github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #4599] Enable concurrency by default for AMD GPU on windows - requires ROCm v6.2 windows release #28647

New Issue
Open
opened 2026-04-22 07:07:31 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-22 07:07:31 -05:00
Owner
Copy Link

Originally created by @dhiltgen on GitHub (May 23, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4599

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

When running model concurrency, the scheduler is unaware of WDDM KMD memory allocations on system memory and just looks like GPU reported memory usage, which can lead to loading too many models and layers, resulting in thrashing and poor performance.

OS

Windows

GPU

Nvidia, AMD

CPU

Intel

Ollama version

No response

Originally created by @dhiltgen on GitHub (May 23, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4599 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? When running model concurrency, the scheduler is unaware of WDDM KMD memory allocations on system memory and just looks like GPU reported memory usage, which can lead to loading too many models and layers, resulting in thrashing and poor performance. ### OS Windows ### GPU Nvidia, AMD ### CPU Intel ### Ollama version _No response_
GiteaMirror added the bugamdwindows labels 2026-04-22 07:07:31 -05:00
GiteaMirror commented 2026-04-22 07:07:35 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Jun 4, 2024):

PR #4517 now includes an update to pull the nvml library in on windows to query GPU VRAM consumption for CUDA cards, which should address the CUDA portion of this. I'm still trying to find the optimal strategy for Radeon cards.

<!-- gh-comment-id:2148190097 --> @dhiltgen commented on GitHub (Jun 4, 2024): PR #4517 now includes an update to pull the nvml library in on windows to query GPU VRAM consumption for CUDA cards, which should address the CUDA portion of this. I'm still trying to find the optimal strategy for Radeon cards.
GiteaMirror commented 2026-04-22 07:07:37 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Jun 22, 2024):

This is resolved now for CUDA. ROCm will require a version bump to v6.2 but that hasn't been released yet by AMD.

<!-- gh-comment-id:2183592740 --> @dhiltgen commented on GitHub (Jun 22, 2024): This is resolved now for CUDA. ROCm will require a version bump to v6.2 but that hasn't been released yet by AMD.
GiteaMirror commented 2026-04-22 07:07:39 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Jul 3, 2024):

I'm re-tasking this issue to track AMD specifically, since NVIDIA support has already merged to main and concurrency will be enabled for CUDA GPUs in the upcoming release.

<!-- gh-comment-id:2207521317 --> @dhiltgen commented on GitHub (Jul 3, 2024): I'm re-tasking this issue to track AMD specifically, since NVIDIA support has already merged to main and concurrency will be enabled for CUDA GPUs in the upcoming release.
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 amd bug windows
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#28647
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?