Tag Archive for: Power capping
Application Notes, Data CenterLIC, BBU and the 1–50 ms Window: Transient Management in the 800 V Sidecar
LIC, BBU and the 1–50 ms Window: Transient Management in the 800 V Sidecar
How Lithium Ion Capacitors, battery backup units and SCMI power capping deal with AI workload power transients in 800 VDC rack architectures.
The 800 V sidecar solves the problem of power distribution to the compute rack. But it does not solve the problem of the power transients that AI silicon workloads generate on millisecond timescales. That window — between 1 and 50 ms — is the territory of the Lithium Ion Capacitor (LIC) and the Battery Backup Unit (BBU), and it is also territory where the discipline of industrial power electronics has a direct and immediate application.
The 1–50 ms window: why it matters
Artificial intelligence workloads generate power transients that did not exist in the previous generation of data centres. When a GPU cluster initiates an intensive compute phase — the first token of an inference, the start of a training step — power demand can ramp in milliseconds from an idle value to the rack’s maximum power. That power step has two consequences.
The first is electrical: the facility rectifier and distribution system cannot respond fast enough. The bus voltage momentarily drops until the supply chain stabilises. If that voltage drop exceeds the silicon’s tolerance threshold — typically ±5% of nominal — the GPU controller activates the throttling mechanism, reducing clock frequency until the voltage recovers. The result is a direct degradation of compute performance.
The second consequence is mechanical and reliability-related: repeated load step and unload cycles accelerate ageing of decoupling capacitors, power modules and rack converters. In racks of 125–140 kW such as NVIDIA’s NVL72, the energy involved in those transients is sufficient to damage components dimensioned for steady-state operation if it is not actively managed.
The SCMI management framework: power capping and PID loop
The OCP EMEA Summit 2026, held in Barcelona in April 2026, added a new dimension to the transient problem: coordinated management from the Data Centre Manager (DCM) level. A joint session by Oracle, AMI and Arm presented Arm SCMI (System Control and Management Interface) as the standard protocol for the DCM to impose power policies at the silicon and power converter level.
The POWERCAP_CAP_SET and POWERCAP_CAP_GET power capping commands allow the DCM to set a consumption ceiling per power domain. Levels CPL0 to CPL3 define four degrees of restriction, from the sustained level (CPL0) to the critical level (CPL3). Concurrent Power Capping (CPC) enables simultaneous restrictions across multiple domains.
The interaction with the closed thermal loop is equally relevant. Without a closed loop coordinating cooling with silicon power demand, the only response to a thermal transient is processor throttling — a performance degradation. With the DCM’s PID loop acting on pumps, valves and CDU flow rate, the system can increase cooling before the silicon reaches its thermal limit, preserving compute performance.
LIC and BBU: the state of the art at rack scale
The reference solution for transient management in the AI rack is a two-stage temporal architecture. The Lithium Ion Capacitor (LIC) covers the ultra-short window of 1 to 50 ms — the energy needed to hold the rack bus while the supply chain responds to the load step. The Battery Backup Unit (BBU) covers the seconds-to-minutes window — the backup needed for an orderly shutdown or to ride through micro-outages.
The LIC is neither a supercapacitor nor a lithium-ion battery. It is a hybrid technology that combines the high specific power of the supercapacitor with an energy density intermediate between the supercapacitor and the battery. In the rack context, a LIC module at rack scale typically provides between 100 and 500 Wh of usable energy with a discharge power of 10–50 kW for the 1–50 ms window, in a format compatible with standard rack U-space. The cycle life of a LIC — typically above 100,000 cycles at 80% depth of discharge — is compatible with the daily cycle count of an AI rack in production.
Implications for the sidecar power converter
For the sidecar converter, the two-stage LIC + BBU architecture implies specific design requirements. The 800 V DC bus must support the connection of LIC and BBU modules without intermediate conversion — which requires that bus voltage tolerance and protection coordination are designed for bidirectional energy flows. The converter supervisor must expose telemetry at the granularity and latency the DCM’s PID loop needs: instantaneous output power, semiconductor temperature, transformer temperature and instantaneous efficiency, updated at intervals below 100 ms.
The converter must also respond to SCMI POWERCAP_CAP_SET/GET commands in times consistent with the control loop dynamics. This is a firmware requirement that today is not standardised in most industrial converters, and one that converter manufacturers will need to address.
Lessons from railway power electronics
The transient management problem in the AI rack is structurally similar to two problems Premium SA has addressed in the railway environment: the management of regenerative braking transients in traction auxiliaries, and the management of load disconnection transients in EN 50155-certified on-board systems. In both cases, the critical window is in the range of tens of milliseconds, the energy involved is in the range of hundreds of watt-hours at power levels of tens of kilowatts, and the reliability requirement is above 99.9% availability over a lifetime of ten to twenty years.
The engineering approach applied to railway traction — conservative derating, accelerated thermal cycling qualification, FMEA at component level — is the same approach that the 1–50 ms transient management problem in the AI rack demands.
References
- Oracle, AMI, Arm — Systems Management session, OCP EMEA Summit 2026 (Barcelona, April 2026): Arm SCMI power capping architecture.
- Open Compute Project Foundation — Data Center Facility: Low Voltage DC Power Distribution white paper v1.0 (approved April 2026).
- NVIDIA — GB200 NVL72 rack power specification.
