NVIDIA AND ME: PASCAL, VR, AND DEEP LEARNING
92 HWM | M AY 2 0 1 6 This year’s NVIDIA GTC (GPU Tech Conference) at San Jose, California was marked with a cornucopia of announcements and presentations that revolved around the application of GPU technology in Virtual Reality (VR), Deep Learning, and of course, (it just wouldn’t be an NVIDIA event without a mention of it), Self-Driving Autonomous Vehicles.
And because this is essentially the year where everyone and everything is going VR, NVIDIA spared no expenses in giving all attendees a chance to experience the various VR demos on the show floor, which we’ll get to momentarily.
Traditionally, any and all announcements of new products, projects, and future endeavors were usually made during GTC (at least, since NVIDIA began hosting its own conference at its home base).
While there was a lot to get excited about this year, almost every tech journalist at GTC 2016 were waiting for just one thing from NVIDIA CEO Jen-Hsun Huang, and that was the announcement and mention of just one name: Pascal.
The GPU itself has long been in the works since it was first announced back when we attended GTC 2014. Compared to the Maxwell architecture (which NVIDIA has reportedly already discontinued), it goes without saying that Pascal is exponentially more powerful and efficient compared to its predecessor. It utilizes a brand new High Bandwidth Memory format known as HBM2, and as promised before, it will come with support for the new NVLink feature, which allows multiple GPUs to link up with each other on a single system without any limitations in performance.
But Pascal’s announcement was met with both a mix of laughter, cheer, and sheepish disapproval when Jen-Hsun unveiled the ark in which NVIDIA’s new GPU architecture would be carried upon, and that, dear folks, is the Tesla P100. Yes, we know: It wasn’t a new GeForce card.
Armed with 16GB of HBM2 that was built on a brand new 16nm FinFET process, 720GB/s of memory bandwidth, and as many as 3,584 FP32 CUDA cores (for single precision), and half of that number for double precision CUDA cores, the Tesla P100 is NVIDIA’s, and by all counts, the world’s most powerful GPU in existence.
The Tesla P100 will be used inside a Deep Learning supercomputer known as the DGX-1.
The DGX-1 supercomputer, in the flesh.
As always, the first keynote of GTC 2016 was by Jen-Hsun Huang, CEO of NVIDIA.
Tesla P100, the first Pascal-powered GPU.