Silicon is stupid without a good compiler.

You all know CUDA is coming to mini Kepler, AKA Tegra, right? How did you think that was going to happen, magic dust?

Nvidia has been working with the Portland group (PGI) since early 2009. They are the folks that brought C++ and FORTRAN to CUDA, they are very very smart, and match Nvidia’s work ethic: code till you drop, sleep when you die.

Founded in 1989, PGI has a long history of innovation in HPC compiler technology and has done work for ARM, Intel, IBM, Linux, OpenMP, and was a pioneer in GPU-compute.

They’ll continue to operate under the PGI flag, developing OpenACC, CUDA Fortran and CUDA x86 for multicore x86 and, of course, GPGPUs. And they will continue to serve their wide-range of customers – including chip makers, research labs and HPC computing centers. PGI did the compilers for the Red Storm supercomputer architecture built by Cray using Opteron processors from Advanced Micro Devices and the SeaStar interconnect developed by Cray to lash them together.

Ian Buck, Nvidia’s GM of GPU Computing Software, said the deal strengthens the OpenACC initiative to create an easy on-ramp to parallel computing by joining the world’s top independent provider of OpenACC compilers with the world’s best GPU designers.

“Behind every great processor technology stands a great compiler team. Ours just got bigger. And better,” said Buck

Our take

It’s about time. This just makes sense. If you’re going to be in the processor business, you have to have a good—no, great—compiler team. Nvidia already has a damn fine GPU compiler group; this makes it even better.

Nvidia bought out ST Micro who originally owned PGI. There’s about 27 folks in Portland who are now getting their arms tattooed with a Nvidia serial number, and grinning while it’s done. I think the deal probably went down for about $30 million; it’s not material, and that’s why Nvidia was able to disclose it now during their quiet period (they’ll announce their quarterly results on August 9).  

Interestingly, AMD was one of PGI’s best customers, and PGI will continue to work with and support AMD—business is business. Same holds true for Intel.

I think it’s a terribly smart and good deal and congratulate both parties.