NVidia’s President and CEO Jen-Hsun Huang gave a short but very clear presentation in the NVidia booth at Supercomputing 2015 about accelerated computing and the impact of deep-learning. In short, look for HPC dominance and a technology transfer that will replace the browser and drive your car.
“Moore’s Law is utterly irrelevant” Huang said. With over $10 billion of total investment to date, Jen-Hsun pointed out that people are seeing “almost unreasonable speedups” though the use of accelerated computing. Meanwhile, “the worst case of accelerated computing is a speedup of only 1x”, which Huang used to humorously highlight the safety of exploring an accelerated computing approach.
The billion dollar return on such an investment is only justified, as Huang noted, by “(1) the need to be commodity and (2) the ability to have a unique value proposition.” In this way, HPC can be brought to the mass market for extreme revenue generation.
For the HPC market: Jen-Hsun told the crowd that “this is a major change year for accelerated computing as 1/3 of the total HPC flops (floating-point operations per second) are now delivered by accelerated computers, with a 50% growth rate per year”.
The migration of technology from HPC to the mass market is mind-boggling. Huang highlighted this by noting that deep-learning has brought speech recognition to the point where, “Microsoft has said the next generation Cortana will replace the browser.” If true, this will change the browsing interface for billions of people and literally the way that people interact with both computers and the web.
Similarly, self-driving cars will change how we transport ourselves and arrange our daily lives. During his talk, Huang acknowledged that NVidia Drive is “the second most accurate self-driving deep-learning network”. It goes without saying that the NVidia team is actively working to make NVidia Drive the accuracy leader. With a credit card sized, TX1 implementation of the Tegra X1 SoC (System On a Chip) now available, NVidia has the hardware platform to bring NVidia Drive to the automotive industry.
On a side-note: The Tegra X1 SoC is generally touted as providing a teraflop inside a 10 watt TDP SoC package. However, TechEnablement received clarification at SC15 that this performance number is actually based on 16-bit floating-point arithmetic. Expect the Tegra X1 to deliver roughly 1/2 TF/s when performing 32-bit, single-precision math.
Huang wrapped-up his presentation by noting that NVidia is bringing Supercomputing everywhere from robots to the browser to the car. HPC (and effectively NVidia) will be part of everyone’s daily life.