Nvidia says its Groq 3 LPX racks delivered 3,400 tokens per second in an Artificial Analysis benchmark running Gemma 4 31B with a 100,000-token input sequence (The Register)

Nvidia's Groq 3 LPX racks achieved 3,400 tokens per second in an Artificial Analysis benchmark. The test used Google's Gemma 4 31B model with a 100,000-token input sequence.

According to Nvidia, this performance makes the Groq 3 LPX racks 4 times faster than the nearest alternative platform. The Register reports that this appears to be a reference to Cerebras, which managed 882 tokens per second under the same conditions.

Groq's LPUs use an SRAM-heavy dataflow architecture designed for high-performance inference serving. This architecture relies on large pools of on-die SRAM, which provides faster memory bandwidth than traditional datacenter GPUs. The third generation of Groq's chips, launched as part of Nvidia's Vera Rubin platform, boasts 150 TB/s of memory bandwidth.

Brief written by urgent.news from Techmeme, The Register — 2 reports on this story. Machine-written — may contain errors; check the original before relying on it.