6 Commits

Author	SHA1	Message	Date
Jedrzej Kosinski	4b93c4360f	Implement persistent thread pool for multi-GPU CFG splitting (#13329) ... Replace per-step thread create/destroy in _calc_cond_batch_multigpu with a persistent MultiGPUThreadPool. Each worker thread calls torch.cuda.set_device() once at startup, preserving compiled kernel caches across diffusion steps. - Add MultiGPUThreadPool class in comfy/multigpu.py - Create pool in CFGGuider.outer_sample(), shut down in finally block - Main thread handles its own device batch directly for zero overhead - Falls back to sequential execution if no pool is available	2026-04-08 05:39:07 -07:00
Jedrzej Kosinski	407a5a656f	Rollback core of last commit due to weird behavior	2025-03-28 02:48:11 -05:00
kosinkadink1@gmail.com	9ce9ff8ef8	Allow chained MultiGPU Work Unit nodes to affect max_gpus present on ModelPatcher clone	2025-03-28 15:29:44 +08:00
Jedrzej Kosinski	6dca17bd2d	Satisfy ruff linting	2025-03-03 23:08:29 -06:00
Jedrzej Kosinski	093914a247	Made MultiGPU Work Units node more robust by forcing ModelPatcher clones to match at sample time, reuse loaded MultiGPU clones, finalize MultiGPU Work Units node ID and name, small refactors/cleanup of logging and multigpu-related code	2025-03-03 22:56:13 -06:00
Jedrzej Kosinski	eda866bf51	Extracted multigpu core code into multigpu.py, added load_balance_devices to get subdivision of work based on available devices and splittable work item count, added MultiGPU Options nodes to set relative_speed of specific devices; does not change behavior yet	2025-01-27 06:25:48 -06:00