At Hot Chips, Nvidia details Groq 3 LPX rack and publishes independent inference test

Nvidia's vice president of hardware, Igor Arsovski, presented the Groq 3 LPX rack architecture at Hot Chips 2026. The company also released its first third-party benchmark for the hardware. The LP30-based rack is already in production, according to Nvidia.

The Hot Chips conference has long served as a venue where chipmakers unveil architectural details ahead of general availability, and Nvidia's presentation of the Groq 3 LPX rack by hardware vice president Igor Arsovski follows that tradition. The LP30-based system's production status suggests the architecture has moved beyond the design phase into tangible deployment, a notable signal given the conference's typical focus on near-future silicon. Publishing a third-party inference benchmark alongside the talk represents a deliberate move toward external validation, particularly relevant as AI inference workloads increasingly drive data-center purchasing decisions and as customers demand verifiable performance figures rather than vendor-supplied estimates. The choice to pair architectural disclosure with independent testing underscores how competitive the inference hardware market has become.