How a 20B MoE was pre-trained for under $10/hour
const_reborn · x · 2026-07-20
A comment highlights a striking training setup from @jondurbin: a 20B MoE model was pre-trained for under $10/hour of compute.
It was not run on a cluster. Instead, the setup used eight rented single-L40S VMs spread across two continents, plus a few 4090s and a 5090, while tolerating about 6 seconds per step. The point is that pre-training — long assumed to require a large centralized cluster — can be done surprisingly cheaply with a distributed setup.
More from Infra
- SpaceX strikes $40B deal with Apollo Global to lease AI chips and data center hardware — Polymarket · 2026-10-26
- India Needs at Least 5GW of Compute by 2028 to Stay Relevant, User Argues — EAccelerate_42 · 2026-10-26
- SpaceX AI revenue up 830%, Claude designs 318K antivenom proteins in 64 hours — PeterDiamandis · 2026-10-26
- Meta plans up to $95 billion stock sale to fund AI buildout — The Decoder · 2026-10-26
- Pandora M1: first prototype of a physical device for distributed GPU monitoring — tarantulae · 2026-10-26
- INT21's AI agent swarm rewrote Drama 3's inference runtime in Rust within one hour, beating SGLang-Omni — bingxu_ · 2026-10-26