Nvidia announced some new CUDA stuff last week, a new developer kit (3.1) and the Parallel Nsight Visual Studio plug-in, both designed to make it easier for ISVs and other coding types to support Nvidia GPUs in their apps. Our pal TPM has a typically detailed story here.
One thing that jumped out from the introductory materials is the penetration that CUDA is achieving. According to Nvidia, there are 100,000 active CUDA developers, and there have been more than 600,000 CUDA development kit downloads.
But high download numbers don't necessarily mean anything. However, we did see quite a bit of pudding-eating-proofing activity at Supercomputing '09 last fall. Walking the show floor, you couldn't swing a cat without hitting a vendor talking about GPUs (almost always Nvidia-branded) and showcasing them in their booths.
Moreover, we also heard a lot of happy GPU chatter from real users. Pushing forward CUDA and the GPU ecosystem is vital for Nvidia's present and future in HPC and, going forward, the enterprise computing market.
The key to big success in accelerators isn't so much the bang for the buck with the accelerator itself but in how much pain, time, and money an organization needs to spend to get that bang on its workload. The more robust the ecosystem, the easier it is for customers to get to the accelerated Promised Land.