github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #12071] Windows: AMD gfx1103 (Radeon 780M / RX 7700S) not accepted by Ollama GPU backend (gfx1103) #70077

New Issue
Open
opened 2026-05-04 20:16:23 -05:00 by GiteaMirror · 9 comments
GiteaMirror commented 2026-05-04 20:16:23 -05:00
Owner
Copy Link

Originally created by @henrylevesque on GitHub (Aug 25, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/12071

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

  • Reproduction summary: Ollama v0.11.6 on Windows reports "amdgpu is not supported (supported types:[])" for gfx1103 (Radeon 780M) and gfx1102 (RX 7700S) and falls back to CPU.
  • Tested: HSA_OVERRIDE_GFX_VERSION=gfx1102 and gfx1103; duplicated rocm kernel filenames; reinstalled Windows AMD drivers.
  • Current behavior: server.log shows "amdgpu is not supported (supported types:[])" and "no compatible GPUs were discovered"; Ollama runs on CPU.

Relevant log output

time=... level=INFO msg="Listening on 127.0.0.1:11434 (version 0.11.6)"
time=... level=INFO msg="looking for compatible GPUs"
time=... level=INFO msg="skipping rocm gfx compatibility check" HSA_OVERRIDE_GFX_VERSION=gfx1102 (or gfx1103)
time=... level=WARN msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1103 gpu=0 library=C:\Users\...\lib\ollama\rocm
time=... level=WARN msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1102 gpu=1 library=C:\Users\...\lib\ollama\rocm
time=... level=INFO msg="no compatible GPUs were discovered"

OS

windows

GPU

GPU 0: gfx1103 — AMD Radeon 780M (reported as gfx1103 in Ollama logs)
GPU 1: gfx1102 — AMD Radeon RX 7700S (reported as gfx1102 in Ollama logs)

CPU

CPU: AMD Ryzen 7 7840HS w/ Radeon 780M Graphics

Ollama version

0.11.6

Originally created by @henrylevesque on GitHub (Aug 25, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/12071 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? - Reproduction summary: Ollama v0.11.6 on Windows reports "amdgpu is not supported (supported types:[])" for gfx1103 (Radeon 780M) and gfx1102 (RX 7700S) and falls back to CPU. - Tested: HSA_OVERRIDE_GFX_VERSION=gfx1102 and gfx1103; duplicated rocm kernel filenames; reinstalled Windows AMD drivers. - Current behavior: server.log shows "amdgpu is not supported (supported types:[])" and "no compatible GPUs were discovered"; Ollama runs on CPU. ### Relevant log output ```shell time=... level=INFO msg="Listening on 127.0.0.1:11434 (version 0.11.6)" time=... level=INFO msg="looking for compatible GPUs" time=... level=INFO msg="skipping rocm gfx compatibility check" HSA_OVERRIDE_GFX_VERSION=gfx1102 (or gfx1103) time=... level=WARN msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1103 gpu=0 library=C:\Users\...\lib\ollama\rocm time=... level=WARN msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1102 gpu=1 library=C:\Users\...\lib\ollama\rocm time=... level=INFO msg="no compatible GPUs were discovered" ``` ### OS windows ### GPU GPU 0: gfx1103 — AMD Radeon 780M (reported as gfx1103 in Ollama logs) GPU 1: gfx1102 — AMD Radeon RX 7700S (reported as gfx1102 in Ollama logs) ### CPU CPU: AMD Ryzen 7 7840HS w/ Radeon 780M Graphics ### Ollama version 0.11.6
GiteaMirror added the amdbugwindows labels 2026-05-04 20:16:24 -05:00
GiteaMirror commented 2026-05-04 20:16:26 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 25, 2025):

