[GH-ISSUE #15721] GPU is not working in screen off mode #72083

Open
opened 2026-05-05 03:25:58 -05:00 by GiteaMirror · 2 comments
Owner

Originally created by @josergc on GitHub (Apr 20, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15721

What is the issue?

I've Ollama 0.21.0 configured to work with the GPU Intel UHD Graphics 620 through vulkan in Windows 11. In the configuration, in power & battery, only "turn my screen off" after 15 minutes is configured and the other options are using "never". Everything works fine until the screen goes off. The API accepts connections but ends in timeout. There is no GPU activity until I turn the screen on again, then API accepts connections and processes the requests.

Basically, the GPU should be available during "turn my screen off" despite of this is off to keep running just like a server.

Relevant log output


OS

Windows

GPU

Intel

CPU

Intel

Ollama version

0.21.0

Originally created by @josergc on GitHub (Apr 20, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15721 ### What is the issue? I've Ollama 0.21.0 configured to work with the GPU Intel UHD Graphics 620 through vulkan in Windows 11. In the configuration, in power & battery, only "turn my screen off" after 15 minutes is configured and the other options are using "never". Everything works fine until the screen goes off. The API accepts connections but ends in timeout. There is no GPU activity until I turn the screen on again, then API accepts connections and processes the requests. Basically, the GPU should be available during "turn my screen off" despite of this is off to keep running just like a server. ### Relevant log output ```shell ``` ### OS Windows ### GPU Intel ### CPU Intel ### Ollama version 0.21.0
GiteaMirror added the bug label 2026-05-05 03:25:58 -05:00
Author
Owner

@mverrilli commented on GitHub (Apr 20, 2026):

I think WDDM power management on integrated Intel graphics can suspend the display adapter when the screen goes off, blocking Vulkan compute. This is a driver/OS-level interaction.

<!-- gh-comment-id:4282615062 --> @mverrilli commented on GitHub (Apr 20, 2026): I think WDDM power management on integrated Intel graphics can suspend the display adapter when the screen goes off, blocking Vulkan compute. This is a driver/OS-level interaction.
Author
Owner

@josergc commented on GitHub (Apr 21, 2026):

can this be managed by the Ollama application that is accessing the driver? is Ollama posting debug logs about this?

<!-- gh-comment-id:4285522540 --> @josergc commented on GitHub (Apr 21, 2026): can this be managed by the Ollama application that is accessing the driver? is Ollama posting debug logs about this?
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#72083