Oracle Cloud Infrastructure (OCI) has achieved NVIDIA Exemplar Cloud for NVIDIA GB200 NVL72 platform, demonstrating OCI’s ability to deliver consistent, reproducible AI performance on NVIDIA Blackwell infrastructure at scale. This milestone builds on OCI’s earlier Exemplar Cloud validation for NVIDIA HGX B200 system and extends that rigor to a rack-scale platform for next-generation AI training and inference.
Why it matters
NVIDIA Exemplar Cloud gives customers a clear way to assess whether cloud infrastructure can deliver benchmarked performance in real-world environments. To earn the designation, providers must show they can run NVIDIA reference workloads within 5% of NVIDIA target results. OCI’s validation on the BM.GPU.GB200v3.4 compute shape powered by GB200 NVL72 shows customers can expect predictable performance at scale across AI training workloads.
High-performance architecture for large-scale AI
The BM.GPU.GB200v3.4 shape is built for both scale-up and scale-out efficiency. Within each rack, fifth-generation NVIDIA NVLink™ fabric connects all 72 Blackwell GPUs in a single scale-up domain with 1.8 TB/s per GPU of all-to-all bandwidth. At the node level, the shape includes a NVIDIA CX-7 with 2×200 Gbps links for the frontend network and 4 NVIDIA ConnectX-8 NICs with 2 x 400 Gbps links, delivering 3.2 Tbps of RDMA networking bandwidth per node, helping customers scale distributed training across OCI AI clusters.
This combination of rack-scale compute and high-throughput networking is critical for modern AI workloads. Customers need infrastructure that can sustain performance across real, distributed workloads, not just deliver peak benchmark numbers. Exemplar Cloud validation helps demonstrate that OCI can provide that consistency on Blackwell-based systems.
Built for scale and reproducibility
OCI’s GB200 NVL72 deployments reflect engineering across compute, networking, systems software, and cloud operations to make rack-scale AI infrastructure practical to deploy and operate. This validation also reflects Oracle’s continued collaboration with NVIDIA to help customers move from experimentation to production-scale AI with greater confidence.
“Achieving NVIDIA Exemplar Cloud validation on GB200 NVL72 reflects OCI’s continued focus on delivering reproducible performance at the scale our AI customers require,” said Ben Scarborough, SVP, OCI Customer Engineering. “With rack-scale NVIDIA Blackwell infrastructure and ultra-high-bandwidth networking on OCI, customers can run the most demanding training workloads on infrastructure engineered for both performance and operational consistency.”
“OCI’s Exemplar Cloud achievement on NVIDIA GB200 NVL72 illustrates how the company’s rack-scale Blackwell infrastructure delivers training workloads with incredible performance at scale,” said Warren Barkley, vice president of product management at NVIDIA. “It gives customers confidence that BM.GPU.GB200v3.4 can support the most demanding AI workloads on infrastructure engineered for consistent results from node to cluster.”
Raising the bar for cloud-scale AI
OCI believes customers should be able to trust that published performance is something they can achieve. Exemplar Cloud validation on GB200 NVL72 is another example of how Oracle and NVIDIA are advancing cloud-scale AI with rack-scale Blackwell compute, high-performance networking, and validated reproducibility.