github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-21 13:11:39 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #10516] No response from Ollama although GPU is at 100% #53431

New Issue
Open
opened 2026-04-29 03:08:36 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-29 03:08:36 -05:00
Owner
Copy Link

Originally created by @slashgob on GitHub (May 1, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/10516

What is the issue?

Trying to run a small model (llama3.2 2gb), seems to load fine but I get no response from ollama. Using btop I can see GPU is at 100% usage, but still no response from chat message. It just stays at 100% usage.

Nobara OS (Fedora spin)
CPU: AMD 5950x
Ram: 32GiB
GPU: AMD 6900 XT

Installed latest Ollama with curl script.

ROCm installed...

─❯ dnf list --installed | grep 'rocm'
hipcc.x86_64 18-10.rocm6.2.1.fc41 nobara
hsakmt.x86_64 1.0.6-46.rocm6.2.1.fc41 nobara
hsakmt-devel.x86_64 1.0.6-46.rocm6.2.1.fc41 nobara
python3-torch-rocm-gfx9.x86_64 2.4.0-10.fc41 nobara
python3-torchaudio-rocm-gfx9.x86_64 2.4.1-2.fc41 nobara
rocm-clinfo.x86_64 6.2.1-5.fc41 nobara
rocm-cmake.noarch 6.2.0-1.fc41 nobara
rocm-comgr.x86_64 18-10.rocm6.2.1.fc41 nobara
rocm-comgr-devel.x86_64 18-10.rocm6.2.1.fc41 nobara
rocm-core.x86_64 6.2.0-1.fc41 nobara
rocm-device-libs.x86_64 18-10.rocm6.2.1.fc41 nobara
rocm-hip.x86_64 6.2.1-5.fc41 nobara
rocm-hip-devel.x86_64 6.2.1-5.fc41 nobara
rocm-meta.x86_64 6.2.1-7.copr.fc41 nobara-updates
rocm-opencl.x86_64 6.2.1-5.fc41 nobara
rocm-opencl-devel.x86_64 6.2.1-5.fc41 nobara
rocm-rpm-macros.x86_64 6.2-1.fc41 nobara
rocm-rpm-macros-modules.x86_64 6.2-1.fc41 nobara
rocm-runtime.x86_64 6.2.1-2.fc41 nobara
rocm-runtime-devel.x86_64 6.2.1-2.fc41 nobara
rocm-smi.x86_64 6.2.1-1.fc41 nobara
rocminfo.x86_64 6.2.1-1.fc41 nobara

