VRAM Leak on AMD R9700 with ComfyUI: A ROCm Troubleshooting Log
BmoreSpinner · reddit · 2026-08-05
A user running ComfyUI on Debian with an AMD Radeon AI PRO R9700 (32GB) and ROCm 7.2 encountered a severe VRAM retention issue.
- The Issue: After generating images/videos using SD 3.5 or LTX 2.3, VRAM usage remains high at around 22.7 GB. Calling ComfyUI's /free endpoint, adding various --disable flags, and configuring PyTorch's PYTORCHALLOCCONF garbage collection thresholds failed to release the memory.
- Diagnostics: A separate Python process confirmed that PyTorch allocated and reserved memory was 0.00 GB, indicating the memory isn't held by PyTorch caching. The only reliable workaround currently is to restart the entire ComfyUI Docker container.
- Next Steps: The user shared detailed system configurations and troubleshooting steps, seeking community help to resolve this underlying memory management conflict.
Related event: Troubleshooting ComfyUI VRAM Leaks on AMD R9700(2 posts)→
More from Infra
- Building a Trusted Compute Cluster: Infrastructure for Safe Frontier AI Evaluation — ohlennart · 2026-08-06
- Running MiniMax H3 on Dual RTX 5060 Ti: VRAM Splitting and Unloading Strategies — Kahvana · 2026-08-06
- Running MiniMax H3 on AMD GPUs Hits a Wall with SageAttention Dependency — itiswhatitiswgatitis · 2026-08-06
- Teaser: Run Any Model via One API Key with Local-Cloud Routing — ycombinator · 2026-08-06
- Inside turbopuffer: How to Build a 256TB Search Index — shakoistsLog · 2026-08-06
- GlobalFoundries Q3 Margin Guides Above 30% as Silicon Photonics Capacity Becomes Scarce — tengyanAI · 2026-08-06