• Unmatched density, flexibility, and customizability that can be optimized for specific compute, GPU acceleration, networking, and storage requirements. 
• A+ Multi-GPU systems with advanced thermal designs support the highest TDP (280W) processors and latest GPUs in a dense form factor in 2U and 4U.
• The unique AIOM enhances fast and flexible networking capabilities for large-scale/disaggregated infrastructure that requires massive dataflow between systems. 

Supermicro offers the broadest portfolio of GPU Systems, powered by dual 3rd Gen AMD EPYC™ Processors. Leveraging our advanced thermal design, including liquid cooling and custom heatsinks, Supermicro 2U and 4U GPU systems feature NVIDIA’s latest A100 or AMD Instinct™ MI200 GPUs in hyper-dense multi-GPU, multi-node systems. Supermicro’s flexible Advanced I/O Module (AIOM) form factor further enhances multi-GPU communication for data-hungry AI applications.

• Supports AIOM
• 32 DIMM slots supporting DDR-3200 up to 8TB, up to 10 (8 - PCI-E switch and 2 - CPU) U.2 NVMe drives + 2 M.2 drives
• 2200W cost effective Platinum level redundant power supplies or 3000W true redundant Titanium level power supplies

• High GPU peer-to-peer communication with 3rd Gen NVLink and NVSwitch
• Dual 3rd  or 2nd Gen AMD EPYC top of the line processors (Max. 280W)
• 1:1 GPUDirect RDMA for large-scale deep learning model training and inference

The latest Ampere generation NVIDIA HGX A100 8-GPU with Supermicro's unique AIOM support enhancing the 8GPU communication and data flow between systems featuring NVIDIA NVLink and NVSwitch, 1:1 GPUDirect RDMA, GPUDirect Storage, and NVMe-oF on InfiniBand.

•  Dual AMD EPYC™ 7003/7002 Series Processors(7003 Series Processor drop-in support requires BIOS version 2.0 or newer) 6.8TB Registered ECC DDR4 3200MHz SDRAM in 32 DIMMs
• 4 PCI-E Gen 4 x16 (LP), 1 PCI-E Gen 4 x8 (LP)
• 4 Hot-swap 2.5" drive bays (SAS/SATA/NVMe Hybrid)
• 2x 2200W Platinum Level power supplies with Smart Power Redundancy

• High Density 2U System with NVIDIA® HGX™ A100 4-GPU; Highest GPU communication using NVIDIA® NVLINK™, 4 NICs for GPUDirect RDMA (1:1 GPU Ratio)
• Supports HGX A100 4-GPU 40GB (HBM2) or 80GB (HBM2e)
• Direct connect PCI-E Gen4 Platform with NVIDIA® NVLink™ v3.0 up to 600GB/s interconnect
• On board BMC supports integrated IPMI 2.0 + KVM with dedicated 10G LAN

Features NVIDIA HGX A100 4-GPU with Supermicro's advanced thermal heatsink design for best performance and flexibility in the compact 2U form factor, enabling high GPU peer-to-peer communication without I/O bottleneck for data hungry workloads.

• Up to 24 Hot-swap 2.5" drive bays.
• 2 GbE LAN ports
• 8 Hot-swap 11.5K RPM cooling fans
• 2000W (2+2) Redundant Power Supplies

  Titanium Level (96%+)

• Supports up to 8 double-width GPUs Direct connect for maximum performance
• Double AMD EPYC™ 7003/7002 Series Processors
• 8TB Registered ECC DDR4 3200MHz SDRAM in 32 DIMMs
• 9 PCI-E 4.0 x16 (Option: 10 PCI-E 4.0 x16 slots without NVMe devices)

Featuring dual-root topology optimized for HPC applications,  the maximum 160 PCI-E 4.0 lanes formed by dual AMD EPYC™ CPU socket configuration drive 8 directly attached PCI-E 4.0 GPUs, 200G networking and U.2 NVMe Storage at full-speed.

• Integrated IPMI 2.0 + KVM with  dedicated LAN
• 2 front Hot-swap 2.5" U.2 NVMe Gen4 drive bays
• AST2600 BMC
• 2600W Redundant (1+1) Power Supplies

  Titanium Level (96%)

• Supports up to 6 single-width GPUs Direct connect for maximum performance
• Single AMD EPYC™ 7003/7002 Series Processors (7003 Series Processor drop-in support requires BIOS version 2.0 or newer)
• 2TB Registered ECC DDR4 3200MHz SDRAM in 8 DIMMs
• Up to 6 PCI-E Gen 4 x16 (4 internal, 2 external) slots, 1 PCI-E 4.0 x8 AIOM slot, 2x NVMe M.2, Form Factor: 2280, 2210, M.2 Key: M-key

Setting a new standard for dense, energy efficient, and resource saving multi-node, multi-GPU system, supporting up to 3 double-width or 6 single-width PCI-E 4.0 GPUs per node,  ideal for multi-instance high-end cloud gaming and many other compute intensive data center applications. Available with AMD EPYC™ 7003/7002 Series or AMD Ryzen™ Threadripper™ PRO processors

• Dual AMD EPYC™ 7003 series processors
• Supports the new AMD Instinct™ MI250 OAM Accelerator
• 4U or 5U configuration32 DIMM slots per node supporting DDR5- 3200MHz
• Flexible Storage configuration with 10 hot-swap 2.5” U.2 NVMe drives
• 4U with optional 1U extension for a 5U system providing PCIe slots expansion with Supermicro AIOM support
• Supports next-generation GPUs in a variety of form factors
• Universal GPU server OCP standards-based design
• Modular design for flexibility/future-proofing
• Optimized thermal capability for 500W/700W GPUs

Supermicro A+ Universal GPU systems are open, modular, standards-based servers which provide superior performance and serviceability with dual AMD EPYC™ 7003 series processors, supporting AMD Instinct™ MI250 OAM Accelerator and various GPU and accelerator form factors, and featuring a hot-swappable, tool-less design. The system's “future proofed” design allows to standardize on one GPU platform with multiple configurations for all data center needs with optimized thermal management.