HSA_OVERRIDE_GFX_VERSION=11.0.3

But since you have multiple devices, you likely want

HSA_OVERRIDE_GFX_VERSION_0=11.0.3
HSA_OVERRIDE_GFX_VERSION_1=11.0.2
<!-- gh-comment-id:3220467885 --> @rick-github commented on GitHub (Aug 25, 2025): `HSA_OVERRIDE_GFX_VERSION=11.0.3` But since you have multiple devices, you likely want ``` HSA_OVERRIDE_GFX_VERSION_0=11.0.3 HSA_OVERRIDE_GFX_VERSION_1=11.0.2 ```
GiteaMirror commented 2026-05-04 20:16:26 -05:00
Author
Owner
Copy Link

@henrylevesque commented on GitHub (Aug 25, 2025):

Tried overriding the gfx version, which ollama picks up, but since the compatible gpu types list is empty in my installation, it still lists my gpus as incompatible and runs on cpu only.

Log output:

skipping rocm gfx compatibility check HSA_OVERRIDE_GFX_VERSION=gfx1102
time=... level=WARN ... msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1103 gpu=0 library=...\rocm
time=... level=WARN ... msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1102 gpu=1 library=...\rocm
time=... level=INFO ... msg="no compatible GPUs were discovered"
time=... level=INFO ... msg="inference compute" id=0 library=cpu

<!-- gh-comment-id:3220878907 --> @henrylevesque commented on GitHub (Aug 25, 2025): Tried overriding the gfx version, which ollama picks up, but since the compatible gpu types list is empty in my installation, it still lists my gpus as incompatible and runs on cpu only. Log output: skipping rocm gfx compatibility check HSA_OVERRIDE_GFX_VERSION=gfx1102 time=... level=WARN ... msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1103 gpu=0 library=...\rocm time=... level=WARN ... msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1102 gpu=1 library=...\rocm time=... level=INFO ... msg="no compatible GPUs were discovered" time=... level=INFO ... msg="inference compute" id=0 library=cpu
GiteaMirror commented 2026-05-04 20:16:26 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 25, 2025):

skipping rocm gfx compatibility check HSA_OVERRIDE_GFX_VERSION=gfx1102

Why are you setting HSA_OVERRIDE_GFX_VERSION=gfx1102?

<!-- gh-comment-id:3220912467 --> @rick-github commented on GitHub (Aug 25, 2025): ``` skipping rocm gfx compatibility check HSA_OVERRIDE_GFX_VERSION=gfx1102 ``` Why are you setting `HSA_OVERRIDE_GFX_VERSION=gfx1102`?
GiteaMirror commented 2026-05-04 20:16:27 -05:00
Author
Owner
Copy Link

@henrylevesque commented on GitHub (Aug 25, 2025):

I set the HSA override only as a diagnostic step to force the Windows ROCm/hip path to treat the devices as known gfx IDs.

What I did

I have two AMD GPUs: gfx1103 (Radeon 780M) and gfx1102 (RX 7700S).
I set per-device overrides in PowerShell (session only): $env:HSA_OVERRIDE_GFX_VERSION_0='11.0.3' $env:HSA_OVERRIDE_GFX_VERSION_1='11.0.2' (the tail / logs also show HSA_OVERRIDE_GFX_VERSION=gfx1102)
Then I ran: & 'C:\Users\leves\AppData\Local\Programs\Ollama\ollama.exe' serve

What I observed

Ollama sees the override (log: "skipping rocm gfx compatibility check HSA_OVERRIDE_GFX_VERSION=gfx1102").
Despite that, each GPU is WARNed as unsupported with an empty supported-types array:
"amdgpu is not supported (supported types:[])" gpu_type=gfx1103 gpu=0
"amdgpu is not supported (supported types:[])" gpu_type=gfx1102 gpu=1
Result: "no compatible GPUs were discovered" and inference falls back to CPU.

Why I tried the override

I used the override to test whether forcing the gfx version would make the packaged ROCm kernels / ggml-hip accept the devices. The override is a low-risk way to test whether the runtime will accept those gfx IDs without changing drivers or packaging.

