Krea 2 SVDQuant checkpoints cut ComfyUI runtime about 2× on NVIDIA GPUs
LightAppropriate624 · reddit · 2026-07-25
A community project releases SVDQuant quantized checkpoints for Krea 2 in ComfyUI, claiming up to 2× speedup with no major quality cliff.
- The author says FP8 is only useful on GPUs with native FP8 tensor cores; on older cards it may be cast back to bf16 and can even be slower.
- Their focus is INT8 and W4A4, which they say work back to Turing / RTX 20-series.
- Benchmarks on an RTX 3090 at 1024×1024, 8 steps show BF16 at 21.3 s warm, while W4A4 variants land around 10.1–10.2 s warm.
- The repo also includes a loader fix: ComfyUI’s standard LoRA loader only patches about 12% of layers on these quantized models, and the custom loader handles the full network.
- The project is community-made and not affiliated with Krea.
Related event: Quantized Krea 2 Boosts ComfyUI Inference Speed by Over 2x(2 posts)→
More from Infra
- Anthropic IPO filing shows $974M Q2 operating profit on $11.6B revenue, up 1,368% YoY — kimmonismus · 2026-10-26
- Pandora M1: first prototype of a physical device for distributed GPU monitoring — tarantulae · 2026-10-26
- E2E Networks to add ~5.4MW for NVIDIA HGX B300, lifting capacity to 14.6MW — cneuralnetwork · 2026-10-26
- INT21's AI agent swarm rewrote Drama 3's inference runtime in Rust within one hour, beating SGLang-Omni — bingxu_ · 2026-10-26
- Anthropic's S-1 reveals $434.5B in contracted compute, with SpaceX the largest counterparty — _AustinCalvert_ · 2026-10-26
- Everyone ships AI sandboxes, yet Modal and Daytona still top growth charts — FinanceYF5 · 2026-10-26