Cardinal Peak Partner Chad Scates writes:
I am in San Jose at the GPU Technology Conference, where I spent yesterday listening to some excellent talks about CUDA.
We’ve blogged before
about CUDA, which is one of a handful of emerging technologies that
allow applications to run compute-intensive tasks on the graphics
processing unit of the computer.
Of course, here at the show we are getting the sales pitch from
Nvidia, which isn’t entirely a bad thing, since it’s also educational.
Something that strikes me is that every card sold by Nvidia since 2006
is CUDA-capable, which means the likelihood of the average desktop
computer having a CUDA-capable card in it is becoming pretty high.
I spent 11 years working for a company that produced desktop
software for scientific analysis in the oil exploration industry. I was
very excited by the idea that every desktop computer now has the
ability to perform highly parallel computations in a very affordable
In his keynote, Nvidia CEO Jen-Hsun Huang showed some slides that
did what he called CEO math. This math showed performance enhancements
on code that was conducive to parallel development; the performance
gain from moving to CUDA was on the order of 100-200x. Wow—that’s two
orders of magnitude.
The drawback to CUDA is still the need for specialized code, but as
we’ve gotten more familiar with this technology, we can attest that—for
certain applications—there can be a pretty dramatic performance gain
from a relatively modest engineering investment.
Pretty exciting stuff.
Learn more about Cardinal Peak at cardinalpeak.com.