Nathan Brookwood, a consultant with the Gerson Lehman Group, has published a response/analysis to the news coming out of NVIDIA’s GTC last week.
While AMD and Nvidia battle for supremacy in the GPU computing arena, there’s one obvious loser, Intel. AMD’s 5870 appeared on schedule. Nvidia’s Fermi is late, but its GTX 280 series still is competitive. Intel’s Larrabee remains a no-show. End users who buy their systems by the teraflop have discovered and validated an alternative approach that requires fewer x86 CPUs, less power, and less space. GPU computing is here to stay, and the market will punish those who lack a competitive offering.
Later in his article, Brookwood seems to argue that AMD has the potential to capture some of the scientific computing market that is interested in GPUs, but then points out enough drawbacks that it seems to me that AMD has a lot of work to do before it has a prayer. For example, there is no C++ (less of a big deal) or Fortran (big deal) support, no ECC, and AMD’s double precision performance is lower than Fermi’s. Perhaps the biggest argument that AMD does not currently have a competitive offering in scientific computing on GPUs is that their current generation product has a DP performance of 500 GFLOPS to NVIDIA’s 80 GFLOPS, and AMD is still really not part of the conversation (the 5870 is too new to structure an argument on installed base).
And this is a key insight, I think. AMD’s GPUs are losing to NVIDIA’s at least in part because of the tools, the support environment, and the network effect that the size of NVIDIA’s GPU ecosystem is creating around it’s products. These are also things that Intel is really, really good at today, and their continuing acquisition of tools providers in the parallel programming space indicates they are only going to get better in the future. I agree with Brookwood that Intel is losing out today, but they clearly have the past track record and current capability to build Larrabee into a powerful force in the computing community when it debuts. And where NVIDIA had to build the infrastructure and ecosystem around an already existing offering as first mover in the space, Intel gets to do both at the same time with second mover advantage — they’ve already had the chance to learn from NVIDIA’s performance.
Intel is starting from behind, but they aren’t out yet. Right now the real absentee player in the GPU computing market looks like AMD.