ComfyUI SeedVR2 Video Upscaler v1.5.7 Keeps TensorRT VRAM Under 6GB
Zestyclose_Bake3680 · reddit · 2026-10-01
ComfyUI-SeedVR2-VideoUpscaler-with-TensorRT shipped v1.5.7, targeting transient VRAM spikes in the DiT stage.
- After reports from RTX 4050 6GB and RTX 5060 8GB users, the author re-measured VRAM across the whole pipeline and found the spikes came from the DiT processing stage, not TensorRT itself.
- The code already supports ConvRot INT8/NVFP4 quantization, which in principle keeps VRAM low, but momentary spikes persisted — fine on the author's RTX 5060 Ti 16GB, a blocker for small-VRAM users.
- With spikes now suppressed, under specific conditions the 3B NVFP4 model runs the entire TensorRT pipeline in under 6GB VRAM, making video upscaling viable on 6GB/8GB cards.
Full technical guide is on the GitHub v1.5.7 release page.
Related event: SeedVR2 Upscaler Plugin Update Cuts TensorRT VRAM Needs Below 6GB(2 posts)→
More from Infra
- Padding trick lets vLLM run tp=6: 27B model at 50 tok/s on six 7900 XTX GPUs — Biomass23 · 2026-10-01
- Chutes team on Bittensor Subnet 64 may have stumbled on a new way to train models while fixing inference economics — markjeffrey · 2026-10-01
- Ornith-1.5 DFlash draft models deliver up to 2.54x lossless inference speedup — alan_ritter · 2026-10-01
- Phonon-2 on-device ASR model with QAT low-bit quantization lands on HF trending — FermionResearch · 2026-10-01
- Manager locked Teams transcripts, employee used Copilot to dig JSON URL out of page source — TheBigCrowbroski · 2026-10-01
- Compute per MW comparison: Nvidia still best price/perf despite prices — Storge2 · 2026-10-01