CRPS Redundant Power Supplies and Cold Redundancy in Dedicated Servers: Datacenter Guide

Operating mission-critical bare-metal servers in Pakistani datacenters (such as facility clusters in Karachi, Lahore, and Islamabad) presents unique power engineering challenges. With municipal grid voltage sags, power outages, diesel generator transfer intervals, and high ambient temperatures, server power infrastructure must deliver uninterrupted DC power rails to multicore CPUs and NVMe storage arrays.

Modern 1U and 2U enterprise rackmount chassis have standardized on the CRPS (Common Redundant Power Supply) open specification established by Intel. Delivering up to 2000W+ in a compact modular form factor with 80 Plus Titanium efficiency (over 96% power conversion at 50% load), CRPS units eliminate single points of failure. Furthermore, by implementing firmware-level Cold Redundancy, servers avoid the efficiency penalties of dual-PSU load sharing during low-utilization idle cycles.

Deploying high-reliability Dedicated Servers in Pakistan and enterprise Dedicated Servers configured with dual CRPS power modules connected to independent A+B utility and UPS power feeds ensures 99.999% uninterrupted uptime.

---

Understanding the CRPS Specification and Modular Design

Before CRPS, server chassis manufacturers (Dell, HPE, Supermicro, Inspur) designed proprietary power supplies with non-interchangeable edge connectors and custom mechanical retention brackets. Replacing a failed PSU required waiting days for proprietary OEM replacements.

The Intel CRPS (Common Redundant Power Supply) specification standardized:

1. Universal Mechanical Dimensions: Standardized 1U width (73.5 mm), height (40.0 mm), and depths (185 mm or 195 mm), enabling cross-vendor compatibility.
2. Unified Golden Finger Edge Connector: Standardized pin assignments for +12V main DC rail, +12Vsb (standby), return ground, and SMBus/I2C communication lines.
3. PMBus (Power Management Bus) Protocol: Standardized registers allowing Baseboard Management Controllers (BMCs) and Linux OS drivers to read real-time voltage, current, temperature, and fan speeds.

+---------------------------------------------------------------+
|               DUAL CRPS A+B FEED ARCHITECTURE                 |
|                                                               |
|  [ Power Feed A (Grid + UPS 1) ]    [ Power Feed B (Gen + UPS 2) ]
|                |                                    |         |
|                v                                    v         |
|       [ CRPS Module 1 (80+ Ti) ]           [ CRPS Module 2 (80+ Ti) ]
|                |                                    |         |
|                +-----------------+------------------+         |
|                                  |                            |
|                       Active +12V DC Common Bus               |
|                                  |                            |
|             +--------------------+--------------------+       |
|             |                                         |       |
|             v                                         v       |
|   [ CPU VRMs (Core Rails) ]                 [ NVMe Backplanes ]
+---------------------------------------------------------------+

For hardware engineers evaluating complementary power and memory reliability technologies, explore our analysis of ECC Memory Patrol Scrubbing Rate and Error Correction Tuning, examine motherboard power distribution in DDR5 On-DIMM PMIC Power Management in Enterprise Racks, and review non-volatile write caching in BBU vs Supercapacitor Flash Cache in RAID.

---

The Efficiency Trap: Load Sharing vs. Cold Redundancy

In standard dual-PSU configurations, servers traditionally operate in Load Sharing (Active/Active) mode: both PSUs contribute 50% of the total load.

However, switching power supply efficiency curves are non-linear:

• At 50% load, an 80 Plus Titanium PSU reaches peak efficiency (> 96%).
• At 10% to 15% load, efficiency plummets to 80–84%, wasting power as parasitic heat.

If a dual 1200W server consumes only 240W at idle, each PSU provides 120W (only 10% load), operating in the most inefficient zone of its conversion curve!

The Cold Redundancy Solution:

Under Cold Redundancy (Active/Cold Standby):

1. Primary PSU (Active): Supplies 100% of the server’s current (240W = 20%–50% load), operating at maximum electrical efficiency.
2. Secondary PSU (Cold Standby): Puts its high-power switching transistors into a low-power quiescent state while keeping internal capacitors charged and the +12Vsb logic bus active.
3. Hardware Microsecond Handover: If the primary PSU fails or detects input AC voltage drop, the secondary PSU wakes up in sub-100 microseconds ($\mu s$), well within the 10-millisecond holdup time of the server motherboard bulk capacitors!

---

Step 1: Querying PMBus Telemetry via Linux ipmitool

Because CRPS modules interface with the motherboard BMC via the PMBus over I2C, system administrators can query live electrical metrics directly from the Linux CLI:

# Query IPMI sensor data for Power Supplies
ipmitool sdr type "Power Supply"

# Query detailed PSU voltage, wattage, and temperature readings
ipmitool sensor | grep -E "PS[12]|POUT|PIN|VIN"

Sample telemetry from a Supermicro dual-CRPS node:

PS1 Input Watts  | 285.000    | Watts      | ok
PS2 Input Watts  | 0.000      | Watts      | ok (Cold Standby)
PS1 Temp         | 34.000     | degrees C  | ok
PS2 Temp         | 26.000     | degrees C  | ok
PS1 Status       | 0x01       | discrete   | 0x0100 (Presence Detected, Failure Asserted: No)
PS2 Status       | 0x01       | discrete   | 0x0100 (Presence Detected, Standby Mode)

Notice that PS2 draws 0 watts of active conversion load while standing by in cold redundancy mode.

---

Step 2: Configuring Redundancy Modes in BIOS/BMC

To toggle between Active/Active (Load Sharing) and Active/Standby (Cold Redundancy), access the server BMC IPMI web console or use the Redfish API:

# Redfish API command to inspect and set Power Supply redundancy mode
curl -k -u admin:password -X PATCH \
  https://bmc-ip/redfish/v1/Chassis/1/Power \
  -H "Content-Type: application/json" \
  -d '{"Redundancy": [{"Mode": "Failover"}]}'

Modes supported:

• Failover (Cold Redundancy): Primary PSU carries the load; secondary sleeps.
• Sharing (Load Balancing): Load divided equally. Recommended only for heavy GPU clusters exceeding 70% continuous power envelope.

---

Step 3: Zero-Downtime Hot-Swap Procedure

When replacing a failed or degraded CRPS module in a live production environment:

1. Verify Feed Redundancy: Check ipmitool sensor to confirm the remaining healthy PSU is actively supporting 100% of the chassis load and reporting ok.
2. Release Safety Latch: Depress the colored (typically green or blue) mechanical retention clip on the failed module.
3. Disengage AC Cord: Unclip and pull the locking C13/C14 power cable retaining clamp.
4. Smooth Extraction: Firmly pull the handle outward. The edge connector disengages smoothly from the midplane golden finger socket without introducing arcing.
5. Insert Fresh Unit: Slide the replacement CRPS module firmly until the latch clicks.
6. Reattach AC Cable: Reconnect power. The LED indicator will flash amber (self-test) and turn solid green within 3 seconds as the BMC synchronizes PMBus registers.

---