Hello all, great work with pytorch. I'm trying to use pytorch in a Jetson TX1, and it sort of works. One issue I'm running into is that it sporadically hangs when using CUDA.

Some details:

• Jetson TX1, a 64-bit ARM architecture with a tegra TX1
• Ubuntu 16.04 aarch64
• CUDA 8.0, CUDNN 5105
• Python 2.7
• Built from source (conda not available yet) from git commit 7c24a (Jun 23) and from tag 1.12. Both have the same issue.

Example program:

import torch
a = torch.FloatTensor(2).cuda()
print(a)

This code will work maybe 8/10 times (subjective belief, not actually measured) and the rest, it will hang on the second line. top indicates the program is using ~18% CPU. The program will not respond to Ctrl-C. Suspecting some kind of concurrency issue, I tried OMP_NUM_THREADS=1, and torch.set_num_threads(1) but the issue still occurs.

When not using .cuda() it never hangs up (and in fact starts relatively quickly compared to the cuda version -- the .cuda() call seems surprisingly slow).