github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-24 14:43:39 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #12725] Optimization with Nvidia DGX Spark #8443

New Issue
Closed
opened 2026-04-12 21:07:27 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-12 21:07:27 -05:00
Owner
Copy Link

Originally created by @VistritPandey on GitHub (Oct 21, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/12725

What is the issue?

I am not sure if this is a bug or just a performance optimization issue, but Ollama seems to be performing slower on the new DGX Spark.

Tested the same example on a Mac Studio with an M3 Ultra, and it was 2.5 times faster. In theory, DGX Spark should be fast, but I am unsure whether it's an Ollama issue or a Spark issue.

Fwiw, Ollama is being utilized properly, when I wanna see the GPU usage using nvidia-smi. Models I tried:

  • Llama 3.2 (3b)
  • Llama 3.2:vision (11b)
  • Llama 3.3 (70b)
  • deepseek-r1 (with thinking enabled) (8b)

Relevant log output

OS

Linux

GPU

Nvidia

CPU

Other

Ollama version

0.12.6

Originally created by @VistritPandey on GitHub (Oct 21, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/12725 ### What is the issue? I am not sure if this is a bug or just a performance optimization issue, but Ollama seems to be performing slower on the new DGX Spark. Tested the same example on a Mac Studio with an M3 Ultra, and it was 2.5 times faster. In theory, DGX Spark should be fast, but I am unsure whether it's an Ollama issue or a Spark issue. Fwiw, Ollama is being utilized properly, when I wanna see the GPU usage using `nvidia-smi`. Models I tried: - Llama 3.2 (3b) - Llama 3.2:vision (11b) - Llama 3.3 (70b) - deepseek-r1 (with thinking enabled) (8b) ### Relevant log output ```shell ``` ### OS Linux ### GPU Nvidia ### CPU Other ### Ollama version 0.12.6
GiteaMirror added the bug label 2026-04-12 21:07:27 -05:00
GiteaMirror commented 2026-04-12 21:07:30 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Oct 21, 2025):

By what measure is ollama slower?

<!-- gh-comment-id:3429438132 --> @rick-github commented on GitHub (Oct 21, 2025): By what measure is ollama slower?
GiteaMirror commented 2026-04-12 21:07:31 -05:00
Author
Owner
Copy Link

@VistritPandey commented on GitHub (Oct 21, 2025):

I meant the time it takes to process a request. For example, if it takes 30 seconds to process a request, it takes approximately 70ish seconds for the same request, settings, and model to process the request

Everything (prompt, settings, model, ctx, etc.) was the same, except the environment and hardware

<!-- gh-comment-id:3429485031 --> @VistritPandey commented on GitHub (Oct 21, 2025): I meant the time it takes to process a request. For example, if it takes 30 seconds to process a request, it takes approximately 70ish seconds for the same request, settings, and model to process the request Everything (prompt, settings, model, ctx, etc.) was the same, except the environment and hardware
GiteaMirror commented 2026-04-12 21:07:32 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Oct 21, 2025):

DGX Spark has a memory bandwidth of 273 GB/s, an M3 Ultra has 819 GB/s. The DGX has a better GPU, but processing time is going to be dominated by how long it takes to get the weights and activations into the GPU.

<!-- gh-comment-id:3429532493 --> @rick-github commented on GitHub (Oct 21, 2025): DGX Spark has a memory bandwidth of 273 GB/s, an M3 Ultra has 819 GB/s. The DGX has a better GPU, but processing time is going to be dominated by how long it takes to get the weights and activations into the GPU.
GiteaMirror commented 2026-04-12 21:07:33 -05:00
Author
Owner
Copy Link

@VistritPandey commented on GitHub (Oct 21, 2025):

Thanks, that makes sense, I was just flabbergasted by the "TFLOPS" lol

<!-- gh-comment-id:3429579584 --> @VistritPandey commented on GitHub (Oct 21, 2025): Thanks, that makes sense, I was just flabbergasted by the "TFLOPS" lol
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 bug
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#8443
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?