ollama

Official

Overview

Official Ollama images for Nvidia or AMD GPUs. Intel Arc GPUs could potentially work with Vulkan enabled. NOTE: Extra Parameters needed for Nvidia: --gpus=all Additionally you can add a Variable called CUDA_VISIBLE_DEVICES to only use specific Nvidia GPU(s) by ID. Example: 0 or 0,1 Extra Parameters needed for AMD: --device='/dev/kfd' --device='/dev/dri' Additionally you can add a Variable called ROCR_VISIBLE_DEVICES to only use specific AMD GPU(s) by UUID. Example: GPU-67246603a11f0a21 or GPU-XXXXXXX,GPU-YYYYYYY Docs: https://docs.ollama.com/

Requirements

Nvidia-Driver plugin (nVidia Support)

Radeon-TOP plugin (AMD Support)

Runtime arguments

Web UI: http://[IP]:[PORT:11434]/
Network: bridge
Shell: bash
Privileged: false

Template configuration

DataPathrw

Target: /root/.ollama
Default: /mnt/user/appdata/ollama
Value: /mnt/user/appdata/ollama

API Interface PortPorttcp

Port number where ollama listens on.

Target: 11434
Default: 11434
Value: 11434

OLLAMA_HOSTVariable

IP and Port the server binds to. Set to 127.0.0.1:11434 for internal only access.

Default: 0.0.0.0:11434
Value: 0.0.0.0:11434

OLLAMA_ORIGINSVariable

Comma-separated list of allowed CORS origins.

Default: *
Value: *

OLLAMA_KEEP_ALIVEVariable

How long a model stays in VRAM, e.g. 60m or 24h (Set to -1 for infinite, 0 for none).

Default: 5m
Value: 5m

OLLAMA_LOAD_TIMEOUTVariable

Timeout for stall detection during model loads.

Default: 5m
Value: 5m

OLLAMA_NUM_PARALLELVariable

Max number of parallel requests a single model can handle.

Default: 1
Value: 1

OLLAMA_CONTEXT_LENGTHVariable

Default context window (tokens) if not specified by the model.

Default: 4096
Value: 4096

OLLAMA_KV_CACHE_TYPEVariable

Quantization type for the K/V cache, e.g. f16, q8_0, q4_0.

Default: f16
Value: f16

OLLAMA_MODELSVariable

The path where model weights and blobs are stored.

Default: /root/.ollama/models
Value: /root/.ollama/models

OLLAMA_MAX_LOADED_MODELSVariable

Maximum number of models loaded per GPU at once (Set to 0 for infinite).

Default: 0
Value: 0

OLLAMA_MAX_QUEUEVariable

Max requests that can wait in line when the server is busy.

Default: 512
Value: 512

OLLAMA_DEBUGVariable

Log detail level: 0 for INFO, 1 for DEBUG, 2 for TRACE.

Default: 0|1|2

OLLAMA_GPU_OVERHEADVariable

Reserved VRAM (in bytes) to leave empty on each GPU.

Default: 0
Value: 0

OLLAMA_FLASH_ATTENTIONVariable

Enables experimental Flash Attention optimizations.

Default: false|true

OLLAMA_SCHED_SPREADVariable

If true, always spreads model layers across all visible GPUs.

Default: false|true

OLLAMA_MULTIUSER_CACHEVariable

Optimizes prompt caching when multiple users share a model.

Default: false|true

OLLAMA_NOPRUNEVariable

If true, does not delete unused model blobs on startup.

Default: false|true

OLLAMA_NOHISTORYVariable

Disables the readline history in the interactive CLI.

Default: false|true

OLLAMA_NEW_ENGINEVariable

Enables the experimental new Ollama engine.

Default: false|true

OLLAMA_VULKANVariable

Enables experimental Vulkan hardware acceleration.

Default: false|true

HTTP_PROXYVariable

Proxy for downloading models over HTTP.

HTTPS_PROXYVariable

Proxy for downloading models over HTTPS.

NO_PROXYVariable

Comma-separate list of hosts/IPs that bypass the proxy.

Links

Template Support Docker Hub

Details

Repository

ollama/ollama

Registry

https://hub.docker.com/r/ollama/ollama/

Last Updated2026-05-27

First Seen2023-11-08

Run Ollama on Unraid.

Ollama is listed in Community Apps for Unraid OS. Explore Unraid to build a flexible home server, NAS, or homelab.

Explore Unraid OS