Running video generation on a single 4090: INT8 weights, FP8 decoder, RAM over VRAM
AIFlow_ML · x · 2026-10-25
A practical recipe for running video generation models on a single RTX 4090: INT8 weights and KV cache, FP8 decoder, staged GPU loading of each component, 18 latent history frames kept on GPU with the rest archived in system RAM — "budget for RAM, not just VRAM." The author also argues generated video looking right isn't physically right: use it for perception variety and camera-path tests in robotics, keep controllers on real simulators plus hardware.
Related event: Running LLM and Video Generation Inference on a Single RTX 4090(2 posts)→
More from Infra
- Running a Q2-quantized LLM on a 24GB MacBook Air: 49 t/s via SSD expert streaming — Inside-Pen-8395 · 2026-10-25
- Researchers rebuild city power grid topology from open data, no utility secrets needed — anselm · 2026-10-25
- 'Pls just one more GPU': the begging-for-H100 meme mocking scaling obsession — miniapeur · 2026-10-25
- SpaceX borrows another $40B for Nvidia chips as bond investors balk at space data centers — jonerp · 2026-10-25
- Musk predicts 'stranded compute' in half-built data centers, questioning infinite demand — mjdramstead · 2026-10-25
- Nvidia open-sources Flora, cuts robot video labeling cost 125x vs GPT-6 Astra — The Decoder · 2026-10-25