Making ComfyUI's INT8 Faster on ROCm
Itchy_Push_3077 · reddit · 2026-07-19
This is a highly specific ComfyUI / AMD ROCm performance optimization guide, aiming to make Krea2 Turbo's INT8 ConvRot run faster on AMD cards.
The author first compares three paths:
- FP8: Stable, warm start around 12.74s
- INT8 ConvRot safe fallback: Stable but very slow, around 65–80s
- Global Triton: Attempts a fast path, but crashes on ROCm
Instead of enabling Global Triton, the author only switches the int8linear path of ConvRot to a Triton implementation, leaving the rest on the stable path. This avoids crashes caused by global enabling while retaining the acceleration of critical operators.
The results:
- Warm start improved from FP8's 12.74s to 10.49–10.54s
- About 6–7.6x speedup compared to the original INT8 fallback path
This type of post is highly valuable for local AMD/ROCm deployment users.
More from Infra
- Cloud capacity crisis is now an AI architecture problem, argues new newsletter — DavidLinthicum · 2026-10-26
- SpaceX files $40B GPU financing commitment with Apollo, already at 1.7GW — XFreeze · 2026-10-26
- Meta plans to raise up to $95B to fund massive AI expansion — Polymarket · 2026-10-26
- Nebius CEO: Debt Markets Closed for Neoclouds Last Week, Leaning on Customer Advances — zephyr_z9 · 2026-10-26
- CoreWeave 5-year CDS spikes to ~1,100bps as market fears AI bubble starts as credit event — kimmonismus · 2026-10-26
- AI demand reshapes cloud computing: scarce capacity and long-term deals replace on-demand — DavidLinthicum · 2026-10-26