Sentdex runs 64 concurrent agents at ~7,000 tok/s on a single desktop GB300
Sentdex · x · 2026-10-21
Sentdex tested Nemotron 120B A12B locally on the ASUS ExpertCenter Pro ET900N G3 (DGX Station architecture, NVIDIA GB300 Grace Blackwell, 748GB coherent memory): 64 concurrent requests at 100 tok/s each, aggregating nearly 7,000 tok/s on one desk-side machine.
- Jokes that "the agents did NOT magically start hacking the planet," and teases that Dario begged him in DMs not to run agent swarms at home
- The tower supports NVIDIA NemoClaw for local agent workflows and local deployment of trillion-parameter models
More from Fun
- steipete laments AI slop flooding X replies, auto-mutes them with Claude — steipete · 2026-10-21
- AI idol and dating shows are now a real thing in China, 92% of fans are young women — ShakeelHashim · 2026-10-21
- Pat Hanrahan's RacketCon keynote on Einstein summation, Penrose diagrams and APL wows viewers — blelbach · 2026-10-21
- Dev: AirPods Announce Notifications is one of the worst product experiences ever — nickbaumann_ · 2026-10-21
- A real larva's connectome — 2,675 cells, 14,066 connections — now swims in a 3D simulation — eptwts · 2026-10-21
- Snap Hires Robert Downey Jr. as Chief Customer Experience Officer to Sell Specs AR Glasses Against Meta — adariostrange · 2026-10-21