Explore AI / LLM RackScale Solutions
Liquid and air cooled rack designs built for large-scale training and inference.
Solution Brief
Large Memory Pool
Up to 18.4 TB HBM3e per rack supports long context windows and large models, with NVLink and NVSwitch ensuring GPUs stay fully utilized during training.
High Speed Networking
800 Gbps InfiniBand per rack delivers low latency communication and rapid synchronization across nodes, ideal for distributed training and inference workloads.
Scalable GPU Density
Air cooled racks scale up to 32 GPUs with 9.2 TB HBM3e. Liquid cooled racks double capacity to 64 GPUs with 18.4 TB HBM3e in a compact rack footprint.
Training advanced voice models at scale
A leading Gen AI developer partnered with AMAX to deploy a DGX SuperPOD built on DGX B200 systems. The cluster delivers 4.6 exaflops for training and 9.2 exaflops for inference, providing a scalable foundation for fast iteration on voice synthesis and multimodal workloads.
Read Full Story
