NVidia's Tesla C870 Graphics Processing Unit (GPU) will be the basis for a "deskside supercomputer" add-on that will provide highly-parallel high performance computing (HPC) capabilities, presumably programmed with \<a href="http://developer.nvidia.com/object/cuda.html"" target="_blank" rel="noopener noreferrer">NVidia's CUDA toolkit.
Dedicated hardware for HPC has always been a treacherous market -- one year's darling is next year's has-been (people used to buy Cray Supercomputers at auction and resell them for the gold in the connectors. \<a href="http://findarticles.com/p/articles/mi_m0NEW/is_1993_April_16/ai_13786747"" target="_blank" rel="noopener noreferrer">True story.). Dedicated processing boards for desktop computers have always been especially troubled, as the system bus is such a bottleneck and Moore's Law used to provide such wonderful free lunches. (No longer true, although the bus issue is potentially more dramatic than ever.)
There is infinite demand for HPC from 3 well-funded sectors: economics (trading), bioinformatics, and chemistry (bio- and otherwise). These sectors will absorb any amount of information processing capacity available. Whether that can be translated into commercial success for NVidia, or whether they unlock additional markets, is far less certain.
I wonder if Google will buy a couple boards.
Takeaway for programmers: Feverish hardware activity relating to concurrency continues. Software lags, with only relatively low-level toolkits available for exploiting the system. Keep your C skills sharp.