OpenAI's Jalapeño Chip Reportedly Beats Nvidia Blackwell on Perf/Watt
petrusenko_max · x · 2026-08-26
OpenAI's in-house Jalapeño chip, built in approximately 16 months, reportedly outperforms Nvidia's Blackwell in performance-per-watt across nearly all inference scenarios involving single-token prediction without model tuning. It is designed as a generalized AI inference chip rather than being specialized for a single model.
More from Infra
- NVIDIA Vera Rubin NVL72 racks enter production with 100% automated assembly — nvidia · 2026-08-26
- Kern: A Fast, Secure Container Runtime for Local LLM Code Execution — quiet-systems · 2026-08-26
- Hardware Prices Won't Recover: Choosing LLM Inference Hardware — nemuro87 · 2026-08-26
- For the price of one DGX Spark, you can build 2x 3090s with 128GB RAM — QuixiAI · 2026-08-26
- Train a private SLM on iMessages locally with MLX — romainhuet · 2026-08-26
- Benchmark Pitfalls: Incorrect Measurement Boundaries Distort Data — WirelessLife · 2026-08-26