A recent article by Joseph Desposito in Electronic Design entitled "Is Your Personal Computer A CUDA-Enabled Speed Merchant?" reminded me about the poll I kicked off on Survey Monkey in August. At that time I hadn’t discovered the wonders of the embedding polling plugin for WordPress. So, I’ve now rehosted the poll by embedding it into my original post ("Straw Poll On Compute Power").
Almost everyone from the SI-list who filled out those first responses thought it very likely or somewhat likely that advanced computing hardware like many-core, server farms, and GPUs would be adopted in the near future, with varying degrees of likelihood.
I was surprised and delighted how many signal integrity folks were aware of the up and coming technology, which originated in the video game segment, but which has potential for an impact in the EDA domain. Many high-end gaming consoles achieve their photo-real images by farming out complex calculations to a graphics processing unit or GPU. It turns out that these GPUs can do more than rendering calculations and can be used as array processors for parallel computation.
One blog named GPUs #2 in the top 20 most influential things to happen to the video game industry. (The internet was #1 of course.)
Unlike vector supercomputers of decades ago (which didn’t go mainstream because they started at the top of the price/performance curve (DARPA funding) and could not migrate down), GPUs have different economic model: they started at the bottom, sold in millions to gamers, and are moving up. The low end eats the high end.
I guess the article in Electronic Design is another data point that the GPU trend is real.
By the way, they quote our press release in the article without linking back to it 🙁 . Here’s the link they omitted:
David A. Patterson, professor of computer science at Berkeley: "No one knows how to design a 15 GHz processor, so the other option is to re-train all the software developers [to do parallel programming]"