<!-- gh-comment-id:3221101642 --> @henrylevesque commented on GitHub (Aug 25, 2025): I set the HSA override only as a diagnostic step to force the Windows ROCm/hip path to treat the devices as known gfx IDs. ### What I did I have two AMD GPUs: gfx1103 (Radeon 780M) and gfx1102 (RX 7700S). I set per-device overrides in PowerShell (session only): $env:HSA_OVERRIDE_GFX_VERSION_0='11.0.3' $env:HSA_OVERRIDE_GFX_VERSION_1='11.0.2' (the tail / logs also show HSA_OVERRIDE_GFX_VERSION=gfx1102) Then I ran: & 'C:\Users\leves\AppData\Local\Programs\Ollama\ollama.exe' serve ### What I observed Ollama sees the override (log: "skipping rocm gfx compatibility check HSA_OVERRIDE_GFX_VERSION=gfx1102"). Despite that, each GPU is WARNed as unsupported with an empty supported-types array: "amdgpu is not supported (supported types:[])" gpu_type=gfx1103 gpu=0 "amdgpu is not supported (supported types:[])" gpu_type=gfx1102 gpu=1 Result: "no compatible GPUs were discovered" and inference falls back to CPU. ### Why I tried the override I used the override to test whether forcing the gfx version would make the packaged ROCm kernels / ggml-hip accept the devices. The override is a low-risk way to test whether the runtime will accept those gfx IDs without changing drivers or packaging.
GiteaMirror commented 2026-05-04 20:16:27 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 25, 2025):

HSA_OVERRIDE_GFX_VERSION=gfx1102 is the wrong format. HSA_OVERRIDE_GFX_VERSION=11.0.3.

<!-- gh-comment-id:3221198204 --> @rick-github commented on GitHub (Aug 25, 2025): `HSA_OVERRIDE_GFX_VERSION=gfx1102` is the wrong format. `HSA_OVERRIDE_GFX_VERSION=11.0.3`.
GiteaMirror commented 2026-05-04 20:16:27 -05:00
Author
Owner
Copy Link

@henrylevesque commented on GitHub (Aug 25, 2025):

I tried that and the HSA numeric override is detected (log shows skipping compatibility check), but the runtime still reports supported types = [] and rejects gfx1103/gfx1102 — so override alone does not make the packaged runtime accept the device

<!-- gh-comment-id:3221310186 --> @henrylevesque commented on GitHub (Aug 25, 2025): I tried that and the HSA numeric override is detected (log shows skipping compatibility check), but the runtime still reports supported types = [] and rejects gfx1103/gfx1102 — so override alone does not make the packaged runtime accept the device
GiteaMirror commented 2026-05-04 20:16:27 -05:00
Author
Owner
Copy Link

@rperezdesiles commented on GitHub (Aug 28, 2025):

Same problem here with gfx1103 (AMD 780M) on Windows...

time=2025-08-28T22:28:57.918+02:00 level=DEBUG source=gpu.go:560 msg="discovered GPU libraries" paths=[]
time=2025-08-28T22:28:57.986+02:00 level=DEBUG source=amd_hip_windows.go:88 msg=hipDriverGetVersion version=60241512
time=2025-08-28T22:28:57.987+02:00 level=DEBUG source=amd_common.go:16 msg="evaluating potential rocm lib dir C:\Users\Rafael\AppData\Local\Programs\Ollama\lib\ollama\rocm"
time=2025-08-28T22:28:57.987+02:00 level=DEBUG source=amd_common.go:44 msg="detected ROCM next to ollama executable C:\Users\Rafael\AppData\Local\Programs\Ollama\lib\ollama\rocm"
time=2025-08-28T22:28:57.987+02:00 level=INFO source=amd_windows.go:70 msg="skipping rocm gfx compatibility check" HSA_OVERRIDE_GFX_VERSION=11.0.3
time=2025-08-28T22:28:57.987+02:00 level=DEBUG source=amd_windows.go:73 msg="detected hip devices" count=1
time=2025-08-28T22:28:57.988+02:00 level=DEBUG source=amd_windows.go:93 msg="hip device" id=0 name="AMD Radeon 780M Graphics" gfx=gfx1103
time=2025-08-28T22:28:58.654+02:00 level=WARN source=amd_windows.go:138 msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1103 gpu=0 library=C:\Users\Rafael\AppData\Local\Programs\Ollama\lib\ollama\rocm
time=2025-08-28T22:28:58.656+02:00 level=INFO source=gpu.go:379 msg="no compatible GPUs were discovered"
time=2025-08-28T22:28:58.656+02:00 level=INFO source=types.go:130 msg="inference compute" id=0 library=cpu variant="" compute="" driver=0.0 name="" total="15.8 GiB" available="9.3 GiB"
time=2025-08-28T22:28:58.656+02:00 level=INFO source=routes.go:1412 msg="entering low vram mode" "total vram"="15.8 GiB" threshold="20.0 GiB"

