Availability of a PyTorch version compatible with CUDA 12.6

Accelerated Computing Intelligent Video Analytics DeepStream SDK
,
1

GPU

deepstream-7.1

JetPack version is R36 with Revision 4.0.

TensorRT version 10.3.0.30-1+cuda12.5

NVIDIA-SMI 540.4.0 Driver Version: 540.4.0 CUDA Version: 12.6

Is there a PyTorch version available that supports CUDA 12.6?

Whenever i run this command

yolo task=detect mode=predict model=yolo11s.pt source=‘/home/orin/Documents/yolo11/h.mp4’ imgsz=720 device=0

I am facing this error

Error in cpuinfo: prctl(PR_SVE_GET_VL) failed
Ultralytics 8.3.53 🚀 Python-3.10.12 torch-2.5.0
Traceback (most recent call last):
File “/home/orin/Documents/yolo11/y11/bin/yolo”, line 8, in
sys.exit(entrypoint())
File “/home/orin/Documents/yolo11/y11/lib/python3.10/site-packages/ultralytics/cfg/init.py”, line 972, in entrypoint
getattr(model, mode)(**overrides) # default args from model
File “/home/orin/Documents/yolo11/y11/lib/python3.10/site-packages/ultralytics/engine/model.py”, line 551, in predict
self.predictor.setup_model(model=self.model, verbose=is_cli)
File “/home/orin/Documents/yolo11/y11/lib/python3.10/site-packages/ultralytics/engine/predictor.py”, line 310, in setup_model
device=select_device(self.args.device, verbose=verbose),
File “/home/orin/Documents/yolo11/y11/lib/python3.10/site-packages/ultralytics/utils/torch_utils.py”, line 192, in select_device
raise ValueError(
ValueError: Invalid CUDA ‘device=0’ requested. Use ‘device=cpu’ or pass valid CUDA device(s) if available, i.e. ‘device=0’ or ‘device=0,1,2,3’ for Multi-GPU.

torch.cuda.is_available(): False
torch.cuda.device_count(): 0
os.environ[‘CUDA_VISIBLE_DEVICES’]: None
See Start Locally | PyTorch for up-to-date torch install instructions if no CUDA devices are seen by torch.

please help me with this at your earliest convenience?

3

In fact, you can use pip to install the corresponding version of CUDA for pytorch separately, which does not conflict

But I think you need to install the correct version of Jetpack first. For DS-7.1, it should be Jetpack 6.1. Please re-flash it using SDKManager

https://docs.nvidia.com/sdk-manager/index.html

4

Is there any pytorch and cuda version that supports deepstream version 7.1 and JetPack version R36 ?

5

R36 Revision 4.0 is a development version, not a release version, so I am not sure if torch can work properly. I installed the latest version of torch (2.5.1) and it works on the AGX orin which I installed JP-6.1.

6

I tried downgrading CUDA to versions 12.1 and 11.8, and installed PyTorch according to the official website instructions for their respective CUDA versions, but PyTorch still doesn’t recognize CUDA. Given that the issue persists, as you mentioned, could the JetPack version I am using be the reason why PyTorch is not working properly with CUDA?

7

Sorry for the long delay because a holiday.

1.If you flash Jetpack 6.1, CUDA-12.6 default version, I’m not sure if other versions will work fine.

  1. Did you switch the CUDA version correctly? Is the output of nvcc --version the CUDA version you expected?
8

It’s fine; hope you had a great holiday!.

I’ll upgrade the JetPack version to 6.1.

Yes, I switched the CUDA version correctly, and the output of nvcc --version matched the expected CUDA version.

9 
pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu126
1 Like
10

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.