Relevant log output

May 01 10:07:17 nobara-pc ollama[10218]: print_info: LF token         = 198 'Ċ'
May 01 10:07:17 nobara-pc ollama[10218]: print_info: EOG token        = 128008 '<|eom_id|>'
May 01 10:07:17 nobara-pc ollama[10218]: print_info: EOG token        = 128009 '<|eot_id|>'
May 01 10:07:17 nobara-pc ollama[10218]: print_info: max token length = 256
May 01 10:07:17 nobara-pc ollama[10218]: load_tensors: loading model tensors, this can take a while... (mmap = true)
May 01 10:07:17 nobara-pc ollama[10218]: load_tensors: offloading 28 repeating layers to GPU
May 01 10:07:17 nobara-pc ollama[10218]: load_tensors: offloading output layer to GPU
May 01 10:07:17 nobara-pc ollama[10218]: load_tensors: offloaded 29/29 layers to GPU
May 01 10:07:17 nobara-pc ollama[10218]: load_tensors:        ROCm0 model buffer size =  1918.35 MiB
May 01 10:07:17 nobara-pc ollama[10218]: load_tensors:   CPU_Mapped model buffer size =   308.23 MiB
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: constructing llama_context
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_seq_max     = 4
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_ctx         = 8192
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_ctx_per_seq = 2048
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_batch       = 2048
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_ubatch      = 512
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: causal_attn   = 1
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: flash_attn    = 0
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: freq_base     = 500000.0
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: freq_scale    = 1
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_ctx_per_seq (2048) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
May 01 10:07:53 nobara-pc ollama[10218]: llama_context:  ROCm_Host  output buffer size =     2.00 MiB
May 01 10:07:53 nobara-pc ollama[10218]: init: kv_size = 8192, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 28, can_shift = 1
May 01 10:07:53 nobara-pc ollama[10218]: init:      ROCm0 KV buffer size =   896.00 MiB
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: KV self size  =  896.00 MiB, K (f16):  448.00 MiB, V (f16):  448.00 MiB
May 01 10:07:53 nobara-pc ollama[10218]: llama_context:      ROCm0 compute buffer size =   424.00 MiB
May 01 10:07:53 nobara-pc ollama[10218]: llama_context:  ROCm_Host compute buffer size =    22.01 MiB
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: graph nodes  = 958
May 01 10:07:53 nobara-pc ollama[10218]: llama_context: graph splits = 2
May 01 10:07:54 nobara-pc ollama[10218]: time=2025-05-01T10:07:54.048+01:00 level=INFO source=server.go:619 msg="llama runner started in 37.87 seconds"
May 01 10:07:54 nobara-pc ollama[10218]: [GIN] 2025/05/01 - 10:07:54 | 200 | 38.161397178s |       127.0.0.1 | POST     "/api/generate"
May 01 10:08:41 nobara-pc ollama[10218]: time=2025-05-01T10:08:41.548+01:00 level=WARN source=ggml.go:152 msg="key not found" key=general.alignment default=32
May 01 10:27:36 nobara-pc ollama[10218]: [GIN] 2025/05/01 - 10:27:36 | 200 |        18m54s |       127.0.0.1 | POST     "/api/chat"
May 01 10:27:42 nobara-pc ollama[10218]: time=2025-05-01T10:27:42.764+01:00 level=WARN source=ggml.go:152 msg="key not found" key=general.alignment default=32
May 01 10:27:45 nobara-pc ollama[10218]: [GIN] 2025/05/01 - 10:27:45 | 200 |  2.613102748s |       127.0.0.1 | POST     "/api/chat"
May 01 10:28:03 nobara-pc ollama[10218]: [GIN] 2025/05/01 - 10:28:03 | 200 |       23.79µs |       127.0.0.1 | HEAD     "/"
May 01 10:28:03 nobara-pc ollama[10218]: [GIN] 2025/05/01 - 10:28:03 | 200 |      70.242µs |       127.0.0.1 | GET      "/api/ps"

