IREN Achieves NVIDIA Exemplar Cloud Certification for HGX B300 Architecture Validating Enterprise AI Performance and Engineering Standards
IREN has obtained the NVIDIA Exemplar Cloud certification on the company HGX B300 system architecture. This verification helps demonstrate the range of performance that the artificial intelligence data center hardware being used in the system architecture can deliver. The hardware platform was evaluated according to the protocol outlined by IREN and verified through end to end cluster training on real world workloads to meet NVIDIA performance requirements. It confirmed the clusters provided by the company perform on par with component level hardware even when operating under sustained load profiles.
The NVIDIA Exemplar Cloud label is one of the few programs run to help validate performance claims using a defined standard. In this instance it came about from end to end training workloads run over a set of multi node GPU clusters that have complex interconnection topologies. This measure includes large pretraining processes across state of the art model architectures and evaluated the overall training speed and scale, network load, and cluster stability under sustained workload.
The certification can only be granted if the infrastructure meets the stated standards across all metrics in this package and not from a peak or max performance at a subset of the tests. This demonstrates that the system has undergone capacity planning and extensive test use to work well under difficult circumstances in specified workload scales.
The extensive engineering effort needed to deliver such a high performance validated system was a joint effort from both vendors. The design encompassed multiple levels of collaboration including physical data center architecture, hardware integration, code tune ups, and network optimization. The end result stems from a shared vertically integrated approach where the same group designs, builds, and operates the data center and hardware enabling rapid design iterations and root cause resolution.
This validation provides enterprise purchasers with third party evidence that the platform can meet the dense workloads needed in cutting edge foundation models and scientific research simulations. Denis Skrivankoff, IREN head engineer, said:
"Achieving NVIDIA Exemplar Cloud status is evidence of the performance and execution standards we are bringing to AI Cloud. Customers running large training and inference workloads need infrastructure that performs predictably under load, and that is what we have engineered for."
Additionally, NVIDIA recognized the considerable engineering effort to achieve this benchmark. Warren Barkley, NVIDIA Vice President of Product Management, said:
"IREN's achievement of NVIDIA Exemplar Cloud status reflects deep engineering collaboration between our teams and the quality of infrastructure behind IREN’s AI Cloud, giving enterprises confidence to run their most demanding training workloads at scale."
This powered up infrastructure may provide a more predictable environment for long range project planning. The system architecture validated HGX B300 clusters are meant for teams building multilingual multimodal systems, performing domain specific model fine tuning, or executing large scale data analysis. By combining validated hardware performance with a fully integrated vendor owned data center infrastructure the platform seeks to eliminate the operational concerns that increase the total cost of adopting more ambitious AI.
