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 #15295] Gemma4 very slow on GB10 #9786

New Issue
Closed
opened 2026-04-12 22:40:03 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-12 22:40:03 -05:00
Owner
Copy Link

Originally created by @phr0gz on GitHub (Apr 3, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15295

What is the issue?

On GB10 platforms (aka DGX Spark), the Gemma 31B models (Q4, Q8, BF16 ) are very slow:

See other model values:
qwen3.5:35b-a3b-q4_K_M, response token/s: 59.68
qwen3.5:122b-a10b-q4_K_M, response token/s: 24
gemma4:26b-a4b-it-q4_K_M, response token/s: 58.67

And see gemma4 31B value:
gemma4:31b-it-q4_K_M, response token/s: 10.34

EDIT: I also tested with llama.cpp, and it's as slow as ollama.

Relevant log output

OS

Docker

GPU

Nvidia

CPU

Other

Ollama version

0.20.0

Originally created by @phr0gz on GitHub (Apr 3, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15295 ### What is the issue? On GB10 platforms (aka DGX Spark), the Gemma 31B models (Q4, Q8, BF16 ) are very slow: See other model values: qwen3.5:35b-a3b-q4_K_M, response token/s: 59.68 qwen3.5:122b-a10b-q4_K_M, response token/s: 24 gemma4:26b-a4b-it-q4_K_M, response token/s: 58.67 And see gemma4 31B value: gemma4:31b-it-q4_K_M, response token/s: 10.34 EDIT: I also tested with llama.cpp, and it's as slow as ollama. ### Relevant log output ```shell ``` ### OS Docker ### GPU Nvidia ### CPU Other ### Ollama version 0.20.0
GiteaMirror added the bug label 2026-04-12 22:40:03 -05:00
GiteaMirror commented 2026-04-12 22:40:05 -05:00
Author
Owner
Copy Link

@3DPJamie commented on GitHub (Apr 3, 2026):

are you aware of difference between dense models and activated parameters models? your examples are:
qwen3.5:35b-a3b-q4_K_M, response token/s: 59.68
qwen3.5:122b-a10b-q4_K_M, response token/s: 24
gemma4:26b-a4b-it-q4_K_M, response token/s: 58.67

which all have 3-10 active parameters, while gemma4:31b is a dense model with 31B active parameters. try your test with qwen3.5:27b model and do the comparison again

<!-- gh-comment-id:4184131573 --> @3DPJamie commented on GitHub (Apr 3, 2026): are you aware of difference between dense models and activated parameters models? your examples are: qwen3.5:35b-a3b-q4_K_M, response token/s: 59.68 qwen3.5:122b-a10b-q4_K_M, response token/s: 24 gemma4:26b-a4b-it-q4_K_M, response token/s: 58.67 which all have 3-10 active parameters, while gemma4:31b is a dense model with 31B active parameters. try your test with qwen3.5:27b model and do the comparison again
GiteaMirror commented 2026-04-12 22:40:06 -05:00
Author
Owner
Copy Link

@phr0gz commented on GitHub (Apr 3, 2026):

You are completely right:
qwen3.5:27b-q4_K_M, response token/s: 11.51

<!-- gh-comment-id:4184241171 --> @phr0gz commented on GitHub (Apr 3, 2026): You are completely right: qwen3.5:27b-q4_K_M, response token/s: 11.51
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 bug
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#9786
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?