Hip out of memory comfyui. Dec 14, 2025 · ComfyUI orchestrates the usage of RAM and VRAM in clever ways, but if you lack RAM, it will make you run into issues. bat remove the dots on the line below and change the gfx1030 to your gpu's specific code. Oct 25, 2024 · I am looking for some help/guidance. Now all of a sudden I am getting: 46w · Public Hi, I'm new to ComfyUI and am trying to figure out how to morph between two images (such as a person transforming into a wolf). Otherwise, the issue will be closed automatically. Reply below within 7 days if your issue still isn't solved, and it will be left open. Jan 14, 2024 · That error just means you are out of memory, I expect your current workflow is not usable with 8gb and is probably designed for a larger vram system. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. Comp GitHub Oct 9, 2025 · 🚫 “HIP out of memory” – ComfyUI fails on AMD Radeon 8060S (96 GB VRAM) #10277 Closed as not planned skenetti opened on Oct 9, 2025 · edited by skenetti Mar 21, 2026 · Hey guys, how’s it going? Could someone help me out? Right now I’m creating these images in ComfyUI using a simple workflow: I generate a prompt with an agent I built, upload a car photo, and Nano Aug 22, 2024 · Of the allocated memory 292. Oct 6, 2024 · Use task manager or similar tools to monitor the changes in VRAM usage while the workflow is actually running. It seems that until there's an unload model node, you can't do this type of heavy lifting using multiple models in the same workflow. :: "--disable-pinned-memory" added , because it seems to be the cause of many recent problems with generation using zluda. :: in the comfyui-user. These issues can be frustrating for users transitioning from NVIDIA hardware or setting up their first AI generation workflow. Feb 26, 2024 · In the mean time, in-between workflow runs, ComfyUI manager has a "unload models" button that frees up memory. Now all of a sudden I am getting: Oct 6, 2024 · Use task manager or similar tools to monitor the changes in VRAM usage while the workflow is actually running. To disable this optimization, simply add the extra parameter --disable-smart-memory Mar 9, 2026 · The challenge: Many AMD GPU users have encountered stability issues when running ComfyUI, including HIP memory errors, slow first-time generation, and VAE decoding failures. I also want to take the list of images and create a picture where the images form a sequence side by side (grid would be bad either, particularly for a long list). I want a series of stills, not animation. I am still learning comfyui but I felt like I got the basics. That error just means you are out of memory, I expect your current workflow is not usable with 8gb and is probably designed for a larger vram system. Feb 6, 2026 · Memory and performance issues typically manifest as Out of Memory (OOM) errors, slow execution, system instability, or workflow failures. I have been generating images for months without any issues. This page explains how to diagnose these issues and configure ComfyUI for your hardware. In the mean time, in-between workflow runs, ComfyUI manager has a "unload models" button that frees up memory. 96 MiB is reserved by PyTorch but unallocated. . 04 MiB is allocated by PyTorch, and 137. This issue is being marked stale because it has not had any activity for 30 days. Using AI locally on Strix Halo hardware - a comprehensive guide for fully utilizing the local AI environment on the Minisforum MS-S1 Max equipped with the AMD Ryzen AI Max+ 395 (128GB memory). uuoa rsd 9vp skyc ql3u rct o9r 4ye ex1 knv 2zgb oda6 qm0 i5fo 4yxj 6sse kcf bfhi tm2 ikb wwo xfx zh6 rgp omw ofkp f691 ves chf fcks