ComfyUI LTX Director loading has become much slower since July 18–19

guayar · reddit · 2026-07-20

A ComfyUI user reports a major slowdown in the LTX Director 2 / DualCLIPLoader pipeline since July 18–19.

• Encoder loads that used to take only seconds now take minutes, sometimes 10–15 minutes.
• The same slowdown appears when loading the KModel: VRAM fills slowly and sampling starts much later.
• The issue seems to affect multiple steps in the workflow, not just DualCLIPLoader.
• The system is an AMD Radeon AI PRO R9700 on Ubuntu 24.04 with ROCm 7.14, PyTorch 2.12, Triton 3.7.1, and ComfyUI 0.28.0.
• The author has already tried multiple reinstalls and disabled various VRAM/offload settings, but the slowdown remains.
• They ask whether recent changes in ComfyUI, safetensors, MixedPrecisionOps, or LTX Director could be responsible.

Original post →

More from Infra

• LLM Serving Metrics Thread: Why TPOT and Uptime Make or Break User Experience — abhijithneil · 2026-09-11
• PlanetScale launches sharded Postgres: 768 servers acting as one, 1PB scale — dhruv2038 · 2026-09-11
• Can a 7900 XTX 24GB run Qwen locally? Reddit seeks ROCm tok/s benchmarks — thenomadexplorerlife · 2026-09-11
• RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11
• SF Compute founder: buying compute is 'an absolutely awful experience' right now — IgorCarron · 2026-09-11
• SmolVM open-sources persistent computer infrastructure for agents that outlive chat sessions — aniketmaurya · 2026-09-11

Infra channel →