wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2026-07-21 23:41:28 +08:00
Code Issues Actions 10 Packages Projects Releases Wiki Activity

Threaded Loader performance fixes / improvements (+ Aimdo 0.4.6) (#14116)

Browse Source 
* memory_management: Add direct to read GPU mode

Make destination optional (or make it optionally GPU) and use aimdo
to file_read direct to GPU.

* ops: Remove stream pin buffers and use aimdo reads

This consumed too much RAM and its better to just take the hit on
the CPU syncing back the stream on a short ring buffer. Aimdo
implements this so just rip the stream pin buffer from comfy.

* model_management: all active pin registration movement

Its better to just let the active model load past the pin limit as
pins and let the pins move around. The saves the HDD and SATA
people disk traffic while only costing a few GPU syncs.

* utils: use aimdo file handle

This opens on windows with more favourable flags

* mp: only count the model proper for loaded_ram and vram

Exclude live loras from the numbers to avoid the case where the reported
loaded memory exceeds the size of the model.

This causes me confusion in the Kijai visualizer when it looked fully
loaded but was hitting disk due to this accounding disrepency.

* utils: add bit reverse utility

useful for max scattering something ordered.

* pinned_memory: Implement offload balancing

Use a max scatter alogorithm to prioritize pins of the same size such
that when doing a little bit of offloading it gets scattered, allowing
the prefetcher to more evenly swollow the offload.

* comfy-aimdo 0.4.7

Aimdo 0.4.7 implement VRAM buffer exhaustion predection to avoid
early speculative load of weights that definately wont fix once the
inference gets further in.

* model-prefetch: consolidate pin ensures on the sync point

This could happen mid prefetch block, cause a sync of the entire
block and lose overlap. Get ahead of the problem with a free down
at the natural compute stream sync point.

* mm: Put a 2GB min on the pin ceiling

This is reasonably bad if it starts causing swap pressure, moreso than
during normal ram-cache proceedings. Clamp it.

* add --fast-disk
This commit is contained in:
rattus
2026-05-30 15:20:04 -04:00
committed by GitHub
parent bb560036b9
commit e154da83b1
9 changed files with 149 additions and 132 deletions
Download Patch File Download Diff File Expand all files Collapse all files
comfy/cli_args.py
+1
View File
@@ -149,6 +149,7 @@ parser.add_argument("--async-offload", nargs='?', const=2, type=int, default=Non
parser.add_argument("--disable-async-offload", action="store_true", help="Disable async weight offloading.")
parser.add_argument("--disable-dynamic-vram", action="store_true", help="Disable dynamic VRAM and use estimate based model loading.")
parser.add_argument("--enable-dynamic-vram", action="store_true", help="Enable dynamic VRAM on systems where it's not enabled by default.")
parser.add_argument("--fast-disk", action="store_true", help="Prefer disk-backed dynamic loading and offload over unpinned RAM. Can be faster for users with fast NVME disks.")
parser.add_argument("--force-non-blocking", action="store_true", help="Force ComfyUI to use non-blocking operations for all applicable tensors. This may improve performance on some non-Nvidia systems but can cause issues with some workflows.")