Jump to content

NVIDIA Announces the Tesla V100 GPU Architecture


Recommended Posts

NVIDIA has announced a new addition to the Tesla product line featuring the Volta V100 GPU architecture. The Volta V100 has been dubbed the "World’s Most Advanced Data Center GPU." At the heart of Volta's GV100 GPU lies 21.1 billion transistors crammed into an 815 mm2 die. Not only does the new Volta architecture offer considerably more power than previous generations, it incorporates many new features, ultimately simplifying GPU programming. Below are some of the features included in the Tesla V100 architecture to give you an idea of just what's inside each card.

  • New Streaming Multiprocessor (SM) Architecture Optimized for Deep Learning
  • Second-Generation NVIDIA NVLink™
  • HBM2 Memory: Faster, Higher Efficiency
  • Volta Multi-Process Service
  • Enhanced Unified Memory and Address Translation Services
  • Maximum Performance and Maximum Efficiency Modes
  • Cooperative Groups and New Cooperative Launch APIs
  • Volta Optimized Software

According to NVIDIA, the Volta architecture offers a leap forward in "deep learning performance" thanks to the new Tensor cores. The GV100 GPU reportedly offers up to 7.5 teraflops of double precision performance and 15 teraflops of single precision performance. The Volta GPU accomplishes this performance by utilizing a total of 640 Tensor cores. According to NVIDIA, these Tensor cores offer six times higher peak TFLOPS over the standard FP16 operations on the older P100 line. For more information on the Volta architecture, hit up the source below to check out the entire white paper!

Source: NVIDIA

Back to original news post

Share this post

Link to post
Share on other sites

  • Create New...