14 Mar NVIDIA GTC ’20 Could Be Massive
As we approach NVIDIA’s annual GPU Technology Conference, everyone is anxious to see what CEO Jensen Huang has up his trademark black leather sleeves. I have no idea, but as usual, I have an opinion.
An intro to GTC
GTC is NVIDIA’s partners’ big annual opportunity to show off and explain how they leverage NVIDIA GPUs in applications—from autonomous vehicles to drones to gaming PCs to the fastest supercomputers in the world. The show floor is dizzying, filled with booths, robots, automobiles and even an occasional semi-truck! If you’ve never been to GTC, and have any interest in this fast-moving segment, I strongly recommend attending.
Jensen kicks off the some 10,000 person tech-fest with a 2+ hour keynote, which is always informative and entertaining. His showmanship is almost as famous as his technology. This is where NVIDIA typically unveils new hardware and software technologies and showcases leading customer use cases spanning cloud, science, automobiles, feature films, gaming and more. The rest of the event is full of workshops and company presentations, geared for technical, business, and investment professionals. It can be exhausting and energizing in equal measure!
Will there be a new champion?
The big question on everyone’s mind this year is whether NVIDIA will launch the successor to Volta, the undisputed leader in AI training in the data center. One could argue that there is no need to rush a new product to market when there are no legitimate competitors. Others would argue that the competition is amassing at the gates of NVIDIA’s Endeavor HQ, and plan to challenge the leader in the data center. Intel’s acquisition of Habana Labs may set the company back from a schedule perspective, but as I have opined previously, Habana Gaudi may be worth the wait. Cerebras and Graphcore also have powerful training chips being readied and tested which will certainly challenge NVIDIA.
I also expect NVIDIA will concentrate more on inference than training this year at GTC, especially on the software and solutions front. Its T4 revenue for inference in Q4 was probably pretty impressive, though it was not disclosed. Also, NVIDIA continues to expand its traditional defensive software moat, CUDA, dramatically. In fact, software now represents NVIDIA’s largest differentiator; it will take years for newcomers to match it.
My bet is that NVIDIA will at least tease the successor to Volta’s crown. Additionally, it will be interesting to learn how NVIDIA has been using its Saturn V Supercomputer to support the development team working on next gen GPUs. I could be wrong. If you look at NVIDIA’s impressive Q4 results, there is no hint of an impending product transition, as NVIDIA grew its data center business at 43% year over year and margins remained strong. There were no signs of discounting the “old” to make room for the “new.”
As I mentioned earlier, there is still zero competition for Data Center AI silicon for training. Three years after NVIDIA unveiled the Volta processor, it remains the fastest production AI silicon in the market.
Do not miss it GTC 2020. Hope to see you there!