search

nitishsrivastava / deepnet

Implementation of some deep learning algorithms.
BSD 3-Clause "New" or "Revised" License
893 stars 438 forks source link

Did this package work well for Mac OS 10.8.2 GeForce 660M GPU? #44

Open odieXin opened 10 years ago

odieXin commented 10 years ago

I got libcudamat.dylib and libcudamat_conv.dylib, then when I tried to test it with "python test_cudamat.py", the result is: Ran 0 tests in 0.000s OK

Then, I tried to run sh runall.sh in the rbm folder and finally, it showed:

No GPU board available.

Could someone help figure out the issue?

Thanks so much!!!

ansyral commented 10 years ago

also has the problem on ubuntu10.04 with the Geforce gtx650 graphic card.is it the reason of the graphic card ?

sanggeol commented 10 years ago

I am running this package on OS X 10.9.1 with GeForce GT 650M, but it requires a little bit of modification. As you might know, deepnet packages uses cudamat for gpu computing, and cudamat has a rather weird way of locking a gpu. ('cudamat/gnu_lock.py') It looks for /dev/nvidia-0 and create a symlink to it, which means that it will only work on Linux systems as macs don't have /dev/anything.

So I have an even uglier solution. Go to /deepnet/trainer.py, and modify LockGPU. Mine looks like this.

board = 0 cm.cuda_set_device(board) cm.cublas_init() return board

Basically I'm skipping gpu locking and telling it to use gpu number 0 (which is the only gpu installed in my case). This is NOT a real solution, but it works for now. I would appreciate if anyone could share a better solution.

danbri commented 10 years ago

I hit this issue too on a mac, and sanggeol's hack works for me.