Free daily briefing on global business news.
The chip, which came from Nvidia's $20 billion Groq acquisition, can generate 3,400 tokens per second in benchmarking tests
Michaela Vatcheva/Bloomberg via Getty Images
Nvidia $NVDA announced Monday that its Groq 3 LPX inference chip has entered full production — a milestone that brings to market the technology behind Nvidia's $20 billion purchase of chip startup Groq's assets in December, the largest deal the company has ever closed.
The Groq 3 LPX is an extension of Nvidia's Vera Rubin platform, designed to accelerate the "decode" phase of AI inference — the stage that determines how fast tokens are generated for individual users. Nvidia packages 256 individual Groq 3 chips into its LPX racks. Samsung fabricates the chip, which packs 500 megabytes of SRAM directly onto the die to avoid the memory bandwidth constraints that slow other inference accelerators, according to CNBC.
