[Bug Report] Check cuda version before Install PyTorch 2.7.0 with CUDA 12.8 support

[luoye2333]

Describe the bug

I am using gpu 4090 and cuda version 12.3. In the install script it forces to install PyTorch 2.7.0 with CUDA 12.8 (commit #2998 ) and this download is also super slow in china. Please consider add a cuda version check before install and use mirror links to download if needed.

Currently comment the code in this commit can solve this.

Steps to reproduce

./isaaclab.sh --install

System Info

• Commit: Updates torch to 2.7.0 with cuda 12.8 blackwell support #2998
• Isaac Sim Version: 4.5.0
• OS: Ubuntu 22.04
• GPU: RTX 4090
• CUDA: 12.3

Checklist

• I have checked that there is no similar issue in the repo (required)
• I have checked that the issue is not in running Isaac Sim itself and is related to the repo

[RandomOakForest]

Thanks for posting this. @kellyguo11 for vis.

[Robinroy-peter]

does it work without any freezing while traning 10M or 20M steps in drl

[ZhouBin818]

I also meet this problem. My cuda version is 12.3 but in the install script it forces to install PyTorch 2.7.0 with CUDA 12.8. Could you tell me if it will affect works and could I downgrade the version of pytorch to adapt to the cuda version?

[15101051]

I have the same issue. The public server driver I’m using supports CUDA 12.7, and for stability the driver isn’t updated frequently. I’d like to ask whether using the Torch 2.7.0+cu126 version would have any impact.