Hsin-Ping

@xbresson I'm running the code with PyTorch 1.0.0 CUDA 10.0 Can you please share the spec of your server or any tips that should be taken care of?

Benchmarking my new RXT2080Ti GPUs :) on ImageNet with ResNet50: github.com/pytorch/exampl… Got 23min/epoch with 4 x RXT2080Ti. Seems okay but I'd appreciate to know if you got similar time?

@xbresson I had serious I/O issue when the dataset was stored on a HDD. I then store the dataset on a SATA SSD which gave significant improvement but still far from your results. I got 1hour/epoch with 4 RTX 2080ti on ImageNet with ResNet50.

@xbresson Hi Xavier, I ran the exact same code. Below are the specs of my server. Supermicro 7048GR-TR 4U server CPU: Intel XEON E5-2620 *2 RAM: DDR4 16G *4 (another 64G swap memory is configured as well) GPU: ASUS Turbo GeForceR RTX 2080 Ti *4 HDD: 4TB *1 SSD: 256GB SATA SSD