I have an Nvidia Geforce 940M GPU, I’ve tried to install CUDA starting from version 7 and above and have installed cuDNN, but why isn’t tensorflow detected? can someone tell me why?
Hello @pengguna109 and welcome to the NVIDIA developer forums!
Can you please be a little bit more specific with your question? Tensorflow is not part of CUDA or NVIDIA drivers, it is an Open Source Deep Learning framework that can be accelerated by using the GPU.
You can find more information about DL frameworks on our Developer pages on the topic. There you will also find further links to Getting Started pages with respect to Tensorflow.
I hope that helps!
Yes i know about it… i was try to install cuda from 5 untill 8 version on my computer wich has Nvidia GPU Geforce 940M, i wanna use it to accelerate my deep learning training model… but till now… the tensorflow cant detect my GPU, what i have to do? Does Geforce 940M support accelerate with cuda on TensorFlow?
The GeForce 940M is CUDA capable, yes.
Did you follow the installation guides on the Tensorflow pages? By far the easiest way to get started is to use a docker image that includes preinstalled dependencies, making it very easy to run Tensorflow projects on GPU. You find pre-made docker images on the Tensorflow pages as well as in the NVIDIA registry.
Wich cuda version will support for Geforce 940M?
The 940M has Hardware support for CUDA version 5, you can look for CUDA Hardware compatibility here: CUDA GPUs - Compute Capability | NVIDIA Developer
I can also recommend CUDA - Wikipedia for a good overview of CUDA version capabilities across HW generations.
I hope that helps
If Nvidia Geforce 940M just Support cuda 5, is this mean that this GPU Type just Support Cuda Toolkit V5?
The version number, in your case 5.0 tells you the version of compute capability. That means the CUDA feature set up until compute v5.0 will run on the 940M.
Since CUDA is, within reason, backwards compatible, even the newest CUDA SDK still supports the 940M. You just cannot expect applications developed against newer compute capability versions than 5.0 to run on the 940M.
Its si poor, because while i begin to install cuda toolkit 5, i get a warning wich say that “this Application will not work on your computer, but you can still install this application on your computer” so i prefer to cancel installation 😭
I am sorry but to be able to help you further please share some more detail.
- What operating system are you using?
- What NVIDIA driver version do you have installed? Not CUDA, just the GPU driver.
Hello, i had an issue about this sample code: https://developer.nvidia.com/blog/even-easier-introduction-cuda/ .
I compiled that code with nvcc and run exe file.
It says Max error: 1 but the author said that it should be 0.
And when i try this exe with nvprof, it says “No kernels were profiled”.
I suspect that the kernel function does not run on GPU.
My setup is this;
OS: Windows 10
Cuda: v12.1
Driver: Version 531.14
Graphics card: Nvidia geforce 940m.
Thank you.
I guess, i cannot use gpu as i suspect. Because i tried this code and it did not print anything.
#include <stdio.h>
global void helloCUDA(float f)
{
printf(“Hello thread %d, f=%f\n”, threadIdx.x, f);
}
int main()
{
helloCUDA<<<1, 5>>>(1.2345f);
cudaDeviceSynchronize();
return 0;
}
“-arch=sm_50” option for nvcc command solved the issue.
I think, in case of newer version of cuda and older version of graphics card,
this kind of options will work.
Thank you.
1 Like
Thank you for sharing this information @azizkeremozkan and welcome to the NVIDIA developer forums!
Great that you cold resolve this issue on your own.
If you have further questions or problems with CUDA, I highly recommend our very active CUDA community in the CUDA category here on this server.
Thanks!
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.