github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-08 17:49:24 -05:00
Code Issues 14.6k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #13261] Make low_vram_mode threshold configurable #34526

New Issue
Closed
opened 2026-04-22 18:10:46 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-22 18:10:46 -05:00
Owner
Copy Link

Originally created by @flrtemis on GitHub (Nov 28, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/13261

What is the issue?

ollama serve drops into low_vram_mode on any machine where the OS reports <20 GiB VRAM, even when the GPU is a high-end 12/16/24 GB card that runs models perfectly. The trigger is a hard-coded literal in server/routes.go:

var lowVRAMThreshold uint64 = 20 * format.GibiByte

Because Windows often reports only ~15–18 GiB “free” on a 24–32 GB card, those users are permanently stuck in low_vram_mode with no escape hatch.

Clarification: the OS isn’t “stealing” VRAM; it reports what remains after display buffers/drivers. The problem is the fixed 20 GiB cutoff instead of a configurable value.

Expected behavior

Allow the cutoff to be set via env/config so normal GPU mode stays active unless memory truly falls below the user’s chosen margin.

Proposed fix

    1. Add var LowVRAMThreshold = Uint64("OLLAMA_LOW_VRAM_THRESHOLD", 20*format.GibiByte) in envconfig/config.go and register it in AsMap().
    2. In server/routes.go, read envconfig.LowVRAMThreshold() instead of the literal.
    3. Document OLLAMA_LOW_VRAM_THRESHOLD (bytes) in the FAQ.

Reference implementation + build steps: https://github.com/flrtemis/ollama (current HEAD 730f0a8). Setting OLLAMA_LOW_VRAM_THRESHOLD=12000000000 keeps a 16 GB RTX 5070 Ti in normal mode.


### Relevant log output

```shell
time=2025-11-27T17:52:11.110-05:00 level=INFO source=routes.go:1545 msg="server config" env="... OLLAMA_LOW_VRAM_THRESHOLD:12000000000 ..."

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

0.3.10 (custom build from commit 730f0a8; “ollama --version” currently reports 0.0.0 after the rebuild)

Originally created by @flrtemis on GitHub (Nov 28, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/13261 ### What is the issue? `ollama serve` drops into low_vram_mode on any machine where the OS reports <20 GiB VRAM, even when the GPU is a high-end 12/16/24 GB card that runs models perfectly. The trigger is a hard-coded literal in server/routes.go: ```go var lowVRAMThreshold uint64 = 20 * format.GibiByte Because Windows often reports only ~15–18 GiB “free” on a 24–32 GB card, those users are permanently stuck in low_vram_mode with no escape hatch. Clarification: the OS isn’t “stealing” VRAM; it reports what remains after display buffers/drivers. The problem is the fixed 20 GiB cutoff instead of a configurable value. Expected behavior Allow the cutoff to be set via env/config so normal GPU mode stays active unless memory truly falls below the user’s chosen margin. Proposed fix 1. Add var LowVRAMThreshold = Uint64("OLLAMA_LOW_VRAM_THRESHOLD", 20*format.GibiByte) in envconfig/config.go and register it in AsMap(). 2. In server/routes.go, read envconfig.LowVRAMThreshold() instead of the literal. 3. Document OLLAMA_LOW_VRAM_THRESHOLD (bytes) in the FAQ. Reference implementation + build steps: https://github.com/flrtemis/ollama (current HEAD 730f0a8). Setting OLLAMA_LOW_VRAM_THRESHOLD=12000000000 keeps a 16 GB RTX 5070 Ti in normal mode. ### Relevant log output ```shell time=2025-11-27T17:52:11.110-05:00 level=INFO source=routes.go:1545 msg="server config" env="... OLLAMA_LOW_VRAM_THRESHOLD:12000000000 ..." ``` ### OS Windows ### GPU Nvidia ### CPU Intel ### Ollama version 0.3.10 (custom build from commit 730f0a8; “ollama --version” currently reports 0.0.0 after the rebuild)
GiteaMirror added the bug label 2026-04-22 18:10:46 -05:00
GiteaMirror commented 2026-04-22 18:10:48 -05:00
Author
Owner
Copy Link

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

https://github.com/ollama/ollama/issues/12143#issuecomment-3242462236

<!-- gh-comment-id:3588244189 --> @rick-github commented on GitHub (Nov 28, 2025): https://github.com/ollama/ollama/issues/12143#issuecomment-3242462236
GiteaMirror commented 2026-04-22 18:10:48 -05:00
Author
Owner
Copy Link

@flrtemis commented on GitHub (Dec 3, 2025):

@rick-github https://flrtemis.github.io/OllamaSourceCodeRebuilt/3d.html

<!-- gh-comment-id:3605132130 --> @flrtemis commented on GitHub (Dec 3, 2025): @rick-github https://flrtemis.github.io/OllamaSourceCodeRebuilt/3d.html
GiteaMirror commented 2026-04-22 18:10:49 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Dec 3, 2025):

You mis-understand the purpose of low VRAM mode. Processing is not slower in low VRAM mode. All that happens is that for a few selected architectures, the default value of num_ctx is a lower value. You can achieve the same result by setting OLLAMA_CONTEXT_LENGTH=8192.

<!-- gh-comment-id:3605960386 --> @rick-github commented on GitHub (Dec 3, 2025): You mis-understand the purpose of low VRAM mode. Processing is not slower in low VRAM mode. All that happens is that for a few [selected architectures](https://github.com/ollama/ollama/blob/cc9555aff0f220748dc761a4302cfaea7c62c9fe/server/routes.go#L146), the default value of `num_ctx` is a lower value. You can achieve the same result by setting `OLLAMA_CONTEXT_LENGTH=8192`.
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#34526
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?