Selecting Servers and GPUs

https://d2l.ai/chapter_appendix-tools-for-deep-learning/selecting-servers-gpus.html


from https://alexiej.github.io/deepnn/#gpu-servers

from https://course.fast.ai/index.html

https://docs.nvidia.com/deeplearning/frameworks/

  1. https://docs.nvidia.com/deeplearning/frameworks/mxnet-release-notes/index.html
  2. https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/index.html
  3. https://docs.nvidia.com/deeplearning/frameworks/tensorflow-release-notes/index.html

NVIDIA Data Center Deep Learning Product Performance
Google pytorch vm


FrameView Performance and Power Benchmarking App

Graphics cards are investments, bought on the promise of delivering excellent performance in that card’s class for at least 2 to 3 years. It’s therefore important that the benchmarks of these cards, which you’re likely using as research, are accurate and cover all the bases, showing frame rates, frame times, power usage, performance per watt, and more.

Comparison of learning and inference speed of different gpu with various cnn models in pytorch

  • 1080TI
  • TITAN V
  • 2080TI

Horovod synthetic benchmarks
Transformers Benchmark