422
edits
Changes
GPU
,update on instruction throughput section
* TensorCores
: With Nvidia [https://en.wikipedia.org/wiki/Volta_(microarchitecture) Volta] series TensorCores were introduced. They offer fp16*fp16+fp32, matrix-matrix-multiplication units, used to accelerate neural networks.<ref>[https://on-demand.gputechconf.com/gtc/2017/presentation/s7798-luke-durant-inside-volta.pdfINSIDE VOLTA]</ref>
==Throughput Examples==