OpenAI says its Jalapeño chip beats Nvidia Blackwell/GB300 on efficiency and speed
OpenAI announced that its custom Jalapeño AI inference chip outperforms Nvidia's Blackwell/GB300 hardware on performance-per-watt and response speed in internal tests. The chip, developed with Broadcom, is inference-focused and not designed to train new models, with no benchmark figures disclosed.
Detected & updated continuously · Source: Nebula
Story subjects
Sources
@PolymarketMoney
BREAKING: OpenAI declares its new Jalapeño chip beat Nvidia Blackwell on delivering “more AI work per watt” and faster response times.
@nknk_3
OPENAI’S JALAPENO CHIP OUTPACES NVIDIA IN TESTS
@CryptoMeterIO
🇺🇸 JUST IN: OpenAI says its custom Jalapeño AI chip shows better performance-per-watt than Nvidia hardware in internal tests. The inference-focused chip is not designed to train new models, and no benchmark figures were disclosed.
@WhaleInsider
JUST IN: OpenAI says its new Jalapeno AI chip beat $NVDA GB300 in power efficiency and response speed during testing. https://t.co/XrrHlz5EGI
@ChainCatcher_
OpenAI 首款自研芯片测试结果曝光：推理能效超过英伟达 据 The Verge 报道，OpenAI 公布了首款自研 AI 芯片 Jalapeño 的最新测试结果。该芯片由 OpenAI 与博通联合开发，专门用于运行已经训练完成的 AI 模型。 基于 SemiAnalysis 的 InferenceX 测试平台，Jalapeño 在运行 gpt-oss-120b、DeepSeek R1