Counterintuitive VRAM trick: keeping Qwen 3.8 loaded doubles video-gen speed on RTX 5090
Modern_Art_Official · reddit · 2026-09-19
A Reddit user with an RTX 5090 and 96GB RAM running a 5-second 0.98MP video workflow (Dasiwa hybrid model, nvfp4 text encoder, Kijai's experimental VAE, 5 image references) noticed runtimes jumping from 174s to 350-430s when Qwen 3.8 was ejected.
- The culprit is ComfyUI's dynamic vram: even when everything fits in VRAM, the experimental VAE still triggers slow Shared GPU Memory usage (0.7GB spillover when running alone).
- With Qwen left resident, dynamic vram only uses 29.4/76.7GB, keeping runtimes under 3 minutes — at the cost of GPU temps hitting 80°C.
Takeaway: keeping a large model loaded can "occupy" the dynamic vram mechanism and prevent slow paging — a useful counterintuitive config for local video generation.
More from Infra
- Malaysia poised to become a compute superpower with both US and Chinese chips by next year — basedjensen · 2026-11-04
- ByteDance details USDB storage system: 85.8% lower write latency for recommender sequences — _reachsumit · 2026-11-04
- Oracle cancels $7B chip leasing deal with Tencent, FT reports — SumitGup · 2026-11-04
- Inference engine shrunk to 1023 bytes, still runs SmolLM2-135M — cneuralnetwork · 2026-11-04
- Where Residential Proxy IPs Come From: AMA with Massive's CEO — Dylan-Hayes · 2026-11-04
- Musk says space data centers are 'looking pretty appealing right now' — elonmusk · 2026-11-04