat without hitting someone or something related to using accelerators to maximize performance. (I know this because I actually did bring in a cat to test this theory… an actual live cat… well, it was alive at the beginning of the testing.)
The chief beneficiary of all of this interest in hybrid computing looks to be NVIDIA who, with its Tesla line of GPU accelerators, is currently offering the most robust product line. In a lot of ways, NVIDIA pioneered the trend and, as a result, has captured mindshare with at least the supercomputing set. There have been many challenges along the way.
The first hurdle was to prove the value of GPU computing, and the second was to put together enough of an ecosystem so that developers could take advantage of GPUs. NVIDIA has succeeded with both to the point that almost all of the system vendors at the show – and an equally large number of ISVs – had their own “NVIDIA + Us = Massive Performance” story.
NVIDIA is releasing its new Fermi line of GPU products sometime in 2Q10. Fermi fixes a lot of the knocks against GPU computing by providing ECC memory error protection and much-improved double precision floating point performance. It can also address up to 1TB of memory with its 512 cores per chip. There is also much to talk about on the software front with CUDA (NVIDIA’s GPU coding environment) supporting C++ and Fortran during the roughly the same time frame.
Our pal Timothy Prickett Morgan has written extensively about Fermi here. The road ahead for NVIDIA seems to be paved with gold, but stuffing its pockets won’t be as easy as it looks. Though the company shrugs off competition from AMD, it will have to deal with Intel and its Larrabee merged CPU/GPU design.
Larrabee has seen delays and stirred rumors of trouble, but Intel is not the type of company that will just cede control of anything that comes close to its turf. But for now, NVIDIA has the field to itself, and this should continue throughout 2010 and maybe into 2011.
This gives NVIDIA even more time to solidify the software side of its story, drive more sales, and entice more developers into the CUDA environment.
We’ve recorded a short video of the NVIDIA SC09 booth and a chat we had with Andy Keane, the head of the Tesla unit. Click here to check it out.
Sponsored: RAID: End of an era?