Back-of-Envelope Estimate Puts DeepSeek V4.1 Pretraining at ~5e24 FLOPs
teortaxesTex · x · 2026-09-10
Citing Astra's estimate, a poster puts DeepSeek V4.1's pretraining compute at 5e24 FLOPs — slightly above V3, or about 3.5M H100-hours, doable in 2 weeks on 4K B300s. With 16K B300s, they muse, one could train a 256B-active/10T-token model, or a 64B-active model at 3-4T tokens in 2 months. Unofficial back-of-envelope speculation.
Related event: Estimate: V4.1 Pretraining Took ~5e24 FLOPs(2 posts)→
More from Infra
- MiniMax H3 runs fully on a Galaxy S26: 5-second video with audio in ~7 minutes — BidFunny4862 · 2026-10-27
- Unconventional AI acquires inference chip firm NeuReality in 1,000x efficiency bid — NaveenGRao · 2026-10-27
- Musk: robot RAM demand will dwarf everything else, and 100% of superintelligence compute goes to deep space — elonmusk · 2026-10-27
- RTX 5090 hits $7K as PC gaming community goes into 'denial' over AI boom — PAstynome · 2026-10-27
- Wan2GP lands on Pinokio: 1-click AI video on 6GB GPUs — cocktailpeanut · 2026-10-27
- AWS on agent prompt caching: explicit + implicit patterns to slash token costs — AWS ML Blog · 2026-10-27