Nebula AI's HGX B300 racks validated by Nvidia for AI cloud use
Summary
NVIDIA has officially validated third-party B300 racks, confirming they meet its reference architecture standards for AI infrastructure deployments. These racks achieved the fastest single-node times in MLPerf 6.0 testing for large language models, including Llama-3.1-8B and GPT-OSS 20B. Production clusters using the cleared hardware are already operational on the Nebius AI Cloud, showcasing the reliability and performance of the new infrastructure.
Tokens
$NBIS$NVDA
Analysis
NVIDIA: NVIDIA designs and supplies graphics processing units along with full-stack AI hardware and software solutions used in data centers and accelerated computing. The company conducted the grading and reference architecture validation for the B300 racks highlighted in the news, confirming they meet performance and compatibility standards. Nebula AI: Nebula AI, operating as Nebius, is a European provider of AI cloud infrastructure and data center services focused on high-performance computing for machine learning workloads. In the reported development, its production HGX B300 racks have been validated by NVIDIA and are now running live clusters on the Nebius AI Cloud platform. Deployment: Production clusters using the cleared hardware are already operational within a major AI cloud environment. Benchmarking: The validated racks achieved leading single-node performance results in MLPerf 6.0 testing for specified large language models. Certification: NVIDIA has completed validation of third-party B300 racks that match its reference design for AI infrastructure deployments.
Categories
techaimachine_learning