OS

Linux

GPU

AMD

CPU

AMD

Ollama version

0.6.6

Originally created by @slashgob on GitHub (May 1, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/10516 ### What is the issue? Trying to run a small model (llama3.2 2gb), seems to load fine but I get no response from ollama. Using btop I can see GPU is at 100% usage, but still no response from chat message. It just stays at 100% usage. Nobara OS (Fedora spin) CPU: AMD 5950x Ram: 32GiB GPU: AMD 6900 XT Installed latest Ollama with curl script. ROCm installed... ─❯ dnf list --installed | grep 'rocm' hipcc.x86_64 18-10.rocm6.2.1.fc41 nobara hsakmt.x86_64 1.0.6-46.rocm6.2.1.fc41 nobara hsakmt-devel.x86_64 1.0.6-46.rocm6.2.1.fc41 nobara python3-torch-rocm-gfx9.x86_64 2.4.0-10.fc41 nobara python3-torchaudio-rocm-gfx9.x86_64 2.4.1-2.fc41 nobara rocm-clinfo.x86_64 6.2.1-5.fc41 nobara rocm-cmake.noarch 6.2.0-1.fc41 nobara rocm-comgr.x86_64 18-10.rocm6.2.1.fc41 nobara rocm-comgr-devel.x86_64 18-10.rocm6.2.1.fc41 nobara rocm-core.x86_64 6.2.0-1.fc41 nobara rocm-device-libs.x86_64 18-10.rocm6.2.1.fc41 nobara rocm-hip.x86_64 6.2.1-5.fc41 nobara rocm-hip-devel.x86_64 6.2.1-5.fc41 nobara rocm-meta.x86_64 6.2.1-7.copr.fc41 nobara-updates rocm-opencl.x86_64 6.2.1-5.fc41 nobara rocm-opencl-devel.x86_64 6.2.1-5.fc41 nobara rocm-rpm-macros.x86_64 6.2-1.fc41 nobara rocm-rpm-macros-modules.x86_64 6.2-1.fc41 nobara rocm-runtime.x86_64 6.2.1-2.fc41 nobara rocm-runtime-devel.x86_64 6.2.1-2.fc41 nobara rocm-smi.x86_64 6.2.1-1.fc41 nobara rocminfo.x86_64 6.2.1-1.fc41 nobara ### Relevant log output ```shell May 01 10:07:17 nobara-pc ollama[10218]: print_info: LF token = 198 'Ċ' May 01 10:07:17 nobara-pc ollama[10218]: print_info: EOG token = 128008 '<|eom_id|>' May 01 10:07:17 nobara-pc ollama[10218]: print_info: EOG token = 128009 '<|eot_id|>' May 01 10:07:17 nobara-pc ollama[10218]: print_info: max token length = 256 May 01 10:07:17 nobara-pc ollama[10218]: load_tensors: loading model tensors, this can take a while... (mmap = true) May 01 10:07:17 nobara-pc ollama[10218]: load_tensors: offloading 28 repeating layers to GPU May 01 10:07:17 nobara-pc ollama[10218]: load_tensors: offloading output layer to GPU May 01 10:07:17 nobara-pc ollama[10218]: load_tensors: offloaded 29/29 layers to GPU May 01 10:07:17 nobara-pc ollama[10218]: load_tensors: ROCm0 model buffer size = 1918.35 MiB May 01 10:07:17 nobara-pc ollama[10218]: load_tensors: CPU_Mapped model buffer size = 308.23 MiB May 01 10:07:53 nobara-pc ollama[10218]: llama_context: constructing llama_context May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_seq_max = 4 May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_ctx = 8192 May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_ctx_per_seq = 2048 May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_batch = 2048 May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_ubatch = 512 May 01 10:07:53 nobara-pc ollama[10218]: llama_context: causal_attn = 1 May 01 10:07:53 nobara-pc ollama[10218]: llama_context: flash_attn = 0 May 01 10:07:53 nobara-pc ollama[10218]: llama_context: freq_base = 500000.0 May 01 10:07:53 nobara-pc ollama[10218]: llama_context: freq_scale = 1 May 01 10:07:53 nobara-pc ollama[10218]: llama_context: n_ctx_per_seq (2048) < n_ctx_train (131072) -- the full capacity of the model will not be utilized May 01 10:07:53 nobara-pc ollama[10218]: llama_context: ROCm_Host output buffer size = 2.00 MiB May 01 10:07:53 nobara-pc ollama[10218]: init: kv_size = 8192, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 28, can_shift = 1 May 01 10:07:53 nobara-pc ollama[10218]: init: ROCm0 KV buffer size = 896.00 MiB May 01 10:07:53 nobara-pc ollama[10218]: llama_context: KV self size = 896.00 MiB, K (f16): 448.00 MiB, V (f16): 448.00 MiB May 01 10:07:53 nobara-pc ollama[10218]: llama_context: ROCm0 compute buffer size = 424.00 MiB May 01 10:07:53 nobara-pc ollama[10218]: llama_context: ROCm_Host compute buffer size = 22.01 MiB May 01 10:07:53 nobara-pc ollama[10218]: llama_context: graph nodes = 958 May 01 10:07:53 nobara-pc ollama[10218]: llama_context: graph splits = 2 May 01 10:07:54 nobara-pc ollama[10218]: time=2025-05-01T10:07:54.048+01:00 level=INFO source=server.go:619 msg="llama runner started in 37.87 seconds" May 01 10:07:54 nobara-pc ollama[10218]: [GIN] 2025/05/01 - 10:07:54 | 200 | 38.161397178s | 127.0.0.1 | POST "/api/generate" May 01 10:08:41 nobara-pc ollama[10218]: time=2025-05-01T10:08:41.548+01:00 level=WARN source=ggml.go:152 msg="key not found" key=general.alignment default=32 May 01 10:27:36 nobara-pc ollama[10218]: [GIN] 2025/05/01 - 10:27:36 | 200 | 18m54s | 127.0.0.1 | POST "/api/chat" May 01 10:27:42 nobara-pc ollama[10218]: time=2025-05-01T10:27:42.764+01:00 level=WARN source=ggml.go:152 msg="key not found" key=general.alignment default=32 May 01 10:27:45 nobara-pc ollama[10218]: [GIN] 2025/05/01 - 10:27:45 | 200 | 2.613102748s | 127.0.0.1 | POST "/api/chat" May 01 10:28:03 nobara-pc ollama[10218]: [GIN] 2025/05/01 - 10:28:03 | 200 | 23.79µs | 127.0.0.1 | HEAD "/" May 01 10:28:03 nobara-pc ollama[10218]: [GIN] 2025/05/01 - 10:28:03 | 200 | 70.242µs | 127.0.0.1 | GET "/api/ps" ``` ### OS Linux ### GPU AMD ### CPU AMD ### Ollama version 0.6.6
GiteaMirror added the bugamdlinux labels 2026-04-29 03:08:42 -05:00
GiteaMirror commented 2026-04-29 03:08:44 -05:00
Author
Owner
Copy Link

