Running Krea-2-Turbo on a GTX 1650: 18 minutes per image cut to 30 seconds
A_NASHEX · reddit · 2026-11-20
A Reddit user documented running Krea-2-Turbo locally on a GTX 1650 Super (4 GB): generation started at nearly 18 minutes per image and was tuned down to 30 seconds on the same card, with LoRAs still working.
Settings: 512px, 8 sampling steps. The model weighs 33 GB while the PC only gives it 11 GB of RAM. The author's takeaway: the bottleneck was never VRAM — default loading/scheduling was the real culprit on low-end machines.
More from Infra
- QuixiInfer author: tell Claude 'defeat is not an option' to make it write kernels for your GPU — QuixiAI · 2026-11-20
- Volcano launches Agent Cloud: self-healing deploy platform built for coding agents — rohanpaul_ai · 2026-11-20
- subapi open-sourced: turn your Claude subscription into a local API endpoint — Last-Transition249 · 2026-11-20
- How BMW governs hundreds of agents across 12,500 developers and 500M lines of code — AWS ML Blog · 2026-11-20
- Together launches free Storage Accelerator: ~20x faster warm reads for GPU clusters — togethercompute · 2026-11-20
- Analyst conjectures compute entry ticket may get cheaper; 200K GB300s 'guaranteed-viable' by 2028 — teortaxesTex · 2026-11-20