OpenAI's Jalapeño chip targets faster, more efficient AI inference

OpenAI says its first custom inference chip, Jalapeño, delivered higher throughput per watt and lower latency across several AI models in new benchmark results.