Self-hosting full GLM 5.3 takes an 8-GPU rig, roughly $136k at today's prices
TheZachMueller · x · 2026-09-30
HackingDave notes that running full GLM 5.3 requires at least 8x H100s and ideally 8x B300s ($400k in hardware), calling it "absolutely incredible" for offensive testing alongside DeepSeek v4.1. TheZachMueller counters that 8x 6000 Pros suffice — roughly $136k at today's pricing — pricey for individuals but feasible for companies that prioritize it. Smaller GLM variants still perform well on cheaper hardware.
More from Infra
- DeepSeek Partners With Huawei on Chip Software to Challenge Nvidia — pstAsiatech · 2026-09-30
- TensorFlow Taught a Generation of Us ML: Moroney's Take on Picking the Right Tool — lmoroney · 2026-09-30
- Open-source Mac meeting notetaker runs 5 local models in 6.8 GB, notes in 33s — stevyhacker · 2026-09-30
- Oído: open-source 13M-param speech model beats Whisper-tiny on a $5 ESP32-S3 — Significant-Price695 · 2026-09-30
- 1GW of compute yields ~$40B annual profit, equal to 1 million Tesla robotaxis, argues analyst — JOBhakdi · 2026-09-30
- Most data centers refusing to disclose their water and electricity use — Thom2503 · 2026-09-30