Transcribing 11.3 hours of Nixon tapes in under 3 minutes with Nemotron 3.5 for ~$0.30
vanstriendaniel · x · 2026-11-06
Developer Daniel van Strien benchmarked NVIDIA's Nemotron 3.5 Transcribe on a single RTX PRO 6000 via vLLM: 11.3 hours of 1971 Nixon tapes transcribed in under 3 minutes, with the whole Hugging Face Job (setup included) costing about $0.30.
- Precision trade-off: NVFP4 finished in 2.1 min (322x realtime), FP8 in 3.0 min (222x), BF16 in 5.0 min (135x); NVIDIA's model card puts NVFP4 at 3.6% higher error than BF16
- Batch workflow: run vLLM inside a Hugging Face Job and drive it with his tool saturate, which auto-tunes concurrency — a 79.7-hour transcription run took just under 10 minutes
- The full recipe is open-sourced as a uv-script on Hugging Face: one command in, outputs include speaker-tagged txt, word-level JSON, srt, and parquet
Related event: Nemotron 3.5 Transcribe Handles 11-Hour Audio in 3 Minutes for $0.30(2 posts)→
More from Infra
- Baidu open-sources ERNIE-5.5-Flash-Base: 280B MoE with 14B active params and 1M context — kimmonismus · 2026-11-06
- Data center NIMBYism is a corporate good-neighbor problem, says ex-Kraft vet — BecauseCulture · 2026-11-06
- Used AMD MI100 home-lab test: a 3D-printed shroud cools 300W at just 2,400 rpm — Much-Farmer-2752 · 2026-11-06
- Intel down 46% in 3 months: the bull case rests on 18A costs down 40% and $10B extra data center revenue — tengyanAI · 2026-11-06
- AI bots hit a site 5x more than humans: 1.23M requests, ChatGPT needs ~100 fetches per visitor — marclou · 2026-11-06
- Qwen3.8-27B as local coding agent burns 16k tokens thinking, never calls tools — previaegg · 2026-11-06