our research lab (robotics) currently is building up a dedicated GPU server and we have first chosen the Nvidia A100 as our first GPU. However this one doesnt seem to have Raytracing cores which might be troublesome when it comes to Isaac Sim. Therefore I would like to ask for experiences which hardware is a good choice to be able to train online-Reinforcement learning models in Isaac sim but also to have computing power for large networks in offline training (image segmentation networks etc.).
Hello @schulze762gr! Here is a link to the technical requirements for Isaac Sim: 1. Isaac Sim Requirements — Omniverse Robotics documentation I hope this helps get you started on what hardware requirements you would need. If you will be processing large amounts of data, obviously, the more GPU VRAM, the better!
thanks for the reply! By ‘server’ I’m more looking towards datacenter-grade GPUs (we also have selected a dedicated server rack etc.) where lots of jobs can run concurrently + a few jobs might ask for Isaac instances. I’m not experienced with setting up proper server infrastructure yet.
The current options I could find so far are either an A100 + A6000 or A4000 (for the people using isaac to strictly train online RL) or an L40 which seems to be very new. It costs half of A100 but also has only half of its bandwidth (at most) as far as I can see.