OpenAI's Jalapeño Chip Benchmarks: 1.9x Efficiency Over Nvidia GB200
AI寒武纪 · wechat · 2026-08-26
OpenAI revealed benchmarks for its first in-house inference chip, Jalapeño. Tested on SemiAnalysis's InferenceX, Jalapeño delivers 1.5x to 1.9x more tasks per watt and 1.7x to 3.6x lower end-to-end latency compared to Nvidia GB200/GB300, with performance in high-interaction scenarios reaching 2.1x to 4.1x. The chip runs at 550W or lower under load despite a 700W rating. Its architecture features full-stack hardware-software co-design to minimize data movement and explicitly manage KV cache, optimizing for agent workloads. The 9-month development cycle leveraged AI for design and verification, with AI-generated code running 1.8x faster than human code in some cases. OpenAI plans deployment by year-end, with Gen 2 and Gen 3 chips already in development.
More from Infra
- SGLang ships Day-0 support for FLUX 3 Image Klein: 10 reference images, BF16/FP8/NVFP4 variants — ying11231 · 2026-10-16
- NVIDIA's Nemotron 3.5 Super VL hits CoreWeave: 12B-active multimodal MoE at $0.25/M input tokens — wandb · 2026-10-16
- Chip equipment beats bull case: 2026 WFE seen at $153B, $278B by 2030 — BenBajarin · 2026-10-16
- Steam now launches in 5.23s, beating Epic's 5.47s — and vLLM gets roasted for slow startup — AlpinDale · 2026-10-16
- Shadeform to Announce Six New Features on Oct 19: Lower GPU Costs, More Suppliers — _jaydeepkarale · 2026-10-16
- Fastino launches Training API for SFT/RL, claiming 25-85% cost cut vs Tinker — kalyan_kpl · 2026-10-16