Out of memory errors
Closed this issue · 5 comments
Hi sir, i just tried your extension today and it works. But it causes frequent out of memory errors, at least on DirectML, as that's what i use. The error shows "Cannot allocate tensor with size". Also the extension stays activated when bypassed or fully removed from the workflow. The only way to disable it, is to disable the whole extension.
Hi, can you send your workflow or steps to reproduce this? Will look into it
Could you provide some info about your hardware setup as well? I have only tested on MPS and CUDA but the node should work on any Comfy supported accelerator.
Hi sir, i amusing a ryzen 5600G processor. An RX 580 with 8GB of VRAM GPU and have 32GB of RAM.
SA_workflow.json
Hi sir, something really weird is goin on, the earlier workflow that i sent you was your sample workflow with the hypertile node attached which constantly gives memory errors. But when i use a workflow incorporating the efficient ksampler nodes and loader, i don't get memory errors, or infrequent. Also , now when i remove the stylealigned node from the workflow, it properly removes from the workflow, not giving any errors as if the node was still attached to the workflow. The only difference is that i had disabled the extension for a few days and enbled it again today, when i saw your replies. Just to do another quick check of its behavior. So that seems to have resolved itself.
Glad it's working, going to close out this issue then.