"cupy.cuda.runtime.CUDARuntimeError" on running with any GPU

[pcgrosen]

Upon running python train.py, I receive this:

data/tinyshakespeare/input.txt
corpus length: 1115394
vocab size: 65
Traceback (most recent call last):
  File "train.py", line 67, in <module>
    cuda.get_device(args.gpu).use()
  File "cupy/cuda/device.pyx", line 73, in cupy.cuda.device.Device.use (cupy/cuda/device.cpp:1804)
  File "cupy/cuda/device.pyx", line 79, in cupy.cuda.device.Device.use (cupy/cuda/device.cpp:1753)
  File "cupy/cuda/runtime.pyx", line 147, in cupy.cuda.runtime.setDevice (cupy/cuda/runtime.cpp:2422)
  File "cupy/cuda/runtime.pyx", line 106, in cupy.cuda.runtime.check_status (cupy/cuda/runtime.cpp:1815)
cupy.cuda.runtime.CUDARuntimeError: cudaErrorUnknown: unknown error

This is a fresh installation of Ubuntu 15.04; I downloaded it for this purpose. Chainer was installed simply with pip install chainer, and I ran the Cuda installer with defaults.

Any help would be greatly appreciated. Thanks!

[deependersingla]

@theradical any update on this? Getting same error.

[yusuketomoto]

it seems gpu is not available on chainer. i changed code to use cpu by default. try again please.

[deependersingla]

For me CPU is working right from start, @yusuketomoto so the GPU doesn't work for anyone?

[yusuketomoto]

gpu is working for me.

try pip install chainer --no-cache after installing CUDA and setting path. following post might be helpful. https://github.com/pfnet/chainer https://goo.gl/sYx4MN

after install, you should try running chainer/examples/mnist/train_mnist.py on gpu to check if gpu works.

[xzyaoi]

same question here. any solutions?

[honnibal]

It's stupid, but have you tried turning it off and on again? Sometimes this happens to me and I simply have to reboot the server. I'm not sure what process needs to be restarted, but it doesn't seem to be anything to do with Chainer or cupy.