Dell PowerEdge XE9712 with NVIDIA GB200 NVL72 (Grace Blackwell)

The Dell PowerEdge XE9712 is a purpose-built, rack-scale AI infrastructure platform engineered around the NVIDIA GB200 NVL72 Grace Blackwell Superchip architecture. The system integrates 36 NVIDIA GB200 Superchips — each pairing one NVIDIA Grace CPU with two NVIDIA Blackwell B200 GPUs — delivering 72 Blackwell GPUs and 36 Grace CPUs within a single NVLink-interconnected rack domain. The GB200 NVL72 configuration provides up to 1.4 exaflops of AI compute at FP4 precision, with the 5th-generation NVLink fabric delivering 130 TB/s of aggregate GPU-to-GPU bandwidth across the entire 72-GPU pool, enabling the system to function as a unified, tightly coupled compute fabric rather than a collection of discrete nodes.

The XE9712 is designed around a direct liquid cooling (DLC) architecture to manage the extreme thermal output of the GB200 NVL72 rack. Each Blackwell B200 GPU features 192 GB of HBM3e memory, yielding a total HBM3e capacity of approximately 13.8 TB across the 72-GPU rack, with aggregate memory bandwidth exceeding 1.1 PB/s. The Grace CPU subsystem contributes additional LPDDR5X system memory per Superchip, connected to the GPU dies via NVIDIA's high-bandwidth NVLink-C2C chip-to-chip interconnect at 900 GB/s bidirectional bandwidth per Superchip, eliminating the traditional PCIe bottleneck between CPU and GPU.

As an enterprise-grade platform, the PowerEdge XE9712 is positioned for organizations deploying frontier-scale AI training, large language model inference, and high-performance simulation workloads at data center scale. Dell's integration of the NVIDIA MGX architecture within the XE9712 ensures compatibility with NVIDIA's full software ecosystem — including CUDA, NCCL, NVLink Switch System support, and NVIDIA AI Enterprise — while Dell's OpenManage systems management and IDRAC9 with Redfish API provide enterprise lifecycle management, telemetry, and operational visibility consistent with the broader PowerEdge portfolio.

Ideal for

• Training and fine-tuning frontier large language models (LLMs) and multimodal foundation models requiring tightly coupled, high-bandwidth multi-GPU compute at scale
• High-throughput generative AI inference serving for enterprise deployments demanding low-latency, high-concurrency token generation across extremely large model parameter counts
• Accelerated scientific simulation and digital twin workloads in energy, life sciences, and climate modeling that require sustained FP64 and FP8 mixed-precision compute
• Sovereign AI and national-scale AI infrastructure deployments where organizations require on-premises, liquid-cooled, rack-native GPU density under direct operational control
• Recommender system training and real-time ranking inference at hyperscale, leveraging the XE9712's extreme memory bandwidth and NVLink all-to-all GPU communication topology
• HPC and AI-converged workloads combining traditional supercomputing simulation with AI-driven analysis pipelines, utilizing the Grace CPU and Blackwell GPU within a unified memory-coherent architecture

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.