The cluster of 64 DGX B200 systems delivers up to 9.2 exaflops of inference performance for a leading developer in synthetic voice and multimodal content.
AMAX, a leading provider of advanced infrastructure solutions in artificial intelligence, HPC, edge computing, and GPU-based systems, has announced the deployment of a powerful NVIDIA DGX SuperPOD composed of 64 NVIDIA DGX B200 systems, along with NVIDIA AI Enterprise software. This deployment achieves 4.6 exaflops of AI training power and 9.2 exaflops for inference and has been allocated to one of the most prominent developers of generative artificial intelligence.
The aim of this infrastructure is to accelerate the development of advanced voice synthesis models and content generation, with an architecture prepared to tackle current and future challenges in the generative AI sector.
Scalability and Exa Power with Modular Architecture
AMAX’s solution is based on a SuperPOD modular architecture, enabling efficient and predictable scalability for intensive workloads. Each Scalable Unit (SU) consists of 32 DGX B200 nodes, allowing for easy capacity expansion or reduction. The 64 deployed nodes total 512 NVIDIA Blackwell GPUs, interconnected via the NVIDIA Quantum-2 InfiniBand network, which offers 400 Gb/s bandwidth and In-Network Computing, ideal for distributed and low-latency workloads.
This infrastructure has been designed to provide shorter deployment times, facilitating the transition from design to execution in weeks rather than months, significantly reducing the time required to iterate models and bring new AI applications to market.
A Turnkey Deployment with Engineering Precision
AMAX has led the entire deployment process, from architecture to final validation, leveraging its experience in high-performance environments. Key phases include:
- Architecture design and optimized placement of racks and power for high-density environments.
- Rack-scale integration and activation, with full cable management and compliance with strict data center requirements.
- Network topology and structured cable management, using NVIDIA Quantum QM9700 switches and a high-performance InfiniBand network.
- Validation and performance testing, including benchmarks, stress testing, and fine-tuning to ensure optimal operation.
- Ongoing enterprise support, with post-deployment maintenance and optimization services.
Driving the Next Generation of Generative AI
Don Wang, Director of Solutions Architecture at AMAX, stated:
“By combining AMAX’s deployment expertise with the latest NVIDIA DGX systems, we offer a turnkey AI solution that maximizes performance, scalability, and efficiency for our clients.”
Additionally, Tony Paikeday, Senior Director of AI Systems at NVIDIA, added:
“The most advanced AI models require platforms capable of delivering fast and innovative results. Thanks to AMAX’s expertise with NVIDIA DGX SuperPOD, organizations can accelerate the entire AI lifecycle, from training to inference, achieving unprecedented returns on investment.”
A Comprehensive Offering to Scale Enterprise AI
AMAX not only provides the deployment of DGX B200 but also comprehensive solutions that include DGX H200, DGX GB200, and the full-stack NVIDIA DGX software, tailored to meet the needs of various artificial intelligence and high-performance computing workloads.
With this implementation, AMAX reinforces its position as a strategic integrator for generative AI projects, establishing itself as a key partner for organizations looking to scale their artificial intelligence capabilities with agility, power, and efficiency.
via: amax