A pre-integrated high-density rack for AI training and HPC workloads: 8–10 dense GPU nodes, direct-to-chip liquid cooling with in-rack CDU, high-power busbar distribution, and 400GbE / NDR InfiniBand fabric — engineered to integrate with your existing chilled-water plant or run standalone via rear-door heat exchanger.
The HR-Series can plumb into your existing chilled-water plant via facility water, or run standalone with a rear-door heat exchanger — dropping into an air-cooled aisle without touching the facility loop. Same rack, different manifold.
A copper busbar runs the full height of the rack, feeding each GPU node via dropdown breakers. No cable spaghetti at 80kW.
Top-of-rack 400GbE spine plus optional NDR InfiniBand for GPU-Direct RDMA workloads. Cabled and validated at factory before it ships.
Deployed for GCC sovereign LLM training programs and university research clusters, with regional firmware baseline signed in-region and Dubai engineering ownership.
Configurable options are marked red. Contact engineering for special-order variants and turnkey integration.
| Height | 42U · deep chassis (1200mm) |
|---|---|
| Width | 800mm |
| Doors | Perforated front · rear-door heat exchanger option |
| Cable Mgmt | Integrated fiber runners + power drops |
| Weight | 1,400 kg fully populated (indicative) |
| Type | Direct-to-chip liquid · in-rack CDU |
|---|---|
| Coolant | PG25 · closed loop |
| CDU Capacity | 80 kW+ per rack · configurable to 120 kW |
| Facility Interface | Chilled water OR RDHx (self-contained) |
| Air Backup | Front intake blowers for ambient loads |
| Distribution | Copper busbar · full-height |
|---|---|
| Input | 3-phase 63A / 100A · configurable |
| Redundancy | Dual feed · A + B · downstream breakers |
| Monitoring | Per-node current + facility kWh · streamed to Fabric |
| Compute Nodes | 8–10× T18 AI8-HPC (4U) · SXM5 |
|---|---|
| ToR Fabric | 400GbE · NDR InfiniBand optional |
| Management | T18 BMC + Redfish · Fabric integration |
| Sensors | Per-node coolant flow + temperature |
Foundation model training clusters for regional cloud programs with data residency requirements.
CFD, seismic, and molecular dynamics clusters across academic and industrial customers.
Batch inference and RAG serving with pooled accelerators.
Drop HR-Series into air-cooled halls via RDHx — no facility redesign.
Iterative fine-tuning workloads for enterprise and government tenants.
Bespoke pods for hyperscale customers with custom fabric and power topologies.