github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #10241] Add support for ROCm 6.4 #32479

New Issue
Closed
opened 2026-04-22 13:47:16 -05:00 by GiteaMirror · 6 comments
GiteaMirror commented 2026-04-22 13:47:16 -05:00
Owner
Copy Link

Originally created by @codeliger on GitHub (Apr 11, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/10241

https://github.com/ROCm/ROCm/releases/tag/rocm-6.4.0#release-highlights

Originally created by @codeliger on GitHub (Apr 11, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/10241 https://github.com/ROCm/ROCm/releases/tag/rocm-6.4.0#release-highlights
GiteaMirror added the feature request label 2026-04-22 13:47:16 -05:00
GiteaMirror commented 2026-04-22 13:47:18 -05:00
Author
Owner
Copy Link

@rjmalagon commented on GitHub (Apr 12, 2025):

I have tested a containerized build with ROCm 6.4 (I currently use 6.3.4) on my branch for AMD APU. Although it builds nicely, it fails to load for GPU inferencing. I can't figure out what the problem is with it.

<!-- gh-comment-id:2799029125 --> @rjmalagon commented on GitHub (Apr 12, 2025): I have tested a containerized build with ROCm 6.4 (I currently use 6.3.4) on my branch for AMD APU. Although it builds nicely, it fails to load for GPU inferencing. I can't figure out what the problem is with it.
GiteaMirror commented 2026-04-22 13:47:18 -05:00
Author
Owner
Copy Link

@robertrosenbusch commented on GitHub (Apr 14, 2025):

@codeliger @rjmalagon : I guess its the very first time since long that GFX803-Owner on ROCm/Ollama are ahead :P

In this way, Ollama 0.6.5 works on ROCm 6.4 at the moment without any probs.

Cheers, Robert Rosenbusch

<!-- gh-comment-id:2801704956 --> @robertrosenbusch commented on GitHub (Apr 14, 2025): @codeliger @rjmalagon : I guess its the very first time since long that [GFX803-Owner on ROCm/Ollama](https://github.com/robertrosenbusch/gfx803_rocm/issues/16#issue-2993061609) are ahead :P In this way, Ollama 0.6.5 works on ROCm 6.4 at the moment without any probs. --- Cheers, Robert Rosenbusch
GiteaMirror commented 2026-04-22 13:47:19 -05:00
Author
Owner
Copy Link

@rjmalagon commented on GitHub (Apr 14, 2025):

@robertrosenbusch Yup, you are ahead (interesting gfx803 patches there). Actually, ROCm 6.4 introduces some interesting changes for integrated APUs (mostly for monsters like MI300A APUs).
Golang is "greek to me", but I suspect something minor on memory handling on Ollama crash on ROCm 6.4 for the minor APUs.
I will wait for more code from Ollama team (no precisely for ROCm 6.4, the very valid "compatible with old distros" mantra of the Ollama project move that support very slow). It is better than my butcher friendly code that this hand will produce.

<!-- gh-comment-id:2802057224 --> @rjmalagon commented on GitHub (Apr 14, 2025): @robertrosenbusch Yup, you are ahead (interesting gfx803 patches there). Actually, ROCm 6.4 introduces some interesting changes for integrated APUs (mostly for monsters like MI300A APUs). Golang is "greek to me", but I suspect something minor on memory handling on Ollama crash on ROCm 6.4 for the minor APUs. I will wait for more code from Ollama team (no precisely for ROCm 6.4, the very valid "compatible with old distros" mantra of the Ollama project move that support very slow). It is better than my butcher friendly code that this hand will produce.
GiteaMirror commented 2026-04-22 13:47:20 -05:00
Author
Owner
Copy Link

@codeliger commented on GitHub (Apr 14, 2025):

I currently have 9070 XT working with rocm 6.4 and ollama.

I would like to know if there is any expected performance improvement implementations available due to the support for new libraries.

gfx1201 now has support for:

  • Composable Kernel (1.1.0)
  • AMDMIGraphX (2.12.0)
  • rocBLAS (4.4.0) On gfx12, all functions now support full rocblas_int dynamic range for batch_count.
  • resolved: gfx12: ger, geam, geam_ex, dgmm, trmm, symm, hemm, ILP64 gemm, and larger data support.

It may need to be implemented in llama.cpp and then rebuilt/integrated with ollama.

<!-- gh-comment-id:2802182956 --> @codeliger commented on GitHub (Apr 14, 2025): I currently have 9070 XT working with rocm 6.4 and ollama. I would like to know if there is any expected performance improvement implementations available due to the support for new libraries. gfx1201 now has support for: - Composable Kernel (1.1.0) - AMDMIGraphX (2.12.0) - rocBLAS (4.4.0) On gfx12, all functions now support full rocblas_int dynamic range for batch_count. - resolved: gfx12: ger, geam, geam_ex, dgmm, trmm, symm, hemm, ILP64 gemm, and larger data support. It may need to be implemented in llama.cpp and then rebuilt/integrated with ollama.
GiteaMirror commented 2026-04-22 13:47:20 -05:00
Author
Owner
Copy Link

@robertrosenbusch commented on GitHub (Apr 14, 2025):

@rjmalagon: its just a hint, not a clue or advice for your prob. you are using ROCm 6.4 and Ollama into a docker container. right?

it could make sense to check out the rocBLAS 4.4 for ROCm 6.4 fresh from GIT and (re-)compile and (re-)install it into your docker with your specific GFX-Version on your APU.

In this way, it wouldnt be a prob on Ollama or llama.cpp ^·^ Let me know if i was wrong or right. I will be appriciate it this information.

Cheers, Robert Rosenbusch

<!-- gh-comment-id:2802400292 --> @robertrosenbusch commented on GitHub (Apr 14, 2025): @rjmalagon: its just a hint, not a clue or advice for your prob. you are using ROCm 6.4 and Ollama into a docker container. right? it could make sense to check out the [rocBLAS 4.4 for ROCm 6.4](https://github.com/ROCm/rocBLAS/releases/tag/rocm-6.4.0) fresh from GIT and (re-)compile and (re-)install it into your docker with your specific GFX-Version on your APU. In this way, it wouldnt be a prob on Ollama or llama.cpp ^·^ Let me know if i was wrong or right. I will be appriciate it this information. --- Cheers, Robert Rosenbusch
GiteaMirror commented 2026-04-22 13:47:21 -05:00
Author
Owner
Copy Link

@garrettbyrd commented on GitHub (Apr 14, 2025):

@rjmalagon is your issue on APUs related to #8735?

<!-- gh-comment-id:2802416974 --> @garrettbyrd commented on GitHub (Apr 14, 2025): @rjmalagon is your issue on APUs related to #8735?
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label feature request
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#32479
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?