Specifications
MPN
TCS-DELL-XE9712
Model
PowerEdge XE9712
Platform
NVIDIA GB200 NVL72 rack-scale system
Rack
IR9048 — 48 RU, 2294 x 750 x 1200 mm, ~1590 kg wet
Sled
1 RU compute sled, 43.6 mm high, ~30 kg
Accelerators
72x NVIDIA Blackwell GPU (4 per sled)
Host CPU
36x NVIDIA Grace (2 per sled, Arm Neoverse V2)
Scale-up
NVLink 5 switch fabric — 1,800 GB/s per accelerator, 72-way domain
Scale-out
800G per sled via ConnectX / BlueField-3 SuperNIC
Sled I/O
2x FHHL PCIe Gen5 front risers, 2x Gen5/Gen6 mezzanine
Networking Card
1x BlueField-3 SuperNIC FHHL per sled
Memory
DDR5 up to 5200 MT/s
Cooling
Liquid cooled (rack-level CDU)
Management
OpenManage Enterprise / iDRAC, Redfish
Rack-scale AI system comparison
| System | Rack | Accelerators | Scale-up domain | Cooling |
|---|---|---|---|---|
| Dell PowerEdge XE9712 | 48U IR9048 | 72x NVIDIA B200 (GB200 NVL72) | 72-way NVLink, 1.8 TB/s | Liquid |
| Dell PowerEdge XE9680 | 6U | 8x HGX H200 SXM5 | 8-way NVSwitch | Air / DLC |
| NVIDIA DGX B200 | 8U | 8x B200 | 8-way NVLink | Air / liquid |
Use Case
The XE9712 is Dell's GB200 NVL72 offering and sits at the top of the Dell AI Factory line-up. Each 1U compute sled carries two Grace CPUs and four Blackwell GPUs; eighteen sleds and nine NVLink switch trays make up the 72-GPU rack, which the NVLink 5 fabric presents as a single 1.8 TB/s-per-GPU domain. That is what allows mixture-of-experts models with trillions of parameters to be trained or served without partitioning across separate 8-GPU islands. The rack is liquid cooled through a rack-level CDU and is normally quoted together with the IR9048 rack, power shelves and a 800G scale-out fabric.
Need Dell PowerEdge XE9712?
Enterprise hardware — configured, tested, deployed. Contact us for pricing and availability.
Request Quote