<!-- gh-comment-id:3234874008 --> @rperezdesiles commented on GitHub (Aug 28, 2025): Same problem here with gfx1103 (AMD 780M) on Windows... time=2025-08-28T22:28:57.918+02:00 level=DEBUG source=gpu.go:560 msg="discovered GPU libraries" paths=[] time=2025-08-28T22:28:57.986+02:00 level=DEBUG source=amd_hip_windows.go:88 msg=hipDriverGetVersion version=60241512 time=2025-08-28T22:28:57.987+02:00 level=DEBUG source=amd_common.go:16 msg="evaluating potential rocm lib dir C:\\Users\\Rafael\\AppData\\Local\\Programs\\Ollama\\lib\\ollama\\rocm" time=2025-08-28T22:28:57.987+02:00 level=DEBUG source=amd_common.go:44 msg="detected ROCM next to ollama executable C:\\Users\\Rafael\\AppData\\Local\\Programs\\Ollama\\lib\\ollama\\rocm" time=2025-08-28T22:28:57.987+02:00 level=INFO source=amd_windows.go:70 msg="skipping rocm gfx compatibility check" HSA_OVERRIDE_GFX_VERSION=11.0.3 time=2025-08-28T22:28:57.987+02:00 level=DEBUG source=amd_windows.go:73 msg="detected hip devices" count=1 time=2025-08-28T22:28:57.988+02:00 level=DEBUG source=amd_windows.go:93 msg="hip device" id=0 name="AMD Radeon 780M Graphics" gfx=gfx1103 time=2025-08-28T22:28:58.654+02:00 level=WARN source=amd_windows.go:138 msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1103 gpu=0 library=C:\Users\Rafael\AppData\Local\Programs\Ollama\lib\ollama\rocm time=2025-08-28T22:28:58.656+02:00 level=INFO source=gpu.go:379 msg="no compatible GPUs were discovered" time=2025-08-28T22:28:58.656+02:00 level=INFO source=types.go:130 msg="inference compute" id=0 library=cpu variant="" compute="" driver=0.0 name="" total="15.8 GiB" available="9.3 GiB" time=2025-08-28T22:28:58.656+02:00 level=INFO source=routes.go:1412 msg="entering low vram mode" "total vram"="15.8 GiB" threshold="20.0 GiB"
GiteaMirror commented 2026-05-04 20:16:28 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 28, 2025):

780M is not currently supported. https://github.com/ollama/ollama/issues/7861

<!-- gh-comment-id:3235000295 --> @rick-github commented on GitHub (Aug 28, 2025): 780M is not currently supported. https://github.com/ollama/ollama/issues/7861
GiteaMirror commented 2026-05-04 20:16:28 -05:00
Author
Owner
Copy Link

@z3rg commented on GitHub (Dec 6, 2025):

here for my build for windows https://github.com/z3rg/ollama-AMD-Radeon-780M

Image
<!-- gh-comment-id:3621362852 --> @z3rg commented on GitHub (Dec 6, 2025): here for my build for windows https://github.com/z3rg/ollama-AMD-Radeon-780M <img width="1680" height="152" alt="Image" src="https://github.com/user-attachments/assets/7bf360d5-40f7-4419-b0e2-382f7849a470" />
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 amd bug windows
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#70077
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?