[GH-ISSUE #14141] Qwen3-vl:32b-instruct using lot of memory #55737

Closed
opened 2026-04-29 09:40:17 -05:00 by GiteaMirror · 2 comments
Owner

Originally created by @JamesInform on GitHub (Feb 7, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14141

I am just testing different models in ollama.

Testing e.g. qwen3-vl:32b-instruct, I see strange memory behavior:
The model itself has a size of 20GB.
But when I run it, it uses 93GB of ram.

I am running the model on an nvidia dgx spark.

See screenshot for details:

Image

Is this the exspected size?

Glad I have a dgx spark with 128GB of total ram :)

Cheers,
James

Originally created by @JamesInform on GitHub (Feb 7, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14141 I am just testing different models in ollama. Testing e.g. qwen3-vl:32b-instruct, I see strange memory behavior: The model itself has a size of 20GB. But when I run it, it uses 93GB of ram. I am running the model on an nvidia dgx spark. See screenshot for details: <img width="844" height="428" alt="Image" src="https://github.com/user-attachments/assets/44446fc0-2260-4e81-923d-4f421595da2b" /> Is this the exspected size? Glad I have a dgx spark with 128GB of total ram :) Cheers, James
Author
Owner

@rick-github commented on GitHub (Feb 7, 2026):

#14116

<!-- gh-comment-id:3865604664 --> @rick-github commented on GitHub (Feb 7, 2026): #14116
Author
Owner

@gzb commented on GitHub (Feb 9, 2026):

me too。

Image
<!-- gh-comment-id:3869038393 --> @gzb commented on GitHub (Feb 9, 2026): me too。 <img width="817" height="286" alt="Image" src="https://github.com/user-attachments/assets/4abc858b-f53f-4cd1-b74b-eaf49159aa4f" />
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#55737