[GH-ISSUE #7109] Downloading models too slow #4516

Open
opened 2026-04-12 15:27:10 -05:00 by GiteaMirror · 5 comments
Owner

Originally created by @rubenmejiac on GitHub (Oct 6, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/7109

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

I have very slow downloads of models since I installed Ollama in Windows 11. No problems running models, etc. it's only the download speeds.
The terminal seems to report a different speed than shown in my network monitor.
I include screens of two downloads and the network monitor, which report 30Mbps approx. while the Ollama progress bars indicates 1.7-1.9MBps. I've got no problems with firewalls or proxies, and downloads from any other clients of big files work usually at 50-300MBps. My network speed is 600MBps

PowerShell 7.4.5 PS C:\Users\v6u2mop> ollama pull deepseek-coder-v2:16b pulling manifest pulling 5ff0abeeac1d... 32% ▕██████████████████ ▏ 2.9 GB/8.9 GB 1.9 MB/s 53m27s

PS C:\Users\ruben_v6u2mop> ollama pull qwen2.5-coder pulling manifest pulling ced7796abcbb... 69% ▕████████████████████████████ ▏ 3.2 GB/4.7 GB 1.8 MB/s 13m0s

Screenshot 2024-10-06 161819

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

0.3.12

Originally created by @rubenmejiac on GitHub (Oct 6, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/7109 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? I have very slow downloads of models since I installed Ollama in Windows 11. No problems running models, etc. it's only the download speeds. The terminal seems to report a different speed than shown in my network monitor. I include screens of two downloads and the network monitor, which report 30Mbps approx. while the Ollama progress bars indicates 1.7-1.9MBps. I've got no problems with firewalls or proxies, and downloads from any other clients of big files work usually at 50-300MBps. My network speed is 600MBps `PowerShell 7.4.5 PS C:\Users\v6u2mop> ollama pull deepseek-coder-v2:16b pulling manifest pulling 5ff0abeeac1d... 32% ▕██████████████████ ▏ 2.9 GB/8.9 GB 1.9 MB/s 53m27s ` `PS C:\Users\ruben_v6u2mop> ollama pull qwen2.5-coder pulling manifest pulling ced7796abcbb... 69% ▕████████████████████████████ ▏ 3.2 GB/4.7 GB 1.8 MB/s 13m0s ` ![Screenshot 2024-10-06 161819](https://github.com/user-attachments/assets/31420f1d-c6bf-43de-8981-b8f1d9cdd424) ### OS Windows ### GPU Nvidia ### CPU Intel ### Ollama version 0.3.12
GiteaMirror added the networkingbugneeds more infowindows labels 2026-04-12 15:27:11 -05:00
Author
Owner

@rick-github commented on GitHub (Oct 7, 2024):

Capital B for bytes, lowercase b for bits. There's still a discrepancy but it's not as large as it looks. If you have other processes using the network at the same time, that may account for it. Or there may just be some extra overhead in the model download.

<!-- gh-comment-id:2397131045 --> @rick-github commented on GitHub (Oct 7, 2024): Capital B for bytes, lowercase b for bits. There's still a discrepancy but it's not as large as it looks. If you have other processes using the network at the same time, that may account for it. Or there may just be some extra overhead in the model download.
Author
Owner

@dhiltgen commented on GitHub (Nov 5, 2024):

I haven't been able to reproduce this. On my windows test system, if I divide the Task Manager value by 8 it's ~identical to what Ollama is reporting.

Did you ever install ollama in WSL? Is it possible the ollama server is inside the WSL VM?

When you're not downloading in Ollama, does network traffic go to ~zero?

If you download large files using other tools, do the metrics match up better?

Can you share your server log? I'm curious if there are lots of failures and retries which might explain the discrepancy.

<!-- gh-comment-id:2458281268 --> @dhiltgen commented on GitHub (Nov 5, 2024): I haven't been able to reproduce this. On my windows test system, if I divide the Task Manager value by 8 it's ~identical to what Ollama is reporting. Did you ever install ollama in WSL? Is it possible the ollama server is inside the WSL VM? When you're not downloading in Ollama, does network traffic go to ~zero? If you download large files using other tools, do the metrics match up better? Can you share your server log? I'm curious if there are lots of failures and retries which might explain the discrepancy.
Author
Owner

@stevenh commented on GitHub (Feb 18, 2025):

Some network card drivers on windows do get bits and bytes wrong, it could be this.

<!-- gh-comment-id:2666779710 --> @stevenh commented on GitHub (Feb 18, 2025): Some network card drivers on windows do get bits and bytes wrong, it could be this.
Author
Owner

@ioquatix commented on GitHub (Jul 26, 2025):

I've seen this happen and it's reproducible.

Image

Apparently, one fix is to press Ctrl-C and retry, it will pick up where it's left off at full speed again:

Image

The Ctrl-C does restore speed for a while, but eventually it slows down again:

Image

Apparently this is due to a bug in the pull implementation:

Image

I'd be happy to take a look at this if I have time.

cc @pdevine

<!-- gh-comment-id:3120980096 --> @ioquatix commented on GitHub (Jul 26, 2025): I've seen this happen and it's reproducible. <img width="1948" height="581" alt="Image" src="https://github.com/user-attachments/assets/4e008f9f-22d3-4069-ac8a-3dac86491c8f" /> Apparently, one fix is to press Ctrl-C and retry, it will pick up where it's left off at full speed again: <img width="1948" height="581" alt="Image" src="https://github.com/user-attachments/assets/b3afe23c-f4da-4800-a587-72ff8ac0f550" /> The `Ctrl-C` does restore speed for a while, but eventually it slows down again: <img width="1948" height="581" alt="Image" src="https://github.com/user-attachments/assets/15f1133d-6aab-4969-abbb-7d44f40b4fec" /> Apparently this is due to a bug in the pull implementation: <img width="767" height="242" alt="Image" src="https://github.com/user-attachments/assets/6a3b342e-ede8-4f01-8145-ec0cbfcf6a8d" /> I'd be happy to take a look at this if I have time. cc @pdevine
Author
Owner

@rick-github commented on GitHub (Jul 26, 2025):

#10331

<!-- gh-comment-id:3120997831 --> @rick-github commented on GitHub (Jul 26, 2025): #10331
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#4516