International Shipping. | Advance Replacement. | Online Support.
3-node Hyperconverged Proxmox VE (HCI) Cluster – MG1A GPU Datacenter
GPU-enabled Hyperconverged Proxmox Cluster, consisting of 3 2U rackmount servers, tested and certified for Proxmox VE. Designed for businesses requiring high reliability and high resource uptime, as well as the ability to run Artificial Intelligence (AI) workloads (training and inference), Machine Learning, and GPU-enabled VDI (Virtual Desktop Infrastructure) workloads distributed across multiple nodes.
Each node is equipped with 2 AMD EPYC™ 5th Generation (9005 Series) CPUs in a dual-socket configuration, supporting up to 384 physical cores per node, up to 6 TB of DDR5 ECC RAM per node, up to 3 double-width PCIe GPUs per node (NVIDIA or AMD Instinct), and 8 Datacenter-class SSD/NVMe drives for hyperconverged Ceph storage. Dedicated connectivity is provided through an AIOM module offering 25 to 200 Gbit/s for hyperconverged networking. Titanium-level redundant power supplies are included for both the nodes and the switches.
The solution is supplied with rack-mount rails for both the servers and switches. Depending on requirements, it can be supplied with a single switch, dual redundant switches, or a Full Mesh configuration. The solution is designed to eliminate any SPOF (Single Point of Failure).
Below, you can choose how to configure the 3 nodes that make up the cluster, including the GPUs.
All 3 servers will have the same hardware configuration. If you are unsure which options to choose, leave the default values selected. Before sending the quotation, our account managers will validate the components you have selected.
Availability:
In stock
SKU
MG1ASUC2S8P2_26
Product Name: Proxmox Datacenter GPU Cluster MG1A
Mainboard: H14DSH
Chassis: 2U rackmount chassis with rack-mount installation rails
CPU: Dual-socket configuration, AMD EPYC™ 9004/9005 Series, up to 384 cores / 768 threads total per node
GPU: Up to 3 double-width GPUs per node. Supported GPUs: NVIDIA PCIe H100 NVL, A16, L4, L40, L40S, RTX PRO™ 6000 Blackwell Max-Q, RTX 2000/4500/5000/6000 Ada Generation; AMD PCIe Instinct™ MI210. GPU-to-GPU interconnection: PCIe.
Memory: 24 DIMM slots; up to 6 TB of 6400 MT/s ECC DDR5 RDIMM (9005 Series) / 4800 MT/s (9004 Series), 1 DPC
Data Storage: Drive bays: up to 24 hot-swap 2.5" NVMe/SAS/SATA drives per node (8 in the tested GPU configuration). NVMe/SATA support may require additional controllers/cables.
OS Storage: 2 × M.2 PCIe 3.0 x4 NVMe (M-Key 2280/22110)
Expansion: Up to 4 × CXL 2.0 x16 devices
Onboard Networking: Connectivity via AIOM module. IPMI: Intelligent Platform Management Interface v2.0 support with virtual media over LAN and KVM-over-LAN
Input/Output: LAN: 1 × dedicated RJ45 1GbE BMC port; USB: 2 × USB 3.0 (rear); Video: 1 × VGA; TPM: 1 × onboard / Port 80
Cooling: Up to 6 × 60 × 60 × 56 mm counter-rotating fans; 2 air shrouds
Power Supply: 2 × 2000W redundant power supplies (1+1), Titanium Level (96%) — per node
Management: SuperCloud Composer® (SCC), Supermicro Server Manager (SSM), Super Diagnostics Offline (SDO), KVM with dedicated LAN, IPMI 2.0, Supermicro Thin-Agent Service (TAS), IPMIView
Operating Temperature: 10–35°C (50–95°F)
Non-Operating Temperature: -40–70°C
Relative Humidity: Operating: 8–90% (non-condensing); Non-operating: 5–95%
Dimensions: Enclosure: 437 × 88.9 × 806.2 mm (W × H × D) — 2
