github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-15 12:48:05 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #6695] Q6_K is slower than Q8_0 #4213

New Issue
Closed
opened 2026-04-12 15:09:04 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-12 15:09:04 -05:00
Owner
Copy Link

Originally created by @napa3um on GitHub (Sep 8, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/6695

What is the issue?

gemma2:9b-instruct-q6_K : gemma2:9b-instruct-q8_0 = 21t/s : 25t/s
mistral-nemo:12b-instruct-2407-q6_K : mistral-nemo:12b-instruct-2407-q8_0 = 17t/s : 21t/s

It used to be different.

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

0.3.9

Originally created by @napa3um on GitHub (Sep 8, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/6695 ### What is the issue? gemma2:9b-instruct-**q6_K** : gemma2:9b-instruct-**q8_0** = **21**t/s : **25**t/s mistral-nemo:12b-instruct-2407-**q6_K** : mistral-nemo:12b-instruct-2407-**q8_0** = **17**t/s : **21**t/s It used to be different. ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version 0.3.9
GiteaMirror added the bug label 2026-04-12 15:09:04 -05:00
GiteaMirror commented 2026-04-12 15:09:05 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Sep 8, 2024):

Server logs of sessions involving both Q6_K and Q8_0 models may give some insight to what's going on, especially if you can also get logs from before the tps rate changed. FWIW a linux+nvidia system shows consistent tps over a range of ollama versions.

Ollama version gemma2:9b-instruct-q6_K gemma2:9b-instruct-q8_0 mistral-nemo:12b-instruct-2407-q6_K mistral-nemo:12b-instruct-2407-q8_0
0.2.8 29.10 24.06 25.12 20.30
0.3.0 29.05 24.04 25.09 20.34
0.3.1 29.07 24.08 25.12 20.36
0.3.2 29.08 24.06 25.07 20.36
0.3.3 28.99 24.03 25.05 20.35
0.3.4 29.00 24.00 25.02 20.32
0.3.5 28.96 24.00 25.00 20.31
0.3.6 29.00 23.99 24.98 20.27
0.3.7 28.99 24.01 24.97 20.31
0.3.8 28.99 23.84 24.96 20.30
0.3.9 29.00 24.00 24.95 20.29
<!-- gh-comment-id:2336676756 --> @rick-github commented on GitHub (Sep 8, 2024): [Server logs](https://github.com/ollama/ollama/blob/main/docs/troubleshooting.md#how-to-troubleshoot-issues) of sessions involving both Q6_K and Q8_0 models may give some insight to what's going on, especially if you can also get logs from before the tps rate changed. FWIW a linux+nvidia system shows consistent tps over a range of ollama versions. | Ollama version | gemma2:9b-instruct-q6_K | gemma2:9b-instruct-q8_0 | mistral-nemo:12b-instruct-2407-q6_K | mistral-nemo:12b-instruct-2407-q8_0 | | --- | --- | --- | --- | --- | | 0.2.8 | 29.10| 24.06| 25.12| 20.30 | | 0.3.0 | 29.05| 24.04| 25.09| 20.34 | | 0.3.1 | 29.07| 24.08| 25.12| 20.36 | | 0.3.2 | 29.08| 24.06| 25.07| 20.36 | | 0.3.3 | 28.99| 24.03| 25.05| 20.35 | | 0.3.4 | 29.00| 24.00| 25.02| 20.32 | | 0.3.5 | 28.96| 24.00| 25.00| 20.31 | | 0.3.6 | 29.00| 23.99| 24.98| 20.27 | | 0.3.7 | 28.99| 24.01| 24.97| 20.31 | | 0.3.8 | 28.99| 23.84| 24.96| 20.30 | | 0.3.9 | 29.00| 24.00| 24.95| 20.29 |
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#4213
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?