10 Rack H200 GPU Compute Cluster | 60 Nodes with 480x H200 SXM5 and 120 EPYC CPUs
Highlights
 Rack Count: | 10 standard-depth racks |
| GPU Configuration: | 480x NVIDIA H200 SXM5 GPUs |
| Number of Nodes: | 60 high-density 8U servers |
| Processor Total: | 120x AMD EPYC 9754 CPUs |
Starting Price: Contact for Price
The 10 Rack H200 GPU Compute Cluster from Hyblox is an enterprise-scale AI supercomputing solution built to deliver the performance required for the most demanding workloads. With 480x NVIDIA H200 SXM5 GPUs distributed across 60 nodes, this cluster provides unmatched density and computational throughput for large language model training, generative AI, and advanced HPC research. Each H200 GPU is equipped with 141GB of HBM3e memory, enabling the system to handle massive datasets and accelerate time-to-insight
Designed for Scalable AI and Data Center Deployment
At the core of each node are dual AMD EPYC 9754 processors with 128 cores, combined with DDR5-6000 ECC memory for a total of more than 1,400 DIMMs across the cluster. High-speed PCIe Gen5 NVMe storage supports both boot and large model datasets, ensuring rapid I/O and balanced performance across workloads.
The system uses a 3-tier InfiniBand NDR fat-tree topology with NVIDIA QM9700 switches, QM9790 spine layers, and ConnectX-7 adapters, delivering ultra-low latency communication across all 60 nodes. With enterprise-grade PDUs, a 750kVA UPS system, and redundant cooling infrastructure, the cluster is ready for continuous large-scale deployment in modern data centers
Key capabilities include:
- Extreme scale with 480x NVIDIA H200 SXM5 GPUs for AI supercomputing and trillion-parameter models
- Optimized for LLM training, generative AI inference, and HPC workloads
- 3-tier InfiniBand NDR networking ensures high-bandwidth, low-latency interconnects across all racks
- Enterprise-ready with redundant power, UPS systems, advanced cooling, and rack-optimized integration
Compute Components (Total for 60 Nodes)
Component | Specification | Quantity |
GPU | NVIDIA H200 SXM5 141GB HBM3e | 480 |
Server Chassis | Supermicro AS-8125GS-TNHR (8U) | 60 |
CPU | AMD EPYC 9754 (128-core, 2.25GHz) | 120 |
Memory | 128GB DDR5-6000 RDIMM ECC | 1,440 |
Boot Storage | Samsung PM1743 2TB PCIe Gen5 NVMe U.2 | 120 |
Model Storage | Western Digital SN861 8TB PCIe Gen5 NVMe U.2 | 360 |
Network Infrastructure
Component | Specification | Quantity |
Leaf Switch | NVIDIA QM9700 64-port NDR 400Gb/s | 15 |
Spine Switch | NVIDIA QM9700 64-port NDR 400Gb/s | 6 |
Super-Spine Switch | NVIDIA QM9700 64-port NDR 400Gb/s | 3 |
Network Adapter | NVIDIA ConnectX-7 NDR 400GbE Dual-port OSFP | 120 |
Node-Leaf Cables | MCP4Y10-NO01 1m OSFP Passive DAC | 120 |
Leaf-Spine Cables | MCP4Y10-NO03 3m OSFP Active Copper | 120 |
Spine-Super-Spine | MCP4Y10-AO05 5m OSFP Active Optical | 24 |
Management Network | Distributed 48-port 10GbE switches | 10 |
Management Cables | Cat6a cables, structured cabling system | 150 |
Power Infrastructure
Component | Specification | Quantity |
Rack PDU | Eaton HDX G4 60kW 415V 3-phase | 20 |
Power Supply Units | 3000W Titanium PSU | 360 |
Power Cables | C19 to C20, 2m | 360 |
UPS System | 750kVA modular UPS architecture | 1 |
Switchgear | 1200A main distribution switchgear | 1 |
Sub-Distribution | 400A panels per 2 racks | 5 |
Generator | 1.2MW diesel generator with ATS | 1 |
About
Designed for Scalable AI and Data Center Deployment
Key capabilities include:
- Extreme scale with 480x NVIDIA H200 SXM5 GPUs for AI supercomputing and trillion-parameter models
- Optimized for LLM training, generative AI inference, and HPC workloads
- 3-tier InfiniBand NDR networking ensures high-bandwidth, low-latency interconnects across all racks
- Enterprise-ready with redundant power, UPS systems, advanced cooling, and rack-optimized integration
Specifications
Compute Components (Total for 60 Nodes)
Component | Specification | Quantity |
GPU | NVIDIA H200 SXM5 141GB HBM3e | 480 |
Server Chassis | Supermicro AS-8125GS-TNHR (8U) | 60 |
CPU | AMD EPYC 9754 (128-core, 2.25GHz) | 120 |
Memory | 128GB DDR5-6000 RDIMM ECC | 1,440 |
Boot Storage | Samsung PM1743 2TB PCIe Gen5 NVMe U.2 | 120 |
Model Storage | Western Digital SN861 8TB PCIe Gen5 NVMe U.2 | 360 |
Network Infrastructure
Component | Specification | Quantity |
Leaf Switch | NVIDIA QM9700 64-port NDR 400Gb/s | 15 |
Spine Switch | NVIDIA QM9700 64-port NDR 400Gb/s | 6 |
Super-Spine Switch | NVIDIA QM9700 64-port NDR 400Gb/s | 3 |
Network Adapter | NVIDIA ConnectX-7 NDR 400GbE Dual-port OSFP | 120 |
Node-Leaf Cables | MCP4Y10-NO01 1m OSFP Passive DAC | 120 |
Leaf-Spine Cables | MCP4Y10-NO03 3m OSFP Active Copper | 120 |
Spine-Super-Spine | MCP4Y10-AO05 5m OSFP Active Optical | 24 |
Management Network | Distributed 48-port 10GbE switches | 10 |
Management Cables | Cat6a cables, structured cabling system | 150 |
Power Infrastructure
Component | Specification | Quantity |
Rack PDU | Eaton HDX G4 60kW 415V 3-phase | 20 |
Power Supply Units | 3000W Titanium PSU | 360 |
Power Cables | C19 to C20, 2m | 360 |
UPS System | 750kVA modular UPS architecture | 1 |
Switchgear | 1200A main distribution switchgear | 1 |
Sub-Distribution | 400A panels per 2 racks | 5 |
Generator | 1.2MW diesel generator with ATS | 1 |
