8× RTX PRO 6000.
The complete AI server.
Eight NVIDIA RTX PRO 6000 Blackwell Server Edition accelerators, dual Xeon host compute, ECC memory, enterprise NVMe, and high-speed networking in a configurable 5U platform.
Supermicro SYS-521GE-TNRT
SYS-521GE-TNRT reference platform. The OEM photo shows a different GPU population; NexTec’s proposed configuration specifies 8× RTX PRO 6000 Server Edition.
Manufacturer reference imageExplore the system in 3D.
Rotate the chassis, separate its layers, and select a component to understand its role.
Loading interactive system…
Reference: Supermicro SYS-521GE-TNRT. Chassis envelope: 438 × 222.5 × 786.1 mm. Component placement, clearances, connectors and cable routing are simplified; this is a detailed reference model, not OEM CAD or an assembly guide. The proposed GPU count is eight. Final BOM, DIMM population and power redundancy require platform validation.
The full system, defined.
NexTec’s proposed 8-GPU build uses the Supermicro SYS-521GE-TNRT as its reference platform. The table identifies the OEM hardware envelope; the configuration builder below records your requested bill of materials. Exact part numbers, firmware, power limits, and acceptance testing are confirmed in the written proposal.
| Reference system | Supermicro SYS-521GE-TNRT · CSE-528GTS-R5K40P chassis · X13DEG-OA motherboard. |
|---|---|
| Specified GPUs | 8× NVIDIA RTX PRO 6000 Blackwell Server Edition, passive air-cooled; 96 GB GDDR7 ECC each. |
| GPU compute / memory | 24,064 CUDA cores; 188 fourth-generation RT cores; 1,597 GB/s memory bandwidth per GPU. 768 GB aggregate memory across eight devices. |
| GPU compatibility | Current OEM matrix: up to 8 double-width GPUs. GPU type, power profile, risers, and harnesses must match the qualified configuration. |
| Host processors | Two LGA-4677 sockets. 4th / 5th Gen Intel Xeon Scalable; up to 64 cores and 350 W per CPU with air cooling. |
| CPU starting options | 2× Xeon Gold 6530: 64 total cores, 270 W TDP each; or 2× Xeon Platinum 8592+: 128 total cores, 350 W TDP each. Final BOM subject to OEM qualification. |
| ECC system memory | 32 DDR5 DIMM slots. OEM maximum: 4 TB at 1 DIMM/channel, up to 5600 MT/s; 8 TB at 2 DIMMs/channel, up to 4400 MT/s. CPU and DIMM selection set actual speed. |
| Balanced memory population | Proposed 512 GB / 1 TB / 2 TB use 16× 32 / 64 / 128 GB RDIMMs, one per channel across both sockets. Gold 6530 limits 1DPC memory to 4800 MT/s. |
| Boot devices | 2× M.2 NVMe slots. Proposal options include paired enterprise boot drives; mirroring method and OS support are itemized. |
| Front storage | Default: 8× 2.5-inch NVMe and 8× 2.5-inch SATA hot-swap bays. An optional additional 8-bay NVMe configuration requires the appropriate controller and cabling. |
| Expansion / topology | 13 PCIe 5.0 x16 FHFL slots; switch-based dual-root architecture. Slot link width is distinct from aggregate switch-uplink bandwidth. |
| Network / management | Onboard 2× 10GbE RJ45 via Intel X710-AT2; dedicated IPMI management. AIOM/OCP 3.0 and PCIe NIC options for the application fabric. |
| High-speed fabric | Quote 100/200/400GbE or InfiniBand with qualified NICs, optics, cables, and switch ports. Define GPU/NIC locality, storage traffic, and GPUDirect requirements. |
| Cooling | Up to 10 heavy-duty fans with speed control and an air shroud. Passive GPUs rely on chassis airflow. Rack inlet temperature and GPU power settings must follow OEM limits. |
| Power hardware | 4× 2700 W Titanium PSUs; 200–240 VAC input. A 2+2 arrangement provides 5.4 kW nominal surviving output after loss of two modules; the complete DC load must fit that envelope. |
| GPU power budget | 400–600 W configurable per GPU: 3.2–4.8 kW for eight GPUs alone. The suggested 450 W planning profile leaves more headroom; exact supported settings and failover require validation. |
| Rack envelope | 5U · 438 mm wide × 222.5 mm high × 786.1 mm deep. Confirm installed weight, rails, cable clearance, PDU connectors, and front-to-rear airflow. |
| Integration deliverables | Exact BOM and serial inventory; firmware/driver versions; GPU, ECC, storage and network results; sustained thermal/power logs; workload acceptance results; and the agreed support path. |
This is a proposed configuration based on published OEM specifications, not an already benchmarked or in-stock NexTec SKU.
Aggregate GPU memory is not one unified memory pool. RTX PRO 6000 provides no HGX NVLink/NVSwitch fabric; validate PCIe peer access for the selected firmware and software stack.
OEM gallery photos show reference populations and older callouts. The written BOM and current support matrix determine the eight-GPU configuration.
Build a useful starting point.
Choose the host and I/O around eight Server Edition GPUs. These are requirements for engineering review, with final compatibility and pricing confirmed in the proposal.
Choices describe requested scope, not guaranteed compatibility or included services. Exact SKUs, software versions, licenses, redundancy and support terms are confirmed in the quotation. The 3D model illustrates a reference population and does not change with these selections.