OpenAI's Jalapeño Chip Beats Nvidia GB200 in Efficiency
Beth_Kindig · x · 2026-08-30
OpenAI claims its new Jalapeño chip delivers 1.5X to 1.9X higher tokens per watt and 1.7X to 3.6X lower latency compared to Nvidia's GB200 and GB300 on models like GPT-OSS 120B, DeepSeek R1, and Kimi K2.5 1T at peak throughput.
More from Infra
- M5 Max + dual 3060s: How to utilize this frankensetup? — Odd-Environment-7193 · 2026-08-30
- How Starship V3 Changes the Scale of Orbital AI Infrastructure — XFreeze · 2026-08-30
- Developer Dumps Qualcomm for Rockchip to Ship Edge AI Devices Faster — kscottz · 2026-08-30
- Empire of AI's datacenter water figure is off by ~4,500x: liters vs cubic meters — altryne · 2026-08-30
- Community fork enables MiniMax H3 on dual GPUs with live preview support — karma3u · 2026-08-30
- Own your harness, and if possible, own the model layer too — omarsar0 · 2026-08-30