Does this code support multi-gpu running?
Opened this issue · 2 comments
Yang-wk commented
Hi, I tried to use multi-gpu to run this code by adding -gpu 0,1, and got this error:
F0628 16:24:14.008252 39462 syncedmem.hpp:31] Check failed: error == cudaSuccess (77 vs. 0) an illegal memory access was encountered.
Is there something going wrong?
leonzfa commented
@Yang-wk Hi, I test the code using "-gpu 0, 1", and it can run without error. But the runtime is twice of that uses only one GPU (batch size is set to 2 in both cases). It seems that there is no acceleration. Maybe the problem comes from the data augmentation layer, which is provided by DispNetC. I still haven't run the code using multi-gpu correctly. If I solve the problem, I will inform you. I'm sorry.