Continuing my series of dinner sessions with HP partners, I spent an evening this week with Nvidia executives including CTO Steve Scott and Executive VP of worldwide sales Jay Puri. The venue was Nvidia’s annual worldwide sales force “bootcamp” with entertainment provided by none other than Nvidia’s own rock band that blasted out quite respectable covers late into the evening. I was one of the few non-Nvidia employees attending the event which turned out to be a great setting to hear not only some of my favorite tunes but also hear about Nvidia’s plans for big growth. I can learn a lot more about a company in a few hours of casual dinner conversation than I can from days of Powerpoint reviews.
HP has partnered with Nvidia for over 10 years, was one of the first companies to design a server specifically around GPUs (the ProLiant SL390s, used in Tokyo Tech’s TSUBAME2.0 system and many other of the world’s most energy efficient supercomputers), and continues to raise the bar for GPU computing with our latest Gen8 SL250s server. Nvidia’s Tesla marketing lead Sumit Gupta already shared on his blog some of the great performance results Nvidia is achieving with HP’s SL250 server configured with three Nvidia M2090 GPUs. Today, HP is not alone in offering GPU-enabled servers, with every major vendor sporting Nvidia GPUs in their catalog, so where is Nvidia looking to for future growth?
While not explicitly stated by any Nvidia exec, compared to the same event I attended last year, it was clear that Nvidia has investing in hiring staff to address the fast growing international market. I talked to a number of Nvidia sales and technical employees from India, China, and Russia that had joined the company in the last twelve months. It is no surprise that HP’s SL390s has seen great sales in each of these regions.
Catching up with Steve Scott, it was clear that he is spending an increasing about of his time thinking about GPU programming models and making it easier than ever for a broader audience of developers to utilize the massive floating point performance delivered by GPUs like the Nvidia M2090 used in HP’s ProLiant SL390s and SL250s servers. Top of mind for Steve was OpenACC, launched by Nvidia and partners last November at SC11.
OpenACC is a new open parallel programming standard designed to enable the millions of scientific and technical Fortran and C programmers to easily take advantage of the transformative power of heterogeneous CPU/GPU computing systems.
OpenACC allows parallel programmers to provide simple hints, known as “directives,” to the compiler, identifying which areas of code to accelerate, without requiring programmers to modify or adapt the underlying code itself. By exposing parallelism to the compiler, directives allow the compiler to do the detailed work of mapping the computation onto the accelerator.
Directives provide a common code base that is multi-platform and multi-vendor compatible, offering an ideal way to preserve investment in legacy applications by enabling an easy migration path to accelerated computing. While it is fair to say the OpenACC initiative is still in its early days, the HPC community has always favored open standards that don’t lock them into a single vendor, on either the hardware or software side, and it is exactly because of this that I believe OpenACC will see rapid adoption over the next twelve months.
Now I just need to sell enough Nvidia GPUs that I get invited back to their event next year!