My GPU is TU117M [GeForce MX450]. What is the latest CUDA version I can use ?
Thanks
Your MX450 uses the Turing architecture with compute capability 7.5 and is thus supported in all currently released stable and nightly PyTorch binaries.
Ok, that’s interesting to know.
I’m using torch 2.0.1
which cuda https://download.pytorch.org/whl/cu118 but I’m still receiving torch.cuda.is_enabled()
False due to torch._C._cuda_getDeviceCount() >
returning 0.
Any insights on what I’m missing here?
[[tool.poetry.source]]
name = "default"
url = "https://pypi.org/simple"
[[tool.poetry.source]]
name = "pytorch"
url = "https://download.pytorch.org/whl/cu118"
secondary = true
[tool.poetry.dependencies]
..
python = "~3.10.12"
pytorch-forecasting = { version = "=1.0.0", source = "default" }
..
tensorboard = { version = "*", source = "default"}
tensorboardX = { version = "*", source = "default"}
torch = { version = "=2.0.1", source = "pytorch" }
What does torch.version.cuda
return?
torch.version.cuda
returns 11.8
Do I need any other prerequisite?
My driver is nvidia-driver-545
I upgraded to torch 2.1.0 (torch.version.cuda = 12.1) and still torch._C._cuda_getDeviceCount()
returns 0.
I also have below libraries installed.
❯ apt list --installed | grep cuda
WARNING: apt does not have a stable CLI interface. Use with caution in scripts.
cuda-cccl-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-command-line-tools-12-3/unknown,now 12.3.0-1 amd64 [installed,automatic]
cuda-compiler-12-3/unknown,now 12.3.0-1 amd64 [installed,automatic]
cuda-crt-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-cudart-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-cudart-dev-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-cuobjdump-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-cupti-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-cupti-dev-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-cuxxfilt-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-documentation-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-driver-dev-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-gdb-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-keyring/unknown,now 1.1-1 all [installed]
cuda-libraries-12-3/unknown,now 12.3.0-1 amd64 [installed,automatic]
cuda-libraries-dev-12-3/unknown,now 12.3.0-1 amd64 [installed,automatic]
cuda-nsight-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-nsight-compute-12-3/unknown,now 12.3.0-1 amd64 [installed,automatic]
cuda-nsight-systems-12-3/unknown,now 12.3.0-1 amd64 [installed,automatic]
cuda-nvcc-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-nvdisasm-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-nvml-dev-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-nvprof-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-nvprune-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-nvrtc-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-nvrtc-dev-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-nvtx-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-nvvm-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-nvvp-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-opencl-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-opencl-dev-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-profiler-api-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-sanitizer-12-3/unknown,now 12.3.52-1 amd64 [installed,automatic]
cuda-toolkit-12-3-config-common/unknown,now 12.3.52-1 all [installed,automatic]
cuda-toolkit-12-3/unknown,now 12.3.0-1 amd64 [installed]
cuda-toolkit-12-config-common/unknown,now 12.3.52-1 all [installed,automatic]
cuda-toolkit-config-common/unknown,now 12.3.52-1 all [installed,automatic]
cuda-tools-12-3/unknown,now 12.3.0-1 amd64 [installed,automatic]
cuda-visual-tools-12-3/unknown,now 12.3.0-1 amd64 [installed,automatic]
Hmm interestingly torch.cuda.current_device()
returned an error.
RuntimeError Traceback (most recent call last)
Cell In[16], line 1
----> 1 torch.cuda.current_device()
File ~/.cache/pypoetry/virtualenvs/vfm-nn-DOmiObSy-py3.10/lib/python3.10/site-packages/torch/cuda/__init__.py:769, in current_device()
767 def current_device() -> int:
768 r"""Returns the index of a currently selected device."""
--> 769 _lazy_init()
770 return torch._C._cuda_getDevice()
File ~/.cache/pypoetry/virtualenvs/vfm-nn-DOmiObSy-py3.10/lib/python3.10/site-packages/torch/cuda/__init__.py:298, in _lazy_init()
296 if "CUDA_MODULE_LOADING" not in os.environ:
297 os.environ["CUDA_MODULE_LOADING"] = "LAZY"
--> 298 torch._C._cuda_init()
299 # Some of the queued calls may reentrantly call _lazy_init();
300 # we need to just return without initializing in that case.
301 # However, we must not let any *other* threads in!
302 _tls.is_initializing = True
RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx
I had a default xorg driver from Ubuntu. I installed the below driver downloaded from NVIDIA and that worked.
@ptrblck thanks for helping me to debug on the right path!
±--------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.113.01 Driver Version: 535.113.01 CUDA Version: 12.2 |
|-----------------------------------------±---------------------±---------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 NVIDIA GeForce MX450 Off | 00000000:01:00.0 Off | N/A |
| N/A 60C P0 N/A / ERR! | 5MiB / 2048MiB | 0% Default |
| | | N/A |
±----------------------------------------±---------------------±---------------------+
±--------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
| 0 N/A N/A 3136 G /usr/lib/xorg/Xorg 4MiB |
±--------------------------------------------------------------------------------------+
Good to hear it’s working now!
Just to answer a few things for clarity:
No, you would only need to install a compatible NVIDIA driver and the PyTorch binaries with CUDA support.