NVIDIA begins mass producing Groq 3 LPX racks for Nebius
NVIDIA just began mass production of its Groq 3 LPX racks, marking a big move.
These racks are built for super-fast AI tasks and will be put to work at Nebius, a neocloud provider, later this year.
The new hardware teams up with NVIDIA's Vera central processors and Rubin graphics processors to boost performance.
256 Groq 3 chips per rack
Each rack packs 256 Groq 3 chips, made by Samsung, and can process up to 3,400 tokens per second: great news for anyone into coding or real-time AI apps.
With high-speed SRAM memory inside, these racks cut out slowdowns and make AI more responsive.
Groq chips primarily focus on the "decode" phase of serving AI models.
NVIDIA expects $1T sales by 2027
NVIDIA is ramping up deliveries of its Vera Rubin systems and expects combined sales from Vera Rubin and Blackwell systems to hit $1 trillion by 2027, according to NVIDIA CEO Jensen Huang.
The company faces tough competition from AMD and OpenAI, which are also chasing low-latency solutions.