I tried to install and use pytorch with a docker container on a gpu server.
When I try to use .cuda() method, it failed with the message:
b = a.cuda()
Traceback (most recent call last):
File “”, line 1, in
File “/root/anaconda3/lib/python3.5/site-packages/torch/_utils.py”, line 51, in _cuda
return self.type(getattr(torch.cuda, self.class.name), async)
File “/root/anaconda3/lib/python3.5/site-packages/torch/_utils.py”, line 24, in type
return new_type(self.size()).copy(self, async)
File “/root/anaconda3/lib/python3.5/site-packages/torch/cuda/init.py”, line 254, in new
_lazy_init()
File “/root/anaconda3/lib/python3.5/site-packages/torch/cuda/init.py”, line 94, in _lazy_init
_check_driver()
File “/root/anaconda3/lib/python3.5/site-packages/torch/cuda/init.py”, line 77, in _check_driver
of the CUDA driver.“”".format(str(torch._C._cuda_getDriverVersion())))
AssertionError:
The NVIDIA driver on your system is too old (found version 8000).
Please update your GPU driver by downloading and installing a new
version from the URL: Official Drivers | NVIDIA
Alternatively, go to: https://pytorch.org/binaries to install
a PyTorch version that has been compiled with your version
of the CUDA driver.
I installed the pytorch with conda and built with the github source as well.
I use ubuntu 14.04, GTX 1080 with driver version of 367.35, CUDA 8.0.44.
Is there any way to run pytorch without updating the driver?
I think 367.35 is enough for CUDA 8 and it worked with torch.