Float8_e4m3fn is only supported for ROCm 6.5 and above
- Dominant language
- Python
- Stars
- 133k
- Forks
- 15.7k
- Avg merge
- 1d 7h
- Merged PRs (30d)
- 158
Description
### Custom Node Testing
- [x] I have tried disabling custom nodes and the issue persists (see [how to disable custom nodes](https://docs.comfy.org/troubleshooting/custom-node-issues#step-1%3A-test-with-all-custom-nodes-disabled) if you need help)
### Expected Behavior
In previous versions of pytorch I could run an fp8 model on my 9070xt without issue, with reasonable perf.
### Actual Behavior
Today in [unrelated troubleshooting](https://github.com/comfyanonymous/ComfyUI/issues/10369) introduced this error. Took me a long time to figure out it wasn't a comfyui update (checking several previous commits) or ROCM (re-install various versions). But some kind of interplay with a recent pytorch update and comfyui.
Now when running an fp8 model, say wan 2.2 for example. I get this (extended logs added in the log section)
```
RuntimeError: Float8_e4m3fn is only supported for ROCm 6.5 and above
```
the ROCm 6.5 and above is especially weird. What is rocm 6.5? I had this issue on ROCM 7.0.1 btw.
Due to the previously linked issue I've downgraded back to rocm 6.4
### Steps to Reproduce
installing pytorch `2.9.0+rocm6.4` or later will cause this issue.
### Debug Logs
```powershell
FETCH ComfyRegistry Data: 40/100
loaded partially 6117.999804687501 6107.603401184082 300
0%| | 0/2 [00:00 master, origin/master, origin/HEAD)
Author: comfyanonymous <121283862+comfyanonymous@users.noreply.github.com>
Date: Fri Oct 17 17:03:28 2025 -0700
Disable torch compiler for cast_bias_weight function (#10384)
* Disable torch compiler for cast_bias_weight function
* Fix torch compile.
`
git log shows how up to date I am.
Apologies if this ends up being a torch issue I will gladly raise an issue over there.
```
Name: AMD Ryzen 9 7950X3D 16-Core Processor
Marketing Name: AMD Ryzen 9 7950X3D 16-Core Processor
Vendor Name: CPU
Name: gfx1201
Marketing Name: AMD Radeon RX 9070 XT
Vendor Name: AMD
Name: amdgcn-amd-amdhsa--gfx1201
Name: amdgcn-amd-amdhsa--gfx12-generic
Name: gfx1036
Marketing Name: AMD Radeon Graphics
Vendor Name: AMD
Name: amdgcn-amd-amdhsa--gfx1036
Name: amdgcn-amd-amdhsa--gfx10-3-generic
torch 2.8.0+rocm6.4
HIP runtime: 6.4.43482-0f2d60242
ROCm detected: True
Device: AMD Radeon RX 9070 XT
```
Contributor guide
Assessment
This issue has not been assessed yet.