Forums

let's see Tesla P100 gpu from nvidia

  • Fri 12th Jan 2018 - 6:28pm
    NVIDIA has just introduced its latest GPU, the Tesla P100, which is currently NVIDIA's fastest GPU. Tesla P100 is made with Pascal architecture that has been introduced by NVIDIA since last year. This GPU will not be present in the consumer-grade graphics card. However, seeing its performance powered by 150 billion transistors and powered by HBM2 memory, this GPU that will emerge as one of today's fastest GPUs. Then what technologies are embedded in it? To PC GAMER, NVIDIA Senior Architect, Lars Nyland, and Chief Technologist for GPU Computing Software NVIDIA, Mark Harris, explained that the number of transistors and HBM2 is not the only feature featured in the Tesla P100. GPU uses Pascal architecture designed for all scenarios. Unlike the previous Maxwell architecture, Pascal has performance for all purposes, ranging from intricate computing and gaming. Pascal also has a better FP64 performance than Maxwell. Looking further, Pascal is more similar to the Fermi architecture that offers FP64 performance of 1/3 of the average FP32 performance. While Maxwell has FP64 performance of 1/32 of the average FP32 performance. That is why, many science applications can not work properly when processed using Tesla K40 or K80 GPUs that use Maxwell architecture. FP64 is a data processing technique using a double-precision floating-point format system. As the name suggests, FP64 takes up 8 inches of space in memory (64 bits) for floating-point. While FP32 is a data processing technique using a single-precision floating-point format system that takes up 4 bytes (32 bits) of space in memory for floating-point. Many science applications use FP64 because it can deliver better simulation performance. In addition comes with some changes in the performance, Pascal also comes with various other technologies. NVILink is one of them. This technology is claimed to improve communication between GPU diapsang on different graphics cards, so the speed of communication does not depend on PCI Express which has limited bandwidth. NVIDIA also uses HBM2 technology, which is the latest generation of High Bandwidth Memory technology in AMD's GPU Fiji. To boost performance, NVIDIA adds Page Migration Engine, so the Tesla P100 GPU is capable of having virtual memory up to 512TB. Need to get the best low power graphics card? justvisit this complete article guide and list for pc gamer From a computational standpoint, Tesla P100 with Pascal architecture does look to have a significant performance leap, in line with what 16nm FinFET processing architecture offers on paper. The ability is certainly very interesting, especially if Pascal has been applied in the consumer class GPU for gaming needs later.

Please register or login to post forum replies