CPU Mode on classification example

[flymer]

Hi,

I've managed to compile classification.cpp without any problem. But I want it to use with the CPU Mode and I cant get it to work. I keep getting this CUDA error:

F1003 19:48:48.537168 1992 cudnn_conv_layer.cpp:31] Check failed: error == cudaSuccess (35 vs. 0) CUDA driver version is insufficient for CUDA runtime version

I tried forcing it with the line "Caffe::set_mode(Caffe::CPU);" and playing with the VS compiling properties but with no success.

I am sure I am missing someting, can you help me pls?

[chipgarner]

See the first line in cudnn_conv_layer.cpp:
#ifdef USE_CUDNN
You have USE_CUDNN defined. In VS go to the caffe project Properties, C/C++, Preprocessor and delete USE_CUDNN. You may need to delete it from one or more of the other projects, you will see if you hit a similar error in one of them.

[happynear]

Yes, you should remove the preprocessor macro USE_CUDNN before compiling.

[Kwongpuikong]

I do delete the all the USE_CUDNN in those projects. What else do I miss?

[happynear]

@Kwongpuikong ,
That's all. I think you have came across another question. Please post more message for us to diagnose.

[Kwongpuikong]

Cause I dont have a gpu, I delete all the USE_CUDNN in all projects. I can compile all the projects successfully. Yes, I can compile all the projects successfully.
The problem comes when I want to run the mnist example.
I set the solver_mode: CPU in lenet_solver.prototxt,
and modify the bat file like:
.\bin\caffe.exe train --solver=examples/mnist/lenet_solver.prototxt
It shows the problem [internal_thread.cpp:26] check failed:error == cudaSuccess(35 vs. 0)
I check the internal_thread.cpp, I think the CPU_ONLY is not defined.
Therefore, it checks my GPU which I do not have.
What can I do?
I am a beginner with caffe. Can you help me pls?

[happynear]

@Kwongpuikong ,
Try to add CPU_ONLY macro. However, I haven't tested it. There may be some more bugs.

[flymer]

Oooh thanks, solve it! I had actually removed USE_CUDNN and added CPU_ONLY but only in the caffe project, I also needed to do it in the caffelib.

Finally it works! Wonderful, I hope it can help others.

Thanks happynear for the port!

[happynear]

@flymer, thanks for the test report. I will add instructions in the ReadMe page.