Routing spatial AI pipelines across heterogeneous compute: CPU to L40S to H100
PrateekVJoshi · x · 2026-11-18
A practical observation on visual/spatial AI: compute within a single trajectory is heterogeneous — CPU preprocessing, 3D reconstruction on an L40S, world model inference on an H100, rendering on another GPU.
Rather than running the whole pipeline on one expensive machine, route each step to the hardware it actually needs. The gnarly part isn't compute but managing state, data movement, and latency across machines.
More from Infra
- NVIDIA Is No Longer Just a Supplier but an Investor, Customer, and Backer of the AI Ecosystem — sidjustice_ · 2026-11-18
- NVIDIA put $30B into OpenAI and billions more — the essay reframing Jensen Huang — sidjustice_ · 2026-11-18
- Running a 180B MoE Model on a 16GB MacBook: Experts Offloaded to SSD, 14× Faster Than llama.cpp — XiongChenyan · 2026-11-18
- Polymarket Puts 90% Odds on Nvidia Beating Earnings in First Report Under Interim CEO — Polymarket · 2026-11-18
- Power is the bottleneck: NYC event explores the grid gap becoming a market of its own — ditzikow · 2026-11-18
- Microsoft's HorizonDB goes GA on Azure with up to 8X self-managed Postgres performance — satyanadella · 2026-11-18