You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When I use a memory-intensive layers like embedding table of DLRM, I face with CUDA out of memory issue. So I came up with model-parallel which may be a solution for this issue.
But it looks like not a fundamental solution of this issue. So can I use the DRAM from CPU side as GPU memory and we only store the most frequently use embedding vectors to GPU?
The text was updated successfully, but these errors were encountered:
When I use a memory-intensive layers like embedding table of DLRM, I face with CUDA out of memory issue. So I came up with model-parallel which may be a solution for this issue.
But it looks like not a fundamental solution of this issue. So can I use the DRAM from CPU side as GPU memory and we only store the most frequently use embedding vectors to GPU?
The text was updated successfully, but these errors were encountered: