NVIDIA GeForce 940MX torch.cuda.is_available() False

Hi I’ve just installed PyTorch through Anaconda distro in Windows 10. And when i call torch.cuda.is_available() it returns False. I have an NVIDIA GeForce 940MX graphic card. It seems that my graphics card is not compatible with the latest version of CUDA.

How can I get a solution that allows me, for example, to allocate a tensor in the memory of the GPU?

What is the output of nvidia-smi command?

I can’t run that command in my cmd promt.

These are my files located in the directory where i had founded a file named nvidia-smi

C:\Program Files\NVIDIA Corporation\Installer2\Display.Driver.{5C61AB78-CA89-4A7E-895C-1D8179B0261F}>ls

DisplayDriver.NVX nv-vk32.json nvdecmftmjpeg.dl_ nvfbcplugin.dl_ nvoglv64.dl_
DisplayDriver.nvi nv-vk64.json nvdecmftmjpegx.dl_ nvfbcplugin64.dl_ nvopencl32.dl_
DisplayDriverExt.dll nvam.cat nvdet.dl_ ** nvidia-smi.1.pd_ ** nvopencl64.dl_
GeForce_iCafe.exe nvam.inf nvdetx.dl_ ** nvidia-smi.ex_ *** nvpciflt.sy_
GeForce_iCafeExt.dll nvapi.dl_ nvdispco64.dll nvifr.dl_ nvptxjitcompiler32.dl_
NvContainerSetup.exe nvapi64.dl_ nvdispgenco64.dll nvifr64.dl_ nvptxjitcompiler64.dl_
NvCplSetupInt.exe nvcompiler32.dl_ nvdrsdb.bi_ nvifropengl32.dl_ nvsetup.exe
NvSplash.exe nvcompiler64.dl_ nvdxgiwrap.dl_ nvifropengl64.dl_ nvumdshim.dl_
NvSplash.wmv nvcuda32.dl_ nvdxgiwrapx.dl_ nvinfo.pb nvumdshimx.dl_
NvSplashService.exe nvcuda64.dl_ nvencmfth264.dl_ nvinit.dl_ nvwgf2um.dl_
NvSplash_ZH.wmv nvcuvid32.dl_ nvencmfth264x.dl_ nvinitx.dl_ nvwgf2um_cfg.dl_
VulkanRT-Installer.exe nvcuvid64.dl_ nvencmfthevc.dl_ nvlddmkm.sy_ nvwgf2umx.dl_
dbInstaller.exe nvd3d9wrap.dl_ nvencmfthevcx.dl_ nvldumd.dl_ nvwgf2umx_cfg.dl_
flxcomm64.dl_ nvd3d9wrapx.dl_ nvencodeapi.dl_ nvldumdx.dl_ opencl32.dl_
flxconnect64.dl_ nvd3dum.dl_ nvencodeapi64.dl_ nvmcumd.dl_ opencl64.dl_
flxcore64.dl_ nvd3dum_cfg.dl_ nvfatbinaryloader32.dl_ nvml.dl_ pvsonydll32.dl_
license.txt nvd3dumx.dl_ nvfatbinaryloader64.dl_ nvoglshim32.dl_ pvsonydll64.dl_
logo.png nvd3dumx_cfg.dl_ nvfbc.dl_ nvoglshim64.dl_
mcu.ex_ nvdebugdump.ex_ nvfbc64.dl_ nvoglv32.dl_

Install CUDA to get nvidia.smi.

I think I already installed it via “conda install pytorch torchvision cudatoolkit = 10.2 -c pytorch” in Anaconda Promt. Should I do the installation again?

You can install it directly from NVidia website which also install the right driver if you don’t have it (https://developer.nvidia.com/cuda-10.2-download-archive).

Hai, i have nvidia geforce 940mx hardware on my laptop. it’s said the cuda version is cuda 5.0. i checked the toolkit driver options. but cuda 5.0 toolkit is not compatible for windows 10. what’s should i do?

The 940mx would have a compute capability of 5.0 so should be supported in the current pip wheels and conda binaries.