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 #8795] Load model into NVME SSD #31471

New Issue
Closed
opened 2026-04-22 11:55:43 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-22 11:55:43 -05:00
Owner
Copy Link

Originally created by @phly95 on GitHub (Feb 3, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/8795

Apparently, it is possible to use NVME SSDs to load model weights, which can be used directly by the GPU.

https://www.reddit.com/r/LocalLLaMA/comments/1iczucy/running_deepseek_r1_iq2xxs_200gb_from_ssd/

From what I've heard, this does not damage the SSD since it's performing frequent reads and not writes, and can allow users with very little VRAM to run much larger models. Apparently, there is also another project called UMbreLLa that can also further speed up inference. Perhaps if ollama were to integrate these two setups, users of small 8GB cards might be able to expect reasonable performance with models like DeepSeek R1, Mistral Small 24b, llama 70b, etc.

Originally created by @phly95 on GitHub (Feb 3, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/8795 Apparently, it is possible to use NVME SSDs to load model weights, which can be used directly by the GPU. https://www.reddit.com/r/LocalLLaMA/comments/1iczucy/running_deepseek_r1_iq2xxs_200gb_from_ssd/ From what I've heard, this does not damage the SSD since it's performing frequent reads and not writes, and can allow users with very little VRAM to run much larger models. Apparently, there is also another project called UMbreLLa that can also further speed up inference. Perhaps if ollama were to integrate these two setups, users of small 8GB cards might be able to expect reasonable performance with models like DeepSeek R1, Mistral Small 24b, llama 70b, etc.
GiteaMirror added the feature request label 2026-04-22 11:55:43 -05:00
GiteaMirror commented 2026-04-22 11:55:44 -05:00
Author
Owner
Copy Link

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

This is just mmap. ollama disables this for some platforms based on performance reports, but can that can be overriden by setting "use_mmap":true. There is one wrinkle in that ollama still does a memory check even if mmap is enabled, a ticket about that includes a workaround for Linux platforms.

<!-- gh-comment-id:2631898310 --> @rick-github commented on GitHub (Feb 3, 2025): This is just `mmap`. ollama disables this for [some platforms](https://github.com/ollama/ollama/blob/ad22ace439eb3fab7230134e56bb6276a78347e4/llm/server.go#L210) based on performance reports, but can that can be overriden by [setting](https://github.com/ollama/ollama/blob/main/docs/api.md#generate-request-with-options:~:text=vocab_only%22%3A%20false%2C%0A%20%20%20%20%22-,use_mmap,-%22%3A%20true%2C%0A%20%20%20%20%22use_mlock) `"use_mmap":true`. There is one wrinkle in that ollama still does a memory check even if `mmap` is enabled, a [ticket](https://github.com/ollama/ollama/issues/8654) about that includes a workaround for Linux platforms.
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 feature request
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#31471
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?