From cd127a23e0c27bfeb0c82ab7cc4f1fa4f7a1fe0c Mon Sep 17 00:00:00 2001 From: poprhythm Date: Sun, 13 Sep 2026 13:27:37 +0000 Subject: [PATCH] Remove obico's GPU reservation to free VRAM for ollama obico's own comment already notes its :cuda tag falls back to CPU inference on this driver/GPU combo, but it was still reserving 826MB of VRAM it never used productively. That squeeze was forcing ollama to offload only 20/29 model layers to GPU, pushing the rest onto CPU and causing severe latency (multi-minute LLM calls) that stalled SUB/WAVE's track picking. Claude-Session: https://claude.ai/code/session_01L7Rwa6guD5wK8F8tWQwcJX --- obico/docker-compose.yml | 14 ++++---------- 1 file changed, 4 insertions(+), 10 deletions(-) diff --git a/obico/docker-compose.yml b/obico/docker-compose.yml index cd9bc31..73bade6 100644 --- a/obico/docker-compose.yml +++ b/obico/docker-compose.yml @@ -7,9 +7,10 @@ services: - obico-internal obico: - # :cuda tag attempts GPU but falls back to CPU if CUDA version mismatch - # (libcudart.so.11.0 required; GTX 1660 SUPER with driver 590 has CUDA 12). - # ML inference still works on CPU - adequate for a single printer. + # :cuda tag falls back to CPU anyway on this GPU (libcudart.so.11.0 + # required; GTX 1660 SUPER with driver 590 has CUDA 12) - adequate for a + # single printer. No GPU reservation here so it doesn't reserve VRAM + # ollama needs for its own model (was forcing partial CPU offload there). image: ghcr.io/imagegenius/obico:cuda container_name: obico restart: unless-stopped @@ -33,13 +34,6 @@ services: - /srv/obico/config/model_cache:/model_cache/ml_api ports: - "3334:3334" - deploy: - resources: - reservations: - devices: - - driver: nvidia - count: 1 - capabilities: [gpu] networks: - npm-network - obico-internal