Kernel code is not working for NVIDIA GPU but working in CPU mode

3
votes

Recently I've tried to run a program, which was correctly compiled and executed on AMD Radeon 5670 GPU which is running Windows 7. Now, when I wanted to run on NVIDIA GPU Tesla 2500, it was not working. But the same Kernel code is giving the output when I change the device to CPU (uisng clGetDeviceIds() argument CL_DEVICE_TYPE_GPU). Here I'm using Ubuntu 10.10.

I tried running some of the sample programs which come with the NVIDIA SDK those are running properly and showing the details of existing GPU's and configurations.

Please respond to this question where is the problem if any one has undergone this kind of problems.

1
copenclgpu
Your question is overly general. What is the smallest snippet of code you can identify that gives different results? - Eli Bendersky
¿Did you use clGetProgramBuildLog to check for errors? - Zhen
We're going to have to see some minimal section of code which exhibits the issue for you. Try to narrow it down as far as possible. Without it we can only make guesses, if we can suggest anything at all. - Bart
Thanks for your suggestions. I checked using clGetProgramBuildLog.. The problem is clGetDeviceIDs() is failing.. Is it the problem of drivers not installed properly??? - Vasanth Raja
I checked even the drivers NVIDIA Tesla drivers those are installed properly.. System has two Graphic cards will that cause any problems? - Vasanth Raja

1 Answers

0
votes

I found the solution for the above problem... First of all thanks to Eli Bendersky, Zhen and Bart..

The problem I was facing was due to clGetDeviceIDs() failure. This is because the platform I chose is AMD platform(I have both AMD and also CUDA but the underlying GPU are NVIDIA GPUs) so This platform could not find the underlying GPUs..

At last, after changing the platforms[0] to platforms[1] while accessing the clGetDeviceIDs() I got the code working...

:)