Nvidia Debuts Tesla P100 Accelerator With 15B Transistors for AI, Deep Learning

Nvidia CEO Jen-Hsun Huang used the hole keynote of the employer’s annual GPU era conference to announce a massive new processor designed specifically for deep learning. The Tesla P100 is the first transport product to apply Nvidia’s new Pascal architecture, and is made up of 15.three billion transistors, which the corporation says makes it the largest microchip ever fabricated.

The Tesla P100 is constructed using a brand new 16nm FinFE production procedure and uses 16GB of HBM2 snap shots reminiscence which is incorporated onto the identical chip substrate, which ends up in memory bandwidth of up to 720GBps. height performance is rated at 21.2 Teraflops for half-precision instructions, 10.6 Teraflops for unmarried-precision and five.three Teraflops for double-precision workloads. Up to 8 Tesla P100 chips can be interconnected the usage of Nvidia’s NVLink bus.

The Tesla P100 is claimed to deliver over 12x the performance of Nvidia’s previous technology Maxwell structure in neural community education eventualities. precise packages, consisting of the AMBER molecular dynamics code, are stated to run faster on one Tesla P100 server node than on forty eight twin-socket CPU server nodes, in keeping with Nvidia.

Huang also stated that the employer has deciced to “move all-in on AI”, and that deep learning and synthetic intelligence are the business enterprise’s fastest developing business location. He named numerous areas of studies, which include finding a therapy for most cancers and know-how climate trade, which require computing resources which could scale infinitely.

Massachusetts popular medical institution has installation a medical datacentre which will use Nvidia’s AI processing era to assist diagnose diseases beginning with the fields of radiology and pathology, and could use its archive of 10 billion medical photographs to create a deep mastering neural network.

The Tesla P100 will first of all be to be had in Nvidia’s new DGX-1 “deep gaining knowledge of supercomputer” in June, and in servers from a number of manufacturers starting in early 2017. The DGX-1 could have eight Tesla P100 chips for a combined a hundred and seventy Teraflops of 1/2-precision overall performance, and is alleged on the way to supply the deep getting to know throughput of 250 conventional x86 servers in a unmarried 3U server enclosure.

Leave a Reply

Your email address will not be published. Required fields are marked *