At last week's GPU Technology Conference, NVIDIA's CEO Jen-Hsun Huang unveiled a couple of key products and technologies that target the High-Performance Computing, or HPC, space, including the Tesla P100 data center accelerator and its companion DGX-1 deep-learning system, which is a powerful server that has up to eight Tesla P100 cards at its core. Today though, we have more information about the underlying architecture employed in the P100, otherwise known NVIDIA's Pascal GPU architecture.

Pascal is the follow-up to the Maxwell architecture, which is leveraged in NVIDIA's current-generation of graphics cards and mobile GPUs. The Pascal-based GPU at the heart of the Tesla P100 is codenamed the GP100 and it promises to be a very different animal...