Power sharing fits 192 GB300 GPUs in 140-GPU power budget, boosting tokens/s by 49%
julsimon · x · 2026-09-28
Citing NVIDIA and Nscale's own test (Sep 27), 140 GB300 GPUs drew only 166.2 kW of a 264.4 kW budget. With power sharing, 192 GPUs fit the same budget, delivering 49.2% more tokens per second — at the cost of a 17% longer wait for the slowest 1% of requests. Part of AI's power shortage, the author notes, is power reserved and never used.
More from Infra
- Sarvam AI Launches Token Factory: India-Hosted Inference for Open Models — itsOmSarraf_ · 2026-11-16
- BIS Issues Interim Final Rule Addressing Remote Access to Advanced AI Chips — apples_jimmy · 2026-11-16
- Jailbroken PS4 runs a 9B open-source model at up to 43 tok/s by talking directly to its 2013-era GPU — AIFlow_ML · 2026-11-16
- Mobile AI Factory: agents auto-build on-device SLMs with 19x less GPU time — pmttyji · 2026-11-16
- Stratechery's Fictional Obituary: Imagining Jensen Huang's 2026 Exit From Nvidia — Gregory_C_Allen · 2026-11-16
- vLLM serves DeepSeek V4.1-Flash at 954K tokens/s per GPU, $38.96/GPU-hour on GB300 NVL72 — bookwormengr · 2026-11-16