Nvidia's MaxLPS squeezes 30-40% more from the same GPUs; Ian Buck says Wall Street misses it
bookwormengr · x · 2026-10-17
Nvidia CUDA creator Ian Buck called out Wall Street at OCP SJ 2026 for failing to model MaxLPS upside. One of Nvidia's most impactful systems, MaxLPS extracts 30-40% more performance from the same GPUs within the same power envelope by dynamically instructing GPUs via BMC to draw more power when others don't need it, while guaranteeing the aggregate budget isn't exceeded (avoiding tripping and fines). The accompanying diagram shows a 540kW budget with static provisioning stranding 170kW, while MaxLPS dynamic provisioning consumes 475kW and fits one extra rack. The author notes the system is easy to copy, so it may not help the stock.
More from Infra
- VaeZip: open-source codec stores VAE latent datasets 5-10x smaller with near-zero quality loss — TensorForger · 2026-10-17
- One plain-text file runs time on every device; UCLA prof maintains tzdata as volunteer since 2005 — PMinervini · 2026-10-17
- Google Cloud ships new docs explaining GPUs vs TPUs, from CUDA cores to use cases — rseroter · 2026-10-17
- Selling 10 iPhones could fund enough GPUs for a local AI lab — AIFlow_ML · 2026-10-17
- E2B founder: most agent sandbox providers are just Firecracker wrappers on AWS — zeeg · 2026-10-17
- Data movement, not compute, is AI's real cost: near-memory investment map in 3 layers — FinanceYF5 · 2026-10-17