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 #11949] Bad performance on Gemma3 with OLLAMA_KV_CACHE_TYPE less than fp16 #33694

New Issue
Closed
opened 2026-04-22 16:36:41 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-22 16:36:41 -05:00
Owner
Copy Link

Originally created by @McBane87 on GitHub (Aug 18, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11949

What is the issue?

When using gemma3:12b-it-qat and set OLLAMA_KV_CACHE_TYPE=q8_0 the token per second performance is very bad compared to fp16.

Results

Quant Token per second
FP16 29.9
Q8_0 5.5
Q4_0 5.5

As far as I know, it is not possible to set the KV cache per model. So it's an all or nothing option. That's why I decided to report this as a bug. Maybe you can disbale KV quantization, for this model, hardcoded in the source code. Same as you do already for other models?

Relevant log output

OS

Linux Ubuntu 24.04 + Docker Container

GPU

2 x Nvidia RTX 3060 12GB

CPU

AMD Ryzen 7900

Ollama version

0.11.5-rc2

Originally created by @McBane87 on GitHub (Aug 18, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11949 ### What is the issue? When using `gemma3:12b-it-qat` and set `OLLAMA_KV_CACHE_TYPE=q8_0` the token per second performance is very bad compared to `fp16`. **Results** | Quant | Token per second | | ------- | ------------------- | | FP16 | 29.9 | | Q8_0 | 5.5 | | Q4_0 | 5.5 | As far as I know, it is not possible to set the KV cache per model. So it's an all or nothing option. That's why I decided to report this as a bug. Maybe you can disbale KV quantization, for this model, hardcoded in the source code. Same as you do already for other models? ### Relevant log output ```shell ``` ### OS Linux Ubuntu 24.04 + Docker Container ### GPU 2 x Nvidia RTX 3060 12GB ### CPU AMD Ryzen 7900 ### Ollama version 0.11.5-rc2
GiteaMirror added the bug label 2026-04-22 16:36:41 -05:00
GiteaMirror commented 2026-04-22 16:36:43 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 18, 2025):

#9683

<!-- gh-comment-id:3195314754 --> @rick-github commented on GitHub (Aug 18, 2025): #9683
GiteaMirror commented 2026-04-22 16:36:43 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Aug 19, 2025):

Going to close as a dupe.

<!-- gh-comment-id:3202840994 --> @pdevine commented on GitHub (Aug 19, 2025): Going to close as a dupe.
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#33694
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?