Key Takeaways
- Supermicro’s new Open Compute Project (OCP) portfolio adds ORv3 and ORW racks, rear‑door heat exchangers (RDHx), and a 48 V DC busbar architecture.
- Centralized DC power and DLC‑2 Direct‑to‑Chip liquid cooling raise compute density to up to 120 kW per rack while cutting cabling by 30‑40 %.
- RDHx units (10 kW‑120 kW) remove residual heat at the rack rear, eliminating the need for additional data‑hall air‑conditioning upgrades.
- Integrated management via SuperCloud Composer and related tools provides end‑to‑end monitoring, automation, and anti‑condensation protection.
- The solution shortens Time‑to‑Market (TTM) for AI/HPC deployments by up to 30 % compared with traditional AC‑powered designs.
Introduction: Why OCP Matters for AI‑Driven Data Centers
The Open Compute Project (OCP) sets community‑driven standards that enable data‑center operators to build high‑density, energy‑efficient racks without vendor lock‑in. Supermicro’s latest portfolio, validated through the DC‑Powered Busbar System (DCBBS), extends these standards to AI and high‑performance computing (HPC) workloads, delivering a turnkey rack‑scale foundation from power distribution to cooling.
Central Innovation: 48 V DC Busbar Architecture
Traditional servers rely on per‑node AC‑to‑DC converters and a maze of power cables. Supermicro’s ORv3 and ORW racks replace that clutter with a single 48 V DC busbar that feeds every node directly. Benefits include:
- Reduced power loss: <2 % conversion loss vs. 5‑7 % for distributed converters.
- Lower cable count: Up to 35 % fewer cables, easing installation and maintenance.
- Higher power headroom: Supports up to 120 kW per rack, enabling dense GPU clusters for AI training.
Liquid‑to‑Chip Cooling: DLC‑2 Technology
The DLC‑2 Direct‑to‑Chip loop circulates coolant directly onto CPUs, GPUs, and high‑power ASICs. Compared with traditional air cooling:
| Metric | Air‑Cooled (baseline) | DLC‑2 Liquid‑Cooled |
|---|---|---|
| CPU/GPU thermal resistance (°C/W) | 0.45 | 0.12 |
| Power density achievable (kW/rack) | 30‑45 | 80‑120 |
| Cooling‑system footprint | 2U (fans) | 1U (pump + reservoir) |
| Energy used for cooling (kW) | 0.12 kW per 10 kW compute | 0.07 kW per 10 kW compute |
The liquid loop extracts heat at the source, allowing the ORv3 RDHx to handle only residual heat, which dramatically reduces the load on data‑hall HVAC.
Rear‑Door Heat Exchangers (RDHx) – From 10 kW to 120 kW
Supermicro’s RDHx units mount directly to the rear of ORv3/ORW racks. Key specifications:
| Model | Cooling Capacity | Fan Redundancy | Condensation Guard | Power Consumption |
|---|---|---|---|---|
| RDHx‑10 | 10 kW | N+1 (2 fans) | Smart anti‑condensation sensor | 0.6 kW |
| RDHx‑60 | 60 kW | N+1 (4 fans) | Same | 3.2 kW |
| RDHx‑120 | 120 kW | N+1 (6 fans) | Same | 6.4 kW |
All units connect to the busbar, eliminating extra AC feeds. The N+1 fan architecture guarantees uptime ≥ 99.9 % (MTBF > 150,000 h). The anti‑condensation system monitors rack humidity and shuts off coolant flow if moisture thresholds are approached, protecting sensitive components.
Unified Management with SuperCloud Suite
Operators can control the entire OCP stack through SuperCloud Composer, which supports SNMP, Redfish, and IPMI. Complementary tools include:
- SuperCloud Director: Centralized resource orchestration for AI workloads.
- SuperCloud Automation Assistant: Policy‑driven provisioning of power and cooling profiles.
- SuperCloud Developer Console: API‑first interface for custom telemetry and integration.
These tools enable real‑time power‑usage analytics, automated fan‑speed adjustments, and predictive maintenance alerts, reducing operational expenses by an estimated 15‑20 %.
Comparison: ORv3 vs. ORW Racks
| Feature | ORv3 (Standard) | ORW (Wide) |
|---|---|---|
| Width | 600 mm (23.6 in) | 800 mm (31.5 in) |
| Max Power | 120 kW | 120 kW |
| Supported GPUs | Up to 8 × NVIDIA H100 per node | Up to 6 × NVIDIA H100 per node (larger spacing) |
| Cooling Integration | RDHx‑10/60/120 | RDHx‑10/60/120 |
| Deployment Target | High‑density AI clusters | Mixed AI + storage or networking |
Both models share the 48 V busbar and DLC‑2 cooling, but the ORW offers a wider chassis for larger interconnect modules or additional storage bays.
Bottom Line
Supermicro’s expanded OCP portfolio delivers a fully open, DC‑powered, liquid‑cooled rack ecosystem that directly addresses the power‑density and thermal challenges of modern AI and HPC workloads. By consolidating power distribution, cutting cabling, and off‑loading heat at the rack rear, operators can achieve up to 120 kW per rack with 30 % lower TTM and **15‑20 %