Intel Corp (INTC)

[chipguy]

Maybe but the tide of history in computing is going against it. HPC
has been moving away from special purpose, low volume silicon
for quite a while. Some of the recent top500 machines have padded
their flop numbers by complementing their Xeons with GPUs but
the reality is real HPC work is a lot harder than simply cranking out
Linpack numbers. Folks who customize apps for these machines
are finding out after a lot of hard work the best results are usually
obtained by ignoring the GPUs and running apps just on the Xeons.
Given the politics of funding /dick size contest nature of the top500
list the use of GPUs in top end supers won't end any time soon but
the reality that it is hard to gain much or any extra performance by
adding Nvidia GPUs for a lot of HPC problem classes is becoming
more widely known. IMO Intel has a better strategy for HPC with
its Larrabee platform of x86 based CMP chips optimized for vector
processing but with a general purpose, coherent cache hierarchy.
Again here Intel has the expertise to pull it off. It is likely far, far
easier to adapt HPC apps to that sort of platform than to a mish
mash of Xeons and GPUs.