← Accueil actu | Salle du marché
2026-09-16 · Source：Nvidia Blog
NVIDIA's Vera Rubin NVL72 debuted with leading performance in MLPerf Inference v6.1, with the company framing performance, scaling and software optimization as the drivers of inference economics.
NVIDIA's Vera Rubin NVL72 platform has posted leading performance in its first appearance in MLPerf Inference v6.1, according to a post on the company's blog. The debut gives AI infrastructure buyers their first benchmark-side view of the new system, one that NVIDIA frames around a simple proposition: the economics of AI inference decide whether deployed hardware pays for itself.
In the post, NVIDIA lays out three levers that determine the economics of AI inference: system performance, efficient infrastructure scaling and continuous software optimization. Higher system performance, the company argues, translates directly into more tokens generated — and therefore higher revenue. That framing ties benchmark numbers to the income side of a data center's ledger rather than to technical bragging rights, positioning throughput on industry benchmarks like MLPerf as a proxy for the commercial return a deployment can extract from its compute.
The second lever is scaling efficiency. NVIDIA contends that efficient scaling means throughput grows proportionally as hardware is added, requiring fewer resources to serve users at scale — a claim aimed at buyers planning multi-system clusters, where linear returns cannot be assumed. The third is continuous software optimization, which the company says generates more value from infrastructure investments already in place, extending returns across the life of a deployment.
For buyers, the significance of the v6.1 results lies less in individual scores than in what a debut establishes: a baseline for how Vera Rubin NVL72 performs at launch, and a starting point for the software gains NVIDIA says accumulate over time. Because MLPerf Inference is designed to let customers compare accelerator platforms on equal footing, a leading first showing gives NVIDIA a strong reference point as procurement teams weigh the new system against alternatives.
Mots-clés: NVIDIA · Vera Rubin NVL72 · MLPerf Inference v6.1 · AI Inference · Benchmarks
萬安算交所 AIXX · AIXX · Bourse compute vérifiée · Accéder à la salle