SeedVR2 TensorRT decoder engine released, big speedups at high resolutions
Zestyclose_Bake3680 · reddit · 2026-09-03
Developer ussoewwin released a TensorRT decoder engine for the ComfyUI SeedVR2 video upscaler. The encoder engine was abandoned due to persistent block noise and blurring, but the decoder is where TensorRT helps most, with dramatic decoding speedups at high resolutions. With ConvRot INT8/NVFP4 models allowing much larger batch sizes, TensorRT encoding offers little benefit — fp16 VAE Encode is usually faster. A Blackwell-specific decoder engine is on Hugging Face; optimal batch size varies by setup (41–256 on RTX 5060 Ti 16GB).
More from Infra
- ARBR open-sources an MIT-licensed routing and governance layer for multi-provider LLM traffic — Genie-Tickle-007 · 2026-09-03
- Qwen 27B on a 4060 Ti at 6-7 t/s: What Else Can Squeeze Local Inference Speed? — thatoneshadowclone · 2026-09-03
- ClickHouse CEO: agents have no personas, latency is the key requirement — bibryam · 2026-09-03
- DeepEP + RoCE Doesn't Work, Warns Engineer Rerunning Benchmarks — TheZachMueller · 2026-09-03
- Ant's OmniTable wins VLDB best paper: 5.6x faster curation of 35PB LLM data — 量子位 · 2026-09-03
- Microsoft to Disclose Azure Revenue Quarterly in Major Financial Reporting Overhaul — tomwarren · 2026-09-03