Speeding Up Local ComfyUI Inference with SSDs
ganrocks007 · reddit · 2026-07-16
The author tested model file placement (SATA SSD vs. NVMe SSD) using their local setup (RTX 3060 12GB, 32GB RAM, --lowvram, Krea 2 + Qwen 3 VL FP8).
Key findings:
- Moving text encoders (like Qwen3 VL) from SATA to NVMe reduced generation time from 70s to 40s, making interactions noticeably faster.
- Placing diffusion models/UNet on NVMe yielded almost no perceptible improvement, as sampling primarily occurs in GPU/VRAM.
Recommendation: With limited NVMe space, prioritize frequently read text encoders, CLIP, and T5 on NVMe, while larger UNets can remain on SATA.
More from Infra
- How to build a PostgreSQL-backed semantic search pipeline with pgvector and Ollama — KhuyenTran16 · 2026-07-21
- NeurIPS 2026 workshop calls papers on on-device intelligence — YiMaTweets · 2026-07-21
- Milled from Solid Aluminum: AI Rig Multi-GPU Case for Local Compute — dee_hw · 2026-07-21
- FutureCaribbean’s Buildathon offers $50K, H200 compute, and an NYSE pitch — HeyAmit_ · 2026-07-21
- A new series tests which data-science workflows can run on GPUs today — pandeyparul · 2026-07-21
- Former AWS operator says Bedrock margins can beat SageMaker as agentic AI lifts CPU demand — RihardJarc · 2026-07-21