Supermicro
Rack-Scale AI Systems
Supermicro NVIDIA GB200 NVL72 SuperCluster
Liquid-cooled 72-GPU NVLink rack with Supermicro DLC and rack-level integration.
- GPUs: 72× NVIDIA Blackwell
- CPUs: 36× NVIDIA Grace
- NVLink: 72-GPU domain, 1.8 TB/s per GPU
Supermicro NVIDIA GB300 NVL72 SuperCluster
Model SRS-GB300-NVL72
End-to-end liquid-cooled rack with 72 Blackwell Ultra GPUs — an exascale supercomputer in a single rack.
- GPUs72× NVIDIA Blackwell Ultra, 288 GB HBM3e each
- CPUs36× NVIDIA Grace (72-core Arm Neoverse V2)
- NVLink1.8 TB/s across all 72 GPUs
- Cooling250 kW liquid-to-liquid CDU or 200 kW liquid-to-air sidecar
Overview
About the NVIDIA GB300 NVL72 SuperCluster
The Supermicro NVIDIA GB300 NVL72 SuperCluster is a complete, liquid-cooled 48U rack with 72 Blackwell Ultra GPUs and 36 Grace CPUs, connected by 1.8 TB/s NVLink into an exascale computer in a single rack. Each GPU carries 288 GB of HBM3e, for up to 21 TB of GPU memory.
Supermicro delivers it end to end, from consultation through rack integration, networking, onsite installation and its direct liquid cooling infrastructure, including in-rack or in-row CDUs and a liquid-to-air sidecar for sites without facility water.
72 Blackwell Ultra GPUs
Via 36 GB300 Grace Blackwell Ultra Superchips, 288 GB HBM3e each.
Up to 21 TB HBM3e
Plus up to 17 TB LPDDR5X system memory in the rack.
Direct liquid cooling
Up to 40% lower data center electricity cost, with CDU or sidecar options.
800 Gb/s compute fabric
ConnectX-8 SuperNICs on Quantum-X800 InfiniBand or Spectrum-X Ethernet.
144 E1.S bays
Up to 144 PCIe 5.0 E1.S drives for local storage.
Full deployment service
All parts, networking and onsite installation from one vendor.
Specifications
Technical Specifications
Rack
| Compute nodes | 18× 1U ARS-121GL-NB3 |
|---|---|
| Enclosure | 48U, 19-inch |
| Dimensions (W × D × H) | 600 × 1068 × 2236 mm |
Compute
| GPUs | 72× NVIDIA B300 (Blackwell Ultra), 288 GB HBM3e each |
|---|---|
| CPUs | 36× NVIDIA Grace |
| GPU memory | Up to 21 TB HBM3e |
| System memory | Up to 17 TB LPDDR5X |
| GPU interconnect | 9× NVLink Switches, 1.8 TB/s GPU-to-GPU |
Networking & storage
| Compute fabric | Quantum-X800 InfiniBand or Spectrum-X Ethernet, ConnectX-8 SuperNICs up to 800 Gb/s |
|---|---|
| In-band management | Up to 200 Gb/s with BlueField-3 DPUs |
| Out-of-band management | 1G / 10G Ethernet |
| Storage | Up to 144 E1.S PCIe 5.0 drive bays |
Power & cooling
| Power shelves | 8× 1U 33 kW (6× 5.5 kW PSUs each), 132 kW total |
|---|---|
| Cooling | 1× in-rack 250 kW CDU with redundant PSUs and dual hot-swap pumps |
| Alternative CDUs | In-row 1.8 MW; liquid-to-air sidecar 200 kW |
Source: official Supermicro product information.
Specifications are set by the manufacturer and may change. Plan Man confirms the final configuration, availability and lead time in your quote.
Use cases
What it's built for
- Foundation model training
- Large-scale reasoning inference
- AI factories and GPU clouds
- Enterprise AI at rack scale
Plan Man services
Supplied and supported in the UAE
- Requirement analysis, sizing and bill of materials
- Genuine supply with manufacturer warranty
- Rack integration, cabling and onsite deployment
- Power and liquid cooling planning for dense GPU racks
- After-sales support, RMA handling and upgrades from Dubai
NVIDIANew
Rack-Scale AI Systems
NVIDIA Vera Rubin NVL72
Next-generation rack-scale platform for agentic AI and reasoning, unifying 72 Rubin GPUs and 36 Vera CPUs in a single NVLink domain.
- GPUs: 72× NVIDIA Rubin
- CPUs: 36× NVIDIA Vera
- Scale-up fabric: Sixth-generation NVLink & NVLink Switch
NVIDIANew
Rack-Scale AI Systems
NVIDIA DGX Vera Rubin NVL72
Turnkey, ready-to-deploy enterprise AI infrastructure built on the Vera Rubin platform and managed with NVIDIA Mission Control.
- GPUs: 72× NVIDIA Rubin
- CPUs: 36× NVIDIA Vera
- Networking: ConnectX-9 SuperNICs, BlueField-4, Spectrum-X
Get a Quote
Tell us your workload and timeline for the Supermicro NVIDIA GB300 NVL72 SuperCluster. We reply within one business day with a configuration, pricing and lead time.