8 Node H200 GPU Compute Cluster | 64x NVIDIA H200

8-Node H200 GPU Compute Cluster | 64x NVIDIA H200 SXM5 with InfiniBand NDR

Highlights

Rack Height:	8U per node
GPU Configuration:	64x NVIDIA H200 SXM5
Number of Nodes:	8
Processor Total:	16x AMD EPYC 9754

Starting Price: Contact for Price

The 8-Node H200 GPU Compute Cluster from Hyblox is engineered for organizations that need extreme scale for AI and HPC. Featuring a total of 64x NVIDIA H200 SXM5 GPUs across eight high-density 8U servers, this cluster delivers unmatched performance for large language model training, generative AI, and advanced simulation workloads. Each GPU comes with 141GB of HBM3e memory, enabling efficient handling of massive datasets and reducing training bottlenecks

Built for Enterprise AI and HPC Infrastructure

Each node is powered by dual AMD EPYC 9754 processors with 128 cores, supported by high-capacity DDR5-6000 ECC memory and ultra-fast PCIe Gen5 NVMe storage. The cluster is interconnected with NVIDIA InfiniBand NDR networking, combining QM9700 leaf switches, a QM9790 spine switch, and ConnectX-7 adapters for ultra-low latency communication across all nodes.

Key capabilities include:

Massive GPU density with 64x NVIDIA H200 SXM5 GPUs for training trillion-parameter models
Optimized for AI supercomputing, LLM development, and scientific HPC workloads
Scalable InfiniBand NDR fat-tree architecture for balanced, high-throughput networking
Enterprise-ready deployment with redundant power, advanced cooling, and rack integration

Compute Components (Total for 3 Nodes)

Component	Specification	Quantity
GPU	NVIDIA H200 SXM5 141GB HBM3e	64
Server Chassis	Supermicro AS-8125GS-TNHR (8U)	8
CPU	AMD EPYC 9754 (128-core, 2.25GHz)	16
Memory	128GB DDR5-6000 RDIMM ECC	192
Boot Storage	Samsung PM1743 2TB PCIe Gen5 NVMe U.2	16
Model Storage	Western Digital SN861 8TB PCIe Gen5 NVMe U.2	48

Network Infrastructure

Component	Specification	Quantity
Leaf Switch	NVIDIA QM9700 64-port NDR 400Gb/s	2
Spine Switch	NVIDIA QM9790 64-port NDR 400Gb/s	1
Network Adapter	NVIDIA ConnectX-7 NDR 400GbE Dual-port OSFP	16
Node-Leaf Cables	MCP4Y10-NO01 1m OSFP Passive DAC	16
Leaf-Spine Cables	MCP4Y10-NO03 3m OSFP Active Copper	4
Management Switch	48-port 10GbE stackable managed switch	2
Management Cables	Cat6a cables, various lengths	20

Power Infrastructure

Component	Specification	Quantity
Rack PDU	Eaton HDX G4 60kW 415V 3-phase	2
Power Supply	3000W Titanium PSU	48
Power Cables	C19 to C20, 2m	48
UPS System	120kVA modular UPS system	1
Power Distribution	200A 3-phase distribution panel	1

Environmental & Rack

Component	Specification	Quantity
Server Rack	48U standard depth	2
Rack Configuration	Rack 1: 6 nodes (48U), Rack 2: 2 nodes (16U) + switches	–
Cooling Solution	Vertiv Liebert CRV 45kW in-row units	2
CDU (Optional)	Coolant Distribution Unit for liquid cooling	1
Cable Management	Full vertical and horizontal management	2 sets
Blanking Panels	Complete airflow optimization kit	2 sets

About