iperov
SAEHD: Not compatible with high core count cpus
Expected behavior
SAEHD trains when you run the file [6) train SAEHD.bat]
Actual behavior
cv2.error: OpenCV(4.1.0) C:\projects\opencv-python\opencv\modules\core\src\alloc.cpp:55: error: (-4:Insufficient memory) Failed to allocate 1048576 bytes in function 'cv::OutOfMemoryError'
my ram and vram barely are being utilised, with about 20GB free of both
Steps to reproduce
Ryzen 5950x, run SAEHD with full core count.
Other relevant information
Specs: Ryzen 5950x, RTX 3090, 32GB Memory
running file [6) train SAEHD.bat] from DeepFaceLab_NVIDIA_RTX3000_series 21/10/2021 (latest release) using prebuilt windows binary
When I switch out my ryzen 9 5950x to my previous cpu, ryzen 5 3600x I have no issue. My 5950x is stable and works fine with all other programs.
If i manually restrict cpu count through windows msconfig, it works fine (although i don't want to have to do this)
quick96 has no issue when using all 32 logical cores
Try to increase paging file.
Already done. set a custom size of 200GB (Massive!) to be absolutely sure and it's still showing the error
You might want to try setting paging file up to 700 GB. Thats what I did and we have, kind of, similar specs: Ryzen 3950x, RTX 3080 Ti, 64 GB Memory. However, it's not about the core count.
Did you ever find the answer? If so, would you mind sharing it and closing this issue?
Same issue with high core count (ryzen 9 5950x). When I switch to ryzen 5 3600x works fine. The only "Fix" i found is running my 5950x with 1ccx (8 cores 16 threads )
