Menu

Supermicro SYS-422GL-NR 4U GPU SuperServer

$84,000.00

The Supermicro SYS-422GL-NR is an air-cooled 4U MGX platform with dual Xeon 6900 P-core sockets and support for up to eight double-width 600W PCIe GPUs. GPU, NIC, memory and drive populations depend on the supplied configuration. Confirm the exact NR model and installed BOM.

BrandSupermicroSKUARS-SRV-SUPERM-422GLNR4USERVE-0280CategoryAI Servers

Explore Related

Related products, categories and research.

SYS-422GL-NR: a dual-root MGX platform for PCIe GPUs

The Supermicro SYS-422GL-NR uses two Intel Xeon 6900 P-core processor sockets on the X14DBG-MAP motherboard. Its 4U air-cooled design supports up to eight double-width PCIe GPU adapters rated up to 600W each. The switch-based dual-root architecture connects accelerator and networking resources to the host processors.

Supermicro SYS-422GL-NR 4U MGX GPU server reference
Manufacturer chassis illustration; accelerator population and other installed equipment require confirmation.

Choose qualified GPUs rather than assuming a fixed accelerator assembly

The model page lists NVIDIA H200 NVL 141GB and NVIDIA RTX PRO 6000 Blackwell Server Edition among supported PCIe options. These are configuration choices, not evidence that either accelerator is included in this listing. Request the installed card models, memory capacity and quantity.

The documented GPU link option is an NVLink bridge where supported. This differs from an HGX SXM assembly with an NVSwitch fabric. Confirm the relevant bridge support for the selected cards and application; the eight-slot capacity does not establish an eight-device all-to-all fabric.

Host processing and memory specifications

Host platform Dual Socket BR LGA-7529, Intel Xeon 6900 processors with P-cores
Processor cooling envelope Up to 500W TDP per supported CPU, air cooled
DIMM positions 24 slots, one DIMM per channel
RDIMM memory Up to 6TB ECC DDR5 at 6400MT/s
MRDIMM memory Up to 6TB ECC DDR5 at 8800MT/s
CPU-to-GPU connection PCIe 5.0 x16 switch-based dual-root topology
Management LAN One RJ45 1GBASE-T connection with ASPEED AST2600

Memory speed and capacity depend on qualified processors and modules. The listed ceilings do not describe the supplied memory population. Check the CPU selection and DIMM part numbers together before planning an upgrade or estimating host-side dataset capacity.

PCIe allocation for accelerators and network cards

Eight PCIe 5.0 x16 positions serve double-width GPUs through the switch fabric. Four additional x16 positions serve NICs through that fabric, and another x16 position for a NIC or DPU connects through a CPU. The slots have different placement and connectivity roles.

Plan network and accelerator traffic against the actual topology. A physical x16 slot does not establish a dedicated end-to-end bandwidth allocation for every concurrent transfer. The software workload, switch connections and adapter selections influence data movement.

The management RJ45 port does not identify the application-network configuration. Request the NIC or DPU models, port speeds, optics and cables separately. Optional network adapter positions describe expansion support rather than a supplied cluster fabric.

E1.S storage and internal boot devices

Up to eight front hot-swap E1.S NVMe drives connect through PCIe switches. Two internal NVMe M.2 devices use the 2280 form factor, with the datasheet identifying PCIe 4.0 x4 interfaces. E1.S and M.2 refer to different physical formats and should not be treated as interchangeable drive choices.

Confirm installed SSD quantities, endurance requirements and usable capacity for the intended workload. Slot and bay availability do not establish a RAID layout or software recovery policy. Account for the storage path alongside GPU and NIC traffic when assessing the complete system.

Rack dimensions, airflow and power

The enclosure measures 438.4mm wide, 176mm high and 800mm deep. Cabinet fit also needs cable clearance, service access and the correct rail arrangement. Review the actual configured assembly when planning handling and rack loading.

The manual specifies five 80mm front fans and five 80mm mid-chassis fans. Four 3200W Titanium supplies provide redundant power. Power-supply capacity and individual component TDP are design ratings, not measured system energy use.

Check the ordered processors, accelerator population and facility feeds as one power and cooling plan. An air-cooled chassis still needs suitable inlet conditions and adequate room heat removal. Do not assume a lower-power card configuration and the maximum eight-card configuration have identical installation needs.

Model and configuration questions

Is NR equivalent to NR2 or FNR2?

Do not substitute those designations. This listing concerns SYS-422GL-NR; a different suffix requires its own model documentation and compatible component list.

What should be checked before accepting a configuration?

Verify the full model label, CPU and DIMM population, GPU card selection, adapter layout, SSDs, software and included accessories. Consult the manufacturer system overview and compare other AI infrastructure platforms against the same workload and facility requirements.

Recently viewed

Add AI Robot Supplier as a preferred source on Google.

WhatsApp Chat with our sales team