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 #3982] CUDA error while trying to run llama3-8B: out of memory #48976

New Issue
Closed
opened 2026-04-28 10:23:40 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-28 10:23:40 -05:00
Owner
Copy Link

Originally created by @piotrfila on GitHub (Apr 27, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/3982

What is the issue?

Hello,

I am trying to run llama3-8B:instruct on 2 * GTX 970 (4GB, CUDA 5.2), no SLI. #1288 led me to believe it should be possible in terms of VRAM requirements (8GB total) and I also have enough RAM (16GB). However, each time I try to run the model the ollama service crashes due to out of memory error and no response is returned (tried with oterm, open-webui and ollama run).

This seems similar to #3765. I tried editing the modelfile as mentioned there without success.

The model runs successfully without GPU accleration. Other applications can use cuda just fine (checked with these examples).

I installed ollama through the NixOS-unstable option.

Modelfile I tried:

# Modelfile generated by "ollama show"
# To build a new Modelfile based on this one, replace the FROM line with:
FROM llama3:instruct

TEMPLATE """{{ if .System }}<|start_header_id|>system<|end_header_id|>

{{ .System }}<|eot_id|>{{ end }}{{ if .Prompt }}<|start_header_id|>user<|end_header_id|>

{{ .Prompt }}<|eot_id|>{{ end }}<|start_header_id|>assistant<|end_header_id|>

{{ .Response }}<|eot_id|>"""
PARAMETER num_ctx 4196
PARAMETER num_gpu 42
PARAMETER num_keep 24
PARAMETER stop "<|start_header_id|>"
PARAMETER stop "<|end_header_id|>"
PARAMETER stop "<|eot_id|>"

Tail of journalctl log: https://pastebin.com/yVH5Sgai

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.1.31

Originally created by @piotrfila on GitHub (Apr 27, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/3982 ### What is the issue? Hello, I am trying to run llama3-8B:instruct on 2 * GTX 970 (4GB, CUDA 5.2), no SLI. #1288 led me to believe it should be possible in terms of VRAM requirements (8GB total) and I also have enough RAM (16GB). However, each time I try to run the model the ollama service crashes due to out of memory error and no response is returned (tried with oterm, open-webui and `ollama run`). This seems similar to #3765. I tried editing the modelfile as mentioned there without success. The model runs successfully without GPU accleration. Other applications can use cuda just fine (checked with [these examples](https://github.com/grahamc/nixos-cuda-example)). I installed ollama through the NixOS-unstable option. Modelfile I tried: ``` # Modelfile generated by "ollama show" # To build a new Modelfile based on this one, replace the FROM line with: FROM llama3:instruct TEMPLATE """{{ if .System }}<|start_header_id|>system<|end_header_id|> {{ .System }}<|eot_id|>{{ end }}{{ if .Prompt }}<|start_header_id|>user<|end_header_id|> {{ .Prompt }}<|eot_id|>{{ end }}<|start_header_id|>assistant<|end_header_id|> {{ .Response }}<|eot_id|>""" PARAMETER num_ctx 4196 PARAMETER num_gpu 42 PARAMETER num_keep 24 PARAMETER stop "<|start_header_id|>" PARAMETER stop "<|end_header_id|>" PARAMETER stop "<|eot_id|>" ``` Tail of journalctl log: https://pastebin.com/yVH5Sgai ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.1.31
GiteaMirror added the bug label 2026-04-28 10:23:40 -05:00
GiteaMirror commented 2026-04-28 10:25:11 -05:00
Author
Owner
Copy Link

@helium729 commented on GitHub (Apr 28, 2024):

Usually a 8B q4_0 model requires at least 5.4 GB of VRAM each, guess yours is not sufficient, you can try some models <= 4b

<!-- gh-comment-id:2081507820 --> @helium729 commented on GitHub (Apr 28, 2024): Usually a 8B q4_0 model requires at least 5.4 GB of VRAM each, guess yours is not sufficient, you can try some models <= 4b
GiteaMirror commented 2026-04-28 10:25:13 -05:00
Author
Owner
Copy Link

@piotrfila commented on GitHub (Apr 28, 2024):

Yeah, phi3 works fine. I didn't know more total vram was needed when using multiple GPUs but it makes sense now that I think about it.

<!-- gh-comment-id:2081593948 --> @piotrfila commented on GitHub (Apr 28, 2024): Yeah, phi3 works fine. I didn't know more total vram was needed when using multiple GPUs but it makes sense now that I think about it.
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#48976
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?