Nvidia has officially validated the production HGX B300 racks operated by cloud provider Nebius. This Exemplar Cloud status confirms that the infrastructure available to customers aligns with Nvidia's reference architecture for training workloads.
The validation process involved an engineering review and analysis of benchmark results. Nebius, which trades on the NASDAQ under the ticker NBIS, demonstrated that its hardware meets the performance standards set by the chipmaker.
In addition to the validation, Nebius achieved the fastest single-node B300 times in the closed division of the MLPerf Training 6.0 benchmark. The tests were conducted using the Llama-3.1-8B and GPT-OSS 20B large language models.
The HGX B300 platform features 8 Blackwell Ultra GPUs per node, approximately 2.1 TB of HBM3e memory, and fifth-generation NVLink interconnect technology. Nebius has previously received Exemplar status for H200 and GB300 NVL72 configurations.
