ENERGY STAR Program Requirements for Computer Servers, Version 4.0 (Partner Commitments, Eligibility Criteria, and Final Test Method Rev. April 2023)

Overview

The document has three parts: partner commitments, eligibility criteria and the final test method. It covers blade, multi-node, rack and pedestal servers with up to four processor sockets. Fully fault-tolerant servers, server appliances, high-performance computing systems, large servers, hyperconverged servers, storage products and large network equipment are excluded. The specification took effect on January 12, 2024, and a product is judged against the specification in force on its date of manufacture.

Power supply requirements come in two tables. Single-output ac-dc supplies rated 750 W or more must reach at least 90%, 94%, 96% and 91% efficiency at 10%, 20%, 50% and 100% load; multi-output units of 750 W or more must reach 90%, 92% and 89% at 20%, 50% and 100%, with no 10% requirement. Power factor limits apply only where output is 75 W or more; for single-output units above 1,000 W the minimums are 0.80 at 10% load, 0.90 at 20%, and 0.95 at 50% and 100%. Testing follows the Generalized Internal Power Supply Efficiency Test Protocol Rev. 6.7.

Active-state efficiency is measured with the SPEC SERT tool. The EffACTIVE score is a weighted geometric mean of processor (65%), memory (30%) and storage (5%) sub-scores, each derived from performance per watt for individual workloads. Minimums vary by product type and processor count: 26.4 for one-processor rack servers, 30.4 for two-processor rack, 29.1 for two-processor blade or multi-node, and 31.9 for rack servers with more than two processors. Storage-heavy servers have no active-state threshold yet. Idle power must also be measured and reported.

Processor power management (DVFS or reduced power states) must be enabled by default, and any pre-installed hypervisor or operating system must offer power management too. A server must expose input power, inlet air temperature and average utilisation of all logical CPUs to third-party management software over a standard network. Power accuracy must be at least ±5% (never required better than ±10 W per PSU), temperature accuracy ±2 °C, sampled at least every 10 seconds. Blade and multi-node systems must offer real-time inlet temperature monitoring and fan speed management.

The test method fixes the conditions: ambient temperature 25 ± 5 °C, relative humidity 15% to 80%, a true-RMS power analyzer accurate to 1% or better, and measurement between the ac source and the unit with no UPS in between. The document covers IT equipment only, not facility infrastructure. Topics flagged for later revisions include dc servers, a review of PSU load points, an active-state criterion for storage-heavy servers and the next SERT release.

Key points

1. Version 4.0 took effect on January 12, 2024; a product is assessed against the specification in force on its date of manufacture.
2. Single-output ac-dc PSUs of 750 W or more must reach at least 90% / 94% / 96% / 91% efficiency at 10% / 20% / 50% / 100% load.
3. Multi-output ac-dc PSUs of 750 W or more must reach at least 90% / 92% / 89% at 20% / 50% / 100% load; there is no 10% requirement.
4. Below 750 W the thresholds are lower: 83% / 90% / 94% / 91% for single-output, and 87% / 90% / 87% for multi-output (20% / 50% / 100% load).
5. For single-output PSUs above 1,000 W the power factor must be at least 0.80 at 10% load, 0.90 at 20% and 0.95 at 50% and 100%, applied only where output is 75 W or more.
6. Minimum EffACTIVE: 26.4 for one-socket rack, 30.4 for two-socket rack, 29.1 for blade or multi-node (two processors); tower and resilient types have separate thresholds.
7. EffACTIVE = EXP(0.65 ln EffCPU + 0.30 ln EffMEMORY + 0.05 ln EffSTORAGE).
8. Servers report input power to ±5%, inlet air temperature to ±2 °C, sampled at least once every 10 seconds.
9. Testing uses 25 ± 5 °C and 15%-80% relative humidity, with power measured between the ac source and the unit and no UPS in the path.
10. Scope is IT equipment only; hyperconverged, large and fully fault-tolerant servers and storage products are excluded.

Using it in design

• Write ENERGY STAR Version 4.0 certification and the PSU efficiency steps (at 10% to 100% load points) into the server procurement specification; look especially at 10% and 20% load efficiency for redundant supplies that run lightly loaded.
• When estimating rack-level PSU losses, use the 96% floor at 50% load for single-output units above 750 W, but set the real load ratio carefully, since redundant feeds leave each supply lightly loaded.
• Check power factor in UPS and distribution design: values of 0.65 to 0.80 at 10% load affect UPS and cable sizing in highly redundant layouts.
• Write the requirement that servers expose input power, inlet temperature and CPU utilisation over a standard network into DCIM or BMS integration requirements.
• Compare products on the complete SERT report and EffACTIVE score rather than a single workload result; EPA prohibits selective reporting.

Who should read it

Designers and IT procurement teams who write server specifications, calculate rack power budgets and PSU losses, or define monitoring integration.

Document contents

1. Definitions
2. Scope
3. Certification Criteria
4. Power Supply Requirements
5. Power Management Requirements
6. Blade and Multi-Node System Criteria
7. Active State Efficiency Criteria
8. Idle State Efficiency Criteria
9. Standard Information Reporting Requirements
10. Standard Performance Data Measurement and Output Requirements
11. Testing
12. Effective Date
13. Considerations for Future Revisions
14. Final Test Method: Test Setup and Test Conduct

Read the document

Publication details

Authors

U.S. EPA ENERGY STAR

Publisher

U.S. EPA / ENERGY STAR

Year

2024

Document language

English

Source type

Government / public

Access

Open licence — downloadable

Licence

U.S. public domain — U.S. Government work; free to reuse

Licence evidence

www.copyright.gov

Citation