@mmbossoni commented on GitHub (Jun 1, 2025):

This is a problem with Nobara kernel handheld patch.

<!-- gh-comment-id:2927367493 --> @mmbossoni commented on GitHub (Jun 1, 2025): This is a problem with Nobara kernel handheld patch.
GiteaMirror commented 2026-04-29 03:08:45 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Jul 5, 2025):

If I had to guess, I'd say there's a bug or problem inside ROCm, maybe the driver, possibly leading the model to get stuck in a loop.

You can try setting AMD_LOG_LEVEL=3 along with OLLAMA_DEBUG=1 to get more verbose logging. dmesg might also have logs from the amdgpu driver.

<!-- gh-comment-id:3040305444 --> @dhiltgen commented on GitHub (Jul 5, 2025): If I had to guess, I'd say there's a bug or problem inside ROCm, maybe the driver, possibly leading the model to get stuck in a loop. You can try setting `AMD_LOG_LEVEL=3` along with `OLLAMA_DEBUG=1` to get more verbose logging. `dmesg` might also have logs from the `amdgpu` driver.
GiteaMirror commented 2026-04-29 03:08:45 -05:00
Author
Owner
Copy Link

@mmbossoni commented on GitHub (Jul 6, 2025):

@dhiltgen , you can see logs at https://github.com/ollama/ollama/issues/10919

The problem apparently lies with nobara kernel patch,
18217d5f97/baseos/kernel/6.14/0001-handheld.patch (L357-L358)

That being said, apparently the "doorbell" change is not on 6.15
I'll retest on 6.15.4 that apparently got rid of this change

<!-- gh-comment-id:3041795534 --> @mmbossoni commented on GitHub (Jul 6, 2025): @dhiltgen , you can see logs at https://github.com/ollama/ollama/issues/10919 The problem apparently lies with nobara kernel patch, https://github.com/Nobara-Project/rpm-sources/blob/18217d5f97bdf6b371ac5b4bccf724ad07373cd6/baseos/kernel/6.14/0001-handheld.patch#L357-L358 That being said, apparently the "doorbell" change is not on 6.15 I'll retest on 6.15.4 that apparently got rid of this change
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 linux
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#53431
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?