AivexaNewsSearch
AI news for builders and product teamsChecked every hour

How NVIDIA Groq 3 LPX Deterministic Execution Drives Power-Efficient High-Interactivity Inference on NVIDIA Vera Rubin

Collected Sep 30, 2026

NVIDIA published a technical explanation of how the Groq 3 LPX rack contributes power efficiency to the NVIDIA Vera Rubin platform, centered on the Groq 3 LPX deterministic execution model.

According to NVIDIA, that model lets the LPU compiler build a schedule of when each piece of data moves to which compute unit and when each operation executes, down to the clock cycle, extending to all 256 LPU chips in the rack. From that schedule, the compiler can predict electrical current draw per cycle.

Two technologies use that predictability. Preemptive Power (PEP) prepares the power-delivery system for demand changes before they arrive, and Clock Period Synthesis (CPS) shapes how abruptly demand rises and falls. CPS can lengthen the clock cycles with the highest current spikes, lowering the rate of current increase. NVIDIA says the two together reduce the voltage guardband, the electrical safety margin maintained for chips.

NVIDIA reports internal testing on Groq 3 LPX systems showed more than 60% less voltage drop. It estimates this will lead to a high single-digit percentage decrease in the baseline voltage the electrical system must continually provide, contributing to higher percentage power decreases because power is proportional to the square of voltage.

NVIDIA states the deterministic execution can reduce power for the same workload by a potentially low-double-digit percentage compared with a similarly specified, nondeterministic system, and that Groq 3 LPX brings these controls to the Vera Rubin platform in the second half of 2026. At the platform level, NVIDIA says pairing Groq 3 LPX with Vera Rubin NVL72 enables up to 35x higher throughput per megawatt versus previous-generation GB200 NVL72 for 2T+ parameter models at long context and high interactivity.

The post describes factory-level NVIDIA DSX MaxLPS software, which NVIDIA says shifts power between racks as demand shifts, allowing operators to recover stranded power, provision up to 40% more GPUs within the same site-power envelope, and deliver 35% higher token throughput. Within each Vera Rubin NVL72, NVIDIA says rack-level capacitors and Intelligent Power Smoothing software absorb bursty power spikes so factories can be planned around sustained demand rather than worst-case peaks.

Read at NVIDIA Developer Blog

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Power is a defining constraint for AI factories. As AI workloads demand a full compute platform to serve them, each component of that platform must maximize...