Supermicro SRS-GB300-NVL72
The Supermicro SRS-GB300-NVL72 is a liquid-cooled rack with 72 NVIDIA B300 GPUs and 36 Grace CPUs, delivering 1.1 ExaFLOPS and 20.7TB of HBM3e for trillion-parameter LLM training and inference.
The Supermicro SRS-GB300-NVL72 is a liquid-cooled, rack-scale AI system based on the NVIDIA Blackwell Ultra platform. It integrates 72 NVIDIA B300 GPUs and 36 NVIDIA Grace CPUs into a single, massive NVLink domain, specifically designed for large-scale AI training and high-throughput reasoning.
Key Technical Specifications
FeatureSpecificationForm Factor48U Liquid-Cooled RackGPU Configuration72x NVIDIA B300 (Blackwell Ultra) Tensor Core GPUsCPU Configuration36x NVIDIA Grace™ CPUs (Arm-based Neoverse V2)Total CPU Cores2,592 Arm Neoverse V2 cores (72 cores per CPU)GPU Memory20.7 TB total HBM3e (~288GB per GPU)System Memory17.2 TB total LPDDR5X (~480GB per CPU)Performance (FP4)~1.1 ExaFLOPS (Dense)NVLink Bandwidth1.8 TB/s per GPU (130 TB/s total rack bandwidth)NetworkingNVIDIA ConnectX-8 SuperNIC (up to 800 Gb/s)DPU SupportNVIDIA BlueField-3 DPUsStorageUp to 144x E1.S PCIe 5.0 NVMe drive bays
Certifications
Red Hat certified hardware is proven to incorporate Red Hat's best practices and provides customers tested interoperability, known life cycle management, and trusted support.
| Compare | Product | Level | |
|---|---|---|---|
Red Hat Enterprise Linux 9.6 - 9.x Architecture: aarch64 View Kbase | |||
Red Hat OpenShift Container Platform 4.19 - 4.x Base OS: Red Hat Enterprise Linux 9.6 Architecture: aarch64 View Kbase | |||
Red Hat AI Inference Server 3.3 - 3.x Base OS: Red Hat Enterprise Linux 9.6 Architecture: aarch64 View Kbase |
