Je comprend pas pourquoi Ollama utilise le CPU ?
retained text
Bonjour,
Pourriez-vous m'aider svp sur la raison que mon ollama consomme mon CPU malgré que le modèle indique 100% GPU.
```yml
###########
services: #
###########
#
#
#############################################################
Ollama: #
# ------------------------------------------------------- #
image: 'ollama/ollama:latest' #
container_name: '01-ollama' #
# ------------------------------------------------------- #
network_mode: 'bridge' #
# ------------------------------------------------------- #
restart: 'unless-stopped' #
# ------------------------------------------------------- #
hostname: 'Ollama' #
# ------------------------------------------------------- #
ipc: 'host' #
# ------------------------------------------------------- #
deploy: #
resources: #
# ------------------------------------------------------ #
reservations: #
# ----------------------------------------------------- #
devices: #
- driver: 'nvidia' #
count: 'all' #
capabilities: ['gpu', 'utility'] #
# ------------------------------------------------------- #
environment: #
# ------------------------------------------------------ #
OLLAMA_DEBUG: 'false' #
# ------------------------------------------------------ #
OLLAMA_FLASH_ATTENTION: '1' #
OLLAMA_NUM_PARALLEL: '1' #
OLLAMA_MAX_LOADED_MODELS: '1' #
OLLAMA_KEEP_ALIVE: '-1' #
# ------------------------------------------------------- #
volumes: #
- "Data:/root/.ollama" #
# ------------------------------------------------------- #
ports: #
- '3001:11434' #
# ------------------------------------------------------- #
extra_hosts: #
- "host.docker.internal:host-gateway" #
# ------------------------------------------------------- #
#############################################################
#
#
#############################################################
volumes: #
Data: #
#############################################################
```
document fetched 2026-11-03 00:42 · canonical https://reddit.com/r/ollama/comments/1xm0vwm/je_comprend_pas_pourquoi_ollama_utilise_le_cpu