mirror of
https://github.com/open-webui/open-webui.git
synced 2026-11-14 05:02:01 -06:00
[GH-ISSUE #16596] feat: Don't add so much ram & commit / pagefile memory usage overhead to the system! #33500
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: github-starred/open-webui#33500
Reference in New Issue
Block a user
Originally created by @tdbe on GitHub (Aug 13, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/16596
Check Existing Issues
Problem Description
I'm running a small Windows machine that has a total of 32GB of RAM (on a ramdisk ie no pagefile), and 8gb GPU VRAM. (why x? why y? Because of reasons, point is I only got 32gb + 8gb)
I can run Ollama and 24-32b parameter models just fine, but if I also add openWebUi (via uv and python), I can't!
Indulge my cynicism: I can remote-desktop into this machine, for a total maximum of 500MB of Commit size RAM, and Ollama uses 140MB. So your base 1.57GB is not much use to me alive is it? BTW, I even caught it reaching well over 4 GB of Commited memory! (but I don't remember what I was doing / configuring) (then there's also uv with another 400MB)
This is a fresh boot and a fresh start of a fresly configured (barely configured) OpenWebUI.
Desired Solution you'd like
Try to use as little RAM & Commit Size as possible. Maybe there's too much overhead via python or various web services, I don't know (and as a dumb user I shouldn't care).
Alternatives Considered
just remote desktop + ollama app uses 640MB.
Additional Context
No response
@tjbck commented on GitHub (Aug 13, 2025):
Most likely caused by built-in embedding inference engine, you should use external engines for minimal memory footprint
@tdbe commented on GitHub (Aug 13, 2025):
Lol insta-closed. Great experience with your product and support. Clap. I installed from your guide and created a user, did NOTHING ELSE.
THIS IS OUT OF THE BOX. Verify it yourself. 1.5GB Commit Size.
@Classic298 commented on GitHub (Aug 13, 2025):
Open WebUI is large. Look at what it can do, what it supports. From OpenAI to Ollama to sentence transformer to the insanely large cuda library that also needs to get loaded.
More features, more memory.
Also: duplicate issue and discussion. This has been raised many times.
Tim said many times, that he would welcome a PR for the introduction of a more minimal or toned-down version with some excluded features like cuda, if you'd wanna work on that, PR welcome.
So yes, issue closed - it is a 100th duplicate.
Search for existing discussions and issues first.
This is open source, the support you get is free - be grateful you get support at all.
If you want real support and CUSTOM installations (incl. smaller footprint) consider paying for it - there are options available.
Additionally, if you are using this "out of the box", then, as Tim said, you are running a local sentence transformer embedding model which also takes a 100MB+ RAM.
@rgaricano commented on GitHub (Aug 13, 2025):
@tdbe
Of course that it can be improved, is on the way.
But Man, don't ask pears from an elm!
For performance keep a 7b model running in GPU and OpenWebUI, Redis, webServer and other staff on CPU.
These are the juggling acts that those of us who have these computers have to do...for openwebui, for moving desktops and for everything we would like to include,
I have even tried using a shoehorn!...but when it doesn't give any more, not even with that!
@Classic298 commented on GitHub (Aug 13, 2025):
The community has been working on PRs and discussions about this topic many times @tdbe
The fact you knowingly chose to tick the box "yes i searched for existing discussions and issues" (just to obviously not search for them), to then be able to open this issue just to harshly and agressively complain is not productive at all.
This is open source community. We don't get money for helping you, in fact, we spend our time on commenting to you, for absolutely nothing in return.
If you have more ideas or feedback on how to improve things, share it in the discussions that already exist for this and help build a PR with others.
This is an insane statement in this context.
@tdbe commented on GitHub (Aug 13, 2025):
Well thanks for jumping in everyone. For the record, all I found in existing issues was this one from 2024 which points to a a docs page which no longer works: https://docs.openwebui.com/tutorial/reduce-ram-usage
And, I solved the issue 🙃: If I need to run OpenWebUi over LAN in minimal mode, I do the following:
(just set it to 0.0.0.0)
@Classic298 commented on GitHub (Aug 13, 2025):
@tdbe
Some discussions around the topic are here (not issues):
https://github.com/open-webui/open-webui/discussions/8598
https://github.com/open-webui/open-webui/discussions/11535
https://github.com/open-webui/open-webui/discussions/2583
And great you found an alternative with ollama-ui (and calling it a minimal mode for open-webui does not make your sarcasm-based facade look any better) - but comparing ollama-ui to open-webui is ... insane. Like comparing a Mini to a Rolls Royce.
Both drive, yeah - but one does SO much more and any comparison fails from the get-go.
Two different products, two different audiences, two different use cases.
If you were searching for the most light-weight and basic-feature-ish UI out there, then you were at the wrong place with open-webui from the beginning.
If you are a basic user with basic needs, then why do you choose the most feature packed and cutting-edge AI interface solution available on Github?
if all you wanted to do from the beginning was talk to the models in Ollama - no user management, no knowledge bases, no file handling, no cuda support, no openAI support, no image gen, no code interpreter, no folders, projects, workspace, notes, roles, permissions, functions, tools, plugins, etc, etc, etc, etc - then why do you even come here, if all you wanted was just a very barebones and light-weight UI for ollama?
That's the beauty of open source - probably you will find what you are looking for tailored to your needs, if you keep looking for it. Someone probably built it.
But complaining and being sarcastic from the get-go in your issue description and not searching for existing issues/discussions and then only being distasteful instead of asking follow-up questions does not help anyone here.