Summary
This case study details the deployment of the NVIDIA DGX SuperPOD platform by AMAX for a generative AI company focused on voice synthesis and multimodal content tools. The customer required a scalable infrastructure to support the rapid training and inference of complex AI models. AMAX collaborated with the customer's engineering and operations teams to evaluate compute needs and long-term goals, ultimately recommending a solution comprising 64 DGX B200 systems equipped with 512 NVIDIA Blackwell GPUs. The interconnected systems utilized NVIDIA Quantum-2 InfiniBand for high-speed networking. The deployment included the design of rack layout and power distribution in accordance with NVIDIA's reference architecture, along with comprehensive testing to ensure reliability. The outcome of this deployment is a robust platform delivering 4.6 exaflops for AI training and 9.2 exaflops for inference, providing the customer with a flexible foundation for future AI workload growth.
