[Windows] RTX 5070 Ti (Blackwell sm_120) - setup and deployment notes

[loongmiaow-pixel]

Environment

• GPU: NVIDIA GeForce RTX 5070 Ti Laptop GPU (Blackwell, compute capability 12.0)
• Driver: 595.79 (CUDA 13.2)
• OS: Windows 11
• Python: 3.14
• accelerate: [latest]

Problem

accelerate on RTX 5070 Ti requires workarounds:

• sm_120 not detected by PyTorch/CUDA runtime
• accelerate config may not list RTX 5070 Ti correctly
• Distributed training may fail without explicit arch list

Solution

$env:TORCH_CUDA_ARCH_LIST = "12.0"
$env:CUDA_VISIBLE_DEVICES = "0"

Then run accelerate config as normal.

Question

Is there a plan to update device detection for Blackwell GPUs?

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[SunMarc]

Hey @loongmiaow-pixel , this is more of an issue on pytorch side.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.