Simple trick to double deep learning speed and CNN and GPU Benchmarks (github.com)

Hi, friends. With NeurIPS behind us and ICML ahead, maybe you want to do some deep learning. Inspired by Justin Johnson's original work benchmarking the older GTX GPUs, I extended this work to the new RTX GPUs with benchmarks for most ResNet architectures on ImageNet and CIFAR. Along the way, I discovered a dramatic difference in performance based on how you position your GPUs. Enjoy :)

Benchmarking post: https://l7.curtisnorthcutt.com/benchmarking-gpus-for-deep-le...

Positioning (speed-up) post: https://l7.curtisnorthcutt.com/gpu-positioning

GitHub with Benchmarks: https://github.com/cgnorthcutt/cnn-gpu-benchmarks