On Tuesday at GPU Technology Conference 2010 nVidia CEO and co-founder Jen-Hsun Huang spoke about the success of GPUs [graphics processing units], the developer momentum behind CUDA, and the importance of Parallel Computing to a broad cross-section of the industry. At the end he showed one of those rare nVidia roadmaps of upcoming products.
The morning started with 3D glasses and a series of high-impact videos. Then Jen-Hsun launched into a nearly non-stop hour and half outline of nVidia’s successes. He clearly is enthusiastic about the CUDA [Compute Unified Device Architecture], nVidia’s programming language for general-purpose GPU computing. He said "we recognize we don’t want to replace the CPU, we want to add performance on it using GPUs."
Jen-Hsun then talked about his definition of visual computing and explained how many organizations are using power of the GPU to tackle difficult parallel computing problems. There were more 3D videos focused around upcoming games. Tony Tamasi, nVidia’s senior VP of content and technology, showed off demos that are well beyond what today’s games are doing. The processing power of the green GPGPU with CUDA has really raised the bar for the red graphics card supporters – and it is great that we have two companies that go the distance to increase the performance and impress the market.
Next up was Dr. Subbiah, VP of Global Business Development at ANSYS, talking about their latest design simulation application, Mechanical R13. ANSYS is using the latest nVidia Tesla GPUs in conjunction with a quad-core processor. The combination can cut overall turnaround time in half on typical workloads, when compared to running solely on the quad-core processor.
We were introduced to how seriously nVidia is focused on the unique needs of engineers and scientists who are developing products or conducting research.
Jen-Hsun then talked about scientific computation application Matlab, a numerical computing environment and programming language developed by MathWorks. Their Release R2010b just announced will support CUDA-accelerated GPUs and make GPU computing instantaneously available to a million users doing important work around the world. In 12 months, Matlab went from being accelerated through JACKET library by AccelerEyes to native support inside the app.
A particularly compelling example was the conversation about real-time heart surgery with Michael Black, MD. That one will be covered in another article.
Parallel Computing – 3rd Pillar of Science
The hardware world was not forgotten. Jen-Hsun announced three more major OEMs to support Tesla based servers: IBM BladeServers, a Tesla addition to the Cray XE6 supercomputer line and T-Platforms, the largest high performance server supplier in the Russian market. HP and Dell are already shipping Tesla-based server products.
We previously wrote about Peer 1 and the Cloud. There was a lot of conversation about the mobile device, the cloud, and remote graphical servers making high-powered desktops less important in the future.
Adobe showed their plenoptic lenses. Multiple lens all capture slightly different parts of a scene from slightly different angles, onto very large digital sensors. Then with the Adobe software the user can do some amazing things to digitally alter focus. Future cameras built with plenoptic lenses could make blurry, out-of-focus images something the kids of today never have to deal with.
Jen-Hsun finally put up a slide with a glimpse of nVidia’s future roadmap after today’s Fermi. Their next generation GPUs are code named "Kepler" and "Maxwell," and will arrive in 2011 and 2013, he said. Maxwell will add about sixteen times the Dual-Precision performance of Fermi, Jen-Hsun said in the keynote. The interesting bit about the performance is that currently, nVidia Fermi architecture yields 1.5 TFLOPS of Single-Precision and 750 GFLOPS in Dual-Precision. If nVidia does not change the architecture or raise efficiency of dual-precision, next-year’s architecture would deliver 3TFLOPS of Dual-Precision and 6TFLOPS of Single-Precision performance, while Maxwell could yield 12 TFLOPS in Dual-Precision and 24 TFLOPS in Single-Precision. Naturally, the margin of error is very large, and the roadmap itself isn’t perfect. After all, Tesla architecture [GT200] was introduced in a briefing to a small group of analysts in June 2008, followed by Editor’s day for the consumer part and public debut of Tesla Architecture in the form of GeForce GTX 280 happened on June 16, 2008. The picture below disagrees with that timeline.
NVIDIA dropped t
he paranoid act and officially announced codenames for the next gen architectures… however, was Tesla [GT200-class GPUs] really launched in 2007? We don’t think so…
Afterward at the press briefing Jen-Hsun admitted that Tegra and the tablets were a year behind his expectations. He said that because of Apple’s tightly integrated operating system, software applications, and hardware, they were out in front for the moment. Jen-Hsun said that Tegra was dependent on five or six different manufacturers, each with their own skill set and priorities, which need to come together to put Tegra-powered tablets into customers hands. He was asked when that would happen. He answered that he was not exactly sure, but felt confident they would be in customers? hands by Q1 2011.
Jen-Hsun claimed "the Tegra 3 design work was done," with the implication samples in the wild. He said "the Tegra 4 is being built" Does that mean ‘taped out’, are test wafers are coming out of someone’s fab, or what? At that point he shook his head and said "cannot comment on unannounced products" and mumbled something about "having said too much already."
Jen-Hsun gave reporters an insight into his personal philosophy. He said that "he need an atmosphere of competition". He explained that he didn’t care who cured cancer – the reporters were stunned and sat and waited for the next shoe to drop. Jen-Hsun explained that he just wanted someone to cure cancer, Alzheimer Disease, or build a real "green car". He didn’t care which company or country got the credit. He said "just do it!"