[GH-ISSUE #7330] Why Does Ollama Use Shard GPU Memory Before Filling Dedicated Ones? #4657

Closed
opened 2026-04-12 15:34:34 -05:00 by GiteaMirror · 5 comments
Owner

Originally created by @0TTA on GitHub (Oct 23, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/7330

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

Hi,

Is there a way to make Ollama uses more of my dedicated GPU memory? Or, can I tell it to start with the dedicated one and only switch to the shared memory if it needs to?

2024-10-23 13_49_51-Task Manager

OS

Windows

GPU

AMD

CPU

AMD

Ollama version

0.3.14

Originally created by @0TTA on GitHub (Oct 23, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/7330 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? Hi, Is there a way to make Ollama uses more of my dedicated GPU memory? Or, can I tell it to start with the dedicated one and only switch to the shared memory if it needs to? ![2024-10-23 13_49_51-Task Manager](https://github.com/user-attachments/assets/90a70e76-45b8-44dd-94c7-b9cf7dec5462) ### OS Windows ### GPU AMD ### CPU AMD ### Ollama version 0.3.14
GiteaMirror added the bug label 2026-04-12 15:34:34 -05:00
Author
Owner

@rick-github commented on GitHub (Oct 23, 2024):

Server logs might give some insight to why this is happening.

<!-- gh-comment-id:2431770496 --> @rick-github commented on GitHub (Oct 23, 2024): [Server logs](https://github.com/ollama/ollama/blob/main/docs/troubleshooting.md#how-to-troubleshoot-issues) might give some insight to why this is happening.
Author
Owner

@7shi commented on GitHub (Oct 23, 2024):

From what I've investigated, this appears to be a general hipMalloc behavior issue when allocating more than 4GB, not specific to Ollama.
https://github.com/ollama/ollama/issues/7328

<!-- gh-comment-id:2431916498 --> @7shi commented on GitHub (Oct 23, 2024): From what I've investigated, this appears to be a general `hipMalloc` behavior issue when allocating more than 4GB, not specific to Ollama. https://github.com/ollama/ollama/issues/7328
Author
Owner

@0TTA commented on GitHub (Oct 23, 2024):

Server logs might give some insight to why this is happening.

server.log

<!-- gh-comment-id:2432123977 --> @0TTA commented on GitHub (Oct 23, 2024): > [Server logs](https://github.com/ollama/ollama/blob/main/docs/troubleshooting.md#how-to-troubleshoot-issues) might give some insight to why this is happening. [server.log](https://github.com/user-attachments/files/17491860/server.log)
Author
Owner

@dhiltgen commented on GitHub (Oct 23, 2024):

This looks like a dup of #7107 - the only workaround for now is to downgrade to a driver OLDER than 24.9.1 until 24.11 comes out which should have the fix.

<!-- gh-comment-id:2432831797 --> @dhiltgen commented on GitHub (Oct 23, 2024): This looks like a dup of #7107 - the only workaround for now is to downgrade to a driver OLDER than 24.9.1 until 24.11 comes out which should have the fix.
Author
Owner

@0TTA commented on GitHub (Oct 23, 2024):

This looks like a dup of #7107 - the only workaround for now is to downgrade to a driver OLDER than 24.9.1 until 24.11 comes out which should have the fix.

Thank you!

<!-- gh-comment-id:2432919465 --> @0TTA commented on GitHub (Oct 23, 2024): > This looks like a dup of #7107 - the only workaround for now is to downgrade to a driver OLDER than 24.9.1 until 24.11 comes out which should have the fix. Thank you!
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#4657