# Nebius Receives NVIDIA Exemplar Cloud Validation for HGX B300 AI Training Infrastructure

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/28921
Published: 2026-10-09T12:02:11.000Z
Updated: 2026-10-09T12:02:11.000Z
Section: Technology

> The production configuration uses eight B300 GPUs and recorded 72.0 minutes for Llama 3.1 8B training in MLPerf Training v6.0.

Nebius received NVIDIA Exemplar Cloud validation on Oct. 8 for production HGX B300 infrastructure, adding a tested Blackwell Ultra configuration to its cloud AI-training offerings.

The validation applies to a single-node setup with eight B300 GPUs. Engineers from Nebius and NVIDIA tested the production configuration against NVIDIA’s reference performance recipes, reviewed the results and refined the platform setup.

Nebius previously received the same validation for systems using NVIDIA H200 GPUs and the HGX GB300 NVL72 platform.

In the closed MLPerf Training v6.0 results, the Nebius HGX B300 configuration completed Llama 3.1 8B training in 72.0 minutes while reaching a 3.3 log-perplexity target. It completed GPT-OSS 20B training in 83.6 minutes at a 3.34 log-perplexity target.

MLPerf Training compares how quickly computing systems train models until they reach a defined quality threshold. The v6.0 round included 24 submitting organizations and 95 unique systems.

The benchmark results offer a standardized comparison of training performance, but they do not establish Nebius’ customer pricing, available capacity or results across every workload. NVIDIA Exemplar Cloud validation reflects performance against defined testing requirements and does not guarantee identical results for every customer deployment.