A Domain-Specific Supercomputer for Training Deep Neural Networks
Google's TPU supercomputers train deep neural networks 50x faster than general-purpose supercomputers running a high-performance computing benchmark.
Norman P. Jouppi
Advertisement
Author Archives
A Domain-Specific Architecture for Deep Neural Networks
Tensor processing units improve performance per watt of neural networks in Google datacenters by roughly 50x.
Technical Perspective: DRAM Errors in the Wild
In order to advance the field, knowledge of the types of memory errors at the system level, their frequencies, and conditions that exacerbate or are unrelated to higher error rates are of critical importance.
Technical Perspective: Software and Hardware Support For Deterministic Replay of Parallel Programs
Parallel programming has long been recognized as a difficult problem. This problem has recently taken on a sense of urgency: the long march of single-thread performance increases has stopped in its tracks. Due to limitations on power dissipation and decreased return on investments in additional processor complexity, additional transistors provided by Moore’s Law are now […]
Shape the Future of Computing
ACM encourages its members to take a direct hand in shaping the future of the association. There are more ways than ever to get involved.
Get Involved