Power designed for your application
Talk to our engineering team about power conversion for 800 VDC data centre architectures.
Application Notes, Data CenterManageable Power Converters: PMBus, Redfish and SCMI in the 800 VDC AI Data Centre
Manageable Power Converters: PMBus, Redfish and SCMI in the 800 VDC AI Data Centre
Which protocols sit between the data centre manager and the power converter, where each one fits, and what the converter firmware needs to provide.
For decades, the relationship between a power converter and the system around it was almost one-way: the load demanded power and the converter supplied it. Signalling was limited to an enable signal, a fault signal and, at most, a voltage setpoint.
In the AI rack, that model is no longer enough. As rack power grows towards 800 VDC distribution architectures (covered in The 800 VDC revolution), the data centre manager needs to know at all times how much power each conversion stage is delivering, how much thermal margin it has and whether there are any alarms. The converter stops being a passive element and becomes a managed node.
This note explains which protocols are involved in that management chain, where each one fits and which firmware capabilities a converter needs to integrate into it.
The management chain: from the data centre manager to the converter
No single protocol links the data centre manager to the converter. Management is organised in layers, and each layer uses a different standard:
| Layer | Typical protocol | Function |
|---|---|---|
| Data centre manager ↔ rack controller / BMC | Redfish (DMTF), REST API over HTTPS | Inventory, telemetry, alarms and power policies at facility scale |
| Rack controller ↔ power shelf / BBU shelf | Modbus over RS-485 or CAN (common in Open Rack V3) | Reading measurements and status from power and backup shelves |
| Controller ↔ power supply or converter | PMBus | Telemetry and control of each conversion unit |
| Inside the compute node | Arm SCMI | Power and performance management of the silicon, including power capping |
For the converter manufacturer, the practical consequence is clear: the equipment must speak PMBus comprehensively and expose the information that the upper layers will translate into Redfish, either directly or through the shelf or rack controller.
PMBus: the converter’s own language
PMBus is the open digital communication standard for power supplies and converters. Its current published revision is 1.5. It defines a common set of commands to read measurements and status and to configure the equipment, for example:
- Measurements: READ_VIN, READ_IOUT, READ_POUT, READ_TEMPERATURE_1.
- Status and alarms: STATUS_WORD and the specific status registers (voltage, current, temperature).
- Configuration: warning and fault thresholds, and operating setpoints.
A converter with a complete, well-documented PMBus implementation simplifies integration: the rack controller can read its data without custom development.
Redfish: the facility view
Redfish is the DMTF standard for infrastructure management through a REST API with JSON data. For power it defines specific schemas, including PowerSubsystem, PowerSupply and PowerSupplyMetrics, as well as a generic sensor model.
The converter does not always need to implement Redfish itself. Typically, the shelf or rack controller collects the PMBus or Modbus data from each unit and publishes it in Redfish. What is essential is that the converter provides all the information those schemas expect: power, voltages, currents, temperatures, efficiency and status.
SCMI: where it really fits
Arm SCMI (System Control and Management Interface) is a standard, transport-independent interface between system software and the platform control firmware. Its natural scope is inside the compute node, not the link to the power supply. It defines protocols for discovery (Base), power domains, system power management, performance, clocks, sensors, reset, voltage and power capping (Powercap).
The Powercap protocol includes, among others, the POWERCAP_CAP_GET and POWERCAP_CAP_SET commands to read and set the power limit of a domain, and POWERCAP_MEASUREMENTS_GET to read average power. Version 4.0 of the specification, still in draft, adds Concurrent Power Limits (CPL) per domain and a telemetry protocol.
How does this relate to the converter? Power capping is executed by the silicon through SCMI, but the decision to cap depends on telemetry from the whole power chain, including the converter: if its thermal or power margin runs out, the manager can reduce compute consumption before the conversion stage operates outside its safe operating area.
Closed-loop thermal control
In liquid-cooled AI racks, thermal control is usually organised in two nested loops. The inner loop, in the coolant distribution unit (CDU), keeps the coolant temperature close to its setpoint by acting on pumps and valves. The outer loop, in the manager, adjusts that setpoint and the power limits according to the thermal telemetry from the racks.
To take part in the outer loop, the converter must publish the temperature of its power semiconductors, the temperature of its magnetic components and its instantaneous efficiency. With that data, the manager can decide whether there is margin for a transient increase in demand or whether power should be limited to protect the equipment’s reliability.
What the converter firmware must provide
- Complete PMBus telemetry: input and output voltage and current, power, internal temperatures and status registers.
- Interface to the rack controller in the protocol specified by the integrator, for example Modbus over RS-485 or CAN.
- Information compatible with the Redfish power model, published directly or through the controller.
- Configurable limits and derating, with event and alarm notification.
- Telemetry latency consistent with the control dynamics defined by the integrator.
- Controlled, traceable firmware updates.
References
- Arm — System Control and Management Interface (SCMI), specification DEN0056.
- DMTF — Redfish: PowerSubsystem, PowerSupply and PowerSupplyMetrics schemas.
- PMBus — Power Management Bus Specification, revision 1.5.
- Open Compute Project — Open Rack V3 specifications.
About Premium PSU
Premium PSU (Premium SA) designs and manufactures custom power supplies and power converters in Barcelona, with over 40 years of experience in railway, energy, industrial, data centre and defence applications.
Talk to our engineering team
Discuss your power conversion requirements with Premium PSU.
