In a nutshell, I want to use TLT (or TAO as it’s called now) but I have a Quadro RTX5000 and a T4. Is there a way to train NVIDIA’s models without buying an RTX30xx or a V100?
Quadro RTX5000 or T4 can work.
I think the tao user guide may confuse end users. We will modify it.
For training, actually at least 1 NVIDIA GPU is available and make sure its compute capability is equal or higher than 6.1. You can check the dgpus and compute capability in CUDA GPUs | NVIDIA Developer
For deployment, all Jetson products are supported. And all dgpus which have compute capability equal or higher than 5.3 are supported.
1 Like
Thank you so much for this :)
One little follow-up question, I run Debian Bullseye on some machines and I know that both Deepstream and TLT are made for ubuntu 20.04. Is there a way around it like building from source to move past the Unsupported OS message?
Officially TAO is supported on Ubuntu 18.04.
For your case, you can pull a Ubuntu 18 docker in your Debian system, then run tao docker in Ubuntu docker(For example, mcr.microsoft.com/azureml/openmpi4.1.0-cuda11.1-cudnn8-ubuntu18.04)
1 Like
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.