This is just a short post to announce a more usable version of the NVIDIA GPU powerlimit setup script that I released a few months ago. This update to version 0.2 uses an interactive mode to set GPU powerlimits and optionally setup a systemd unit file to set these limits on subsequent reboots.
This post presents testing data showing that power-limit reduction on NVIDIA GPUs have give significant benefits for both high wattage and lower wattage GPUs. Power-limit vs Performance data is presented for 1-4 A5000 and 1-4 RTX3090 GPUs.
In this post I am referencing a Bash shell script I recently put together for setting up automatic NVIDIA GPU power-limit lowering at system boot. This allows a reliable way to configure and maintain multi-GPU systems for stable operation under heavy load.
I have been qualifying a 4 GPU workstation for Machine Learning and HPC use. The last confirmation testing I wanted to do was running it with TensorFlow benchmarks on 4 NVIDIA Titan V GPU’s. I have that systems up and running and the multi-GPU scaling looks very good.
Batch size is an important hyper-parameter for Deep Learning model training. When using GPU accelerated frameworks for your models the amount of memory available on the GPU is a limiting factor. In this post I look at the effect of setting the batch size for a few CNN’s running with TensorFlow on 1080Ti and Titan V with 12GB memory, and GV100 with 32GB memory.
Tensor-cores are one of the compelling new features of the NVIDIA Volta architecture. In this post I discuss the some thought on mixed precision and FP16 related to Tensor-cores. I have some performance results for large convolution neural network training that makes a good argument for trying to use them. Performance looks very good.
Another great GTC meeting. NVIDIA does this right! The most interesting aspects for me this year were the talks on “Deep Learning” (Artificial Neural Networks) and OpenPOWER. I have some observations and links to recordings of the keynotes and talks. Enjoy!