Meta’s custom AMD MI400 chip is said to use 144GB of HBM, not 432GB
rwang07 · x · 2026-07-22
A SemiAnalysis report says Meta’s custom AMD MI400-series chip will be about half the size of a normal MI455X chip.
- The chip is reportedly optimized for recommender workloads and cost per memory bandwidth.
- Meta is said to use about 144GB of HBM instead of 432GB.
- The post is a reaction to that hardware-specific design choice, which points to Meta’s continuing push toward custom inference hardware.
More from Infra
- AWS on agent prompt caching: explicit + implicit patterns to slash token costs — AWS ML Blog · 2026-10-27
- Thieves spoofing GPS steal $100M+ in data center cargo since June — Ars Technica AI · 2026-10-27
- Lam Research: weekly token usage up ~250x in 18 months — Beth_Kindig · 2026-10-27
- AWS shows how to build enterprise agentic retrieval on Bedrock with CDK — AWS ML Blog · 2026-10-27
- Encrypted GPT-2 speedrun challenges FHE inference: best is 3.031 s/token, goal sub-1s — tensorqt · 2026-10-27
- Open-source Latch compiles PyTorch into training megakernels, 2.75x faster on H100 — sguada · 2026-10-27