[AINews] Hot Chips: OpenAI’s Jalapeño, Cerebras CS-5, Groq 3 LPX, Apple M6
OpenAI’s first custom inference chip is being presented as a direct efficiency and latency challenge to NVIDIA’s current systems.
Latent Space says Jalapeño delivered 1.5–1.9x more work per watt at peak throughput and 1.7–3.6x lower end-to-end latency than GB200/GB300 systems in OpenAI’s tests. The chip is rated at 700W but reportedly stayed at or below 550W in the tested runs. OpenAI says deployment into its own infrastructure starts by year-end, with second- and third-generation chips already in development. The note also says GPT-Astra and Codex helped optimize low-level kernels for Jalapeño, in some cases beating human-written implementations. Latent Space's note
Latent Space says Jalapeño delivered 1.5–1.9x more work per watt at peak throughput and 1.7–3.6x lower end-to-end latency than GB200/GB300 systems in OpenAI’s tests. The chip is rated at 700W but reportedly stayed at or below 550W in the tested runs. OpenAI says deployment into its own infrastructure starts by year-end, with second- and third-generation chips already in development. The note also says GPT-Astra and Codex helped optimize low-level kernels for Jalapeño, in some cases beating human-written implementations. Latent Space's note
score 6