27-Node H200 PCIe GPU Cluster | 216× H200 (141 GB) & 54× EPYC CPUs
Highlights
 Racks: | As configured |
| GPUs: | 216× NVIDIA H200 PCIe (141 GB) |
| Nodes: | 27 |
| CPUs: | 54× AMD EPYC 9374F |
- Condition: Cancelled new servers — factory-sealed, unused
- Warranty: 3-Year International Warranty included
Sale Price: $287,000
Huge discounts on bulk order
An enterprise-grade GPU cluster built for large-scale AI training, generative AI, and HPC. Each node delivers eight NVIDIA H200 PCIe GPUs with NVLink (4-way) and high-bandwidth 400 Gb/s connectivity, balanced by dual EPYC 9374F processors, DDR5-4800 ECC memory, and NVMe storage. The 27-node configuration scales to 216 GPUs with consistent performance, serviceability, and remote management.
Designed for Scalable AI and Data Center Deployment
- Compute architecture: 8× H200 PCIe GPUs per node with NVLink bridges (4-way), dual AMD EPYC 9374F CPUs, DDR5-4800 ECC RDIMM.
- Storage tiers: Per node: 2× 1.92 TB NVMe (PM9A3) + 2× 3.84 TB NVMe (PM9A3); optional RAID via CRA4960.
- Networking: Per node: 3× ConnectX-7 VPI single-port OSFP (400 Gb/s) adapters (PCIe 5.0 x16).
- Operations: Remote/server management software (license-free), assembly & testing included. Power & cooling readiness: Standard data-center power cabling and CPU/GPU power harnesses included; rack rails provided.
Key capabilities include:
- Scales to 216 H200 GPUs for multi-node distributed training.
- 400 Gb/s host fabric per node for low-latency collective ops.
- NVLink (4-way) intra-node GPU connectivity.
- Balanced CPU/memory for data-pipeline throughput.
- Enterprise support (3-year standard warranty; RMA/remote tech support).
Compute Components (Total for 27 Nodes)
Component | Specification | Quantity |
GPU | NVIDIA H200 PCIe GPU (141 GB VRAM), 8× per node | 216 |
Server (Barebone) | G494-ZB4-ACP2 | 27 |
CPU | AMD EPYC 9374F, 2× per node | 54 |
Memory | 64 GB DDR5-4800 ECC RDIMM, 24× per node | 648 |
NVLink Bridges | NVIDIA NVLink Bridge Board (4-Way), P/N 900-23946-0000-000 / 25FB0-H20000-N2R | 54 |
NVMe SSD (Boot/Data 1) | PCIe NVMe SSD 1.92 TB PM9A3, 2× per node | 54 |
NVMe SSD (Data 2) | PCIe NVMe SSD 3.84 TB PM9A3, 2× per node | 54 |
SSD RAID Card | CRA4960 | 27 |
Network Infrastructure
Component | Specification | Quantity |
Host Adapters | NVIDIA ConnectX-7 VPI 400 Gb/s (InfiniBand/Ethernet), single-port OSFP, HHHL, PCIe 5.0 x16, Secure Boot enabled, crypto disabled, MCX75310AAS-NEAT | 81 |
Fabric Cabling/Switching | As configured (not specified) | — |
Management Network | As configured (not specified) | — |
Services & Software
Item | Description | Quantity |
Accessories | 4× server power cables; 1× slide-rail kit; 2× CPU heatsinks; 8× GPU power cables; 2× SlimSAS cables for CRA4960 | 27 |
Service 1 | Assembly and testing | 27 |
Service 2 | 3-Year standard warranty (parts & labor, remote technical support, standard RMA return-to-base) | 27 |
Software | Server Management (GSM) Software — license-free | 27 |
About
An enterprise-grade GPU cluster built for large-scale AI training, generative AI, and HPC. Each node delivers eight NVIDIA H200 PCIe GPUs with NVLink (4-way) and high-bandwidth 400 Gb/s connectivity, balanced by dual EPYC 9374F processors, DDR5-4800 ECC memory, and NVMe storage. The 27-node configuration scales to 216 GPUs with consistent performance, serviceability, and remote management.
Designed for Scalable AI and Data Center Deployment
- Compute architecture: 8× H200 PCIe GPUs per node with NVLink bridges (4-way), dual AMD EPYC 9374F CPUs, DDR5-4800 ECC RDIMM.
- Storage tiers: Per node: 2× 1.92 TB NVMe (PM9A3) + 2× 3.84 TB NVMe (PM9A3); optional RAID via CRA4960.
- Networking: Per node: 3× ConnectX-7 VPI single-port OSFP (400 Gb/s) adapters (PCIe 5.0 x16).
- Operations: Remote/server management software (license-free), assembly & testing included. Power & cooling readiness: Standard data-center power cabling and CPU/GPU power harnesses included; rack rails provided.
Key capabilities include:
- Scales to 216 H200 GPUs for multi-node distributed training.
- 400 Gb/s host fabric per node for low-latency collective ops.
- NVLink (4-way) intra-node GPU connectivity.
- Balanced CPU/memory for data-pipeline throughput.
- Enterprise support (3-year standard warranty; RMA/remote tech support).
Specifications
Compute Components (Total for 27 Nodes)
Component | Specification | Quantity |
GPU | NVIDIA H200 PCIe GPU (141 GB VRAM), 8× per node | 216 |
Server (Barebone) | G494-ZB4-ACP2 | 27 |
CPU | AMD EPYC 9374F, 2× per node | 54 |
Memory | 64 GB DDR5-4800 ECC RDIMM, 24× per node | 648 |
NVLink Bridges | NVIDIA NVLink Bridge Board (4-Way), P/N 900-23946-0000-000 / 25FB0-H20000-N2R | 54 |
NVMe SSD (Boot/Data 1) | PCIe NVMe SSD 1.92 TB PM9A3, 2× per node | 54 |
NVMe SSD (Data 2) | PCIe NVMe SSD 3.84 TB PM9A3, 2× per node | 54 |
SSD RAID Card | CRA4960 | 27 |
Network Infrastructure
Component | Specification | Quantity |
Host Adapters | NVIDIA ConnectX-7 VPI 400 Gb/s (InfiniBand/Ethernet), single-port OSFP, HHHL, PCIe 5.0 x16, Secure Boot enabled, crypto disabled, MCX75310AAS-NEAT | 81 |
Fabric Cabling | As configured (not specified) | — |
Management Network | As configured (not specified) | — |
Services & Software
Item | Description | Quantity |
Accessories | 4× server power cables; 1× slide-rail kit; 2× CPU heatsinks; 8× GPU power cables; 2× SlimSAS cables for CRA4960 | 27 |
Service 1 | Assembly and testing | 27 |
Service 2 | 3-Year standard warranty (parts & labor, remote technical support, standard RMA return-to-base) | 27 |
Software | Server Management (GSM) Software — license-free | 27 |
