DirectML and GPU Shared memory #19395
NarutoUA
started this conversation in
Ideas / Feature Requests
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Once onnxruntime-dml fully utilizes dedicated VRAM, it resorts to utilizing shared system memory, leading to a significant decrease in performance. I suggest introducing an execution provider option to toggle the utilization of system memory
Beta Was this translation helpful? Give feedback.
All reactions