Liquid to Liquid CDU Test Methodology and Performance Rating, Revision 1.0

Overview

The document says CDU performance definitions vary from supplier to supplier because of a lack of standardisation in design, rating and reporting. A liquid-to-liquid CDU is heat exchanger hardware that isolates the IT-side flow network from the facility loop and controls flow, pressure and temperature. Two main types are defined: the in-rack CDU, a standard U form factor chassis (for example 4U or 5U) connected to rack manifolds, and the in-row CDU, usually of larger capacity and serving several racks. Both connect to the facility water system through a heat exchanger.

The basic requirements table gives operating coolant temperature of 17-65 C (coolant must be delivered at least 2 C above dew point per ASHRAE), non-operating -40 to 85 C, relative humidity 10-90%, a 10-year service life (taken as the time for 10% of a pump population to fail), a primary filter of at most 200 microns and a secondary filter of at most 50 microns. Pump redundancy (N+1 or more) is essential to keep service during repair; wetted parts must be compatible with the coolant and 304/316 stainless steel is commonly used. Pressure supplied to the TCS can be as high as 100 psi (690 kPa), so TCS components must be designed for it.

Thermal performance is stated through approach temperature difference (ATD = TCS supply temperature minus FWS supply temperature) against the heat load supported at a given flow. Pressure head (TCS P-Q curves) and facility-side flow impedance are measured separately, with and without quick disconnects shown. The recommended design flow for rating is 1.5 LPM/kW, said to support most current chips. An in-rack CDU rating is X kW where TCS head is at least 25 psi at X x 1.5 LPM, FWS impedance is at most 75 psi and ATD is 5 C. For in-row units the TCS head is at least 40 psi.

Reliability tests cover hydrostatic pressure (IEC 62368-1 or ASME B31.1/B31.3), chemical compatibility (coupon tests, ASTM B117 salt spray, fluid analysis by ICP-MS, GC-MS, LC-MS and ion chromatography), dynamic testing (vibration IEC 60068-2-64, shock IEC 60068-2-27, ISTA shipping) and thermal cycling, with a pressure test after every mechanical and thermal test to check for leaks. Typical filter sizes are 25-50 microns on the TCS and 150-250 microns on the facility side. The paper presents itself as general test requirements and sets no vendor-specific acceptance threshold; it lists typical regulatory requirements for safety, EMC and material compliance.

Key points

1. Basic requirements: operating coolant 17-65 C (at least 2 C above dew point), non-operating -40 to 85 C, 10-90% RH, 10-year service life, primary filter at most 200 microns, secondary at most 50 microns.
2. Thermal performance is specified as ATD = TCS supply temperature minus FWS supply temperature versus heat load at given flows.
3. A recommended design flow of 1.5 LPM/kW is said to support most current chips.
4. In-rack CDU rating: X kW where TCS head is at least 25 psi at X x 1.5 LPM, FWS impedance at most 75 psi and ATD 5 C.
5. In-row CDU rating: as in-rack, but TCS head at least 40 psi at X x 1.5 LPM.
6. Reporting must state fluid type (water on the FWS, propylene glycol on the TCS), FWS temperature or ASHRAE water class, and flow rate.
7. Pressure supplied to the TCS can be up to 100 psi (690 kPa), and TCS components should be designed for it.
8. Typical filtration is 25-50 microns on the TCS and 150-250 microns on the facility side; hot-swappable filters and pumps are best practice.
9. Hydrostatic testing per IEC 62368-1 or ASME B31.1/B31.3, a pressure test after every performance test, then shock and vibration followed by leak checks.
10. N+1 or greater pump redundancy keeps service during repair, and a leak sensor rope is the common protection.

Using it in design

• Store CDU capacity in the catalogue using this rating basis (kW at 5 C ATD, 1.5 LPM/kW, head and FWS impedance) so units from different vendors can be compared, and derate when the design ATD or flow differs.
• In the liquid-loop calculator default to 1.5 LPM/kW and check that TCS supply is at least 2 C above room dew point.
• Specify 25-50 micron filtration on the TCS and 150-250 micron on the FWS.
• When selecting a CDU, compare the TCS P-Q curve (including quick disconnects) with in-rack network impedance, and the FWS curve with facility pump capacity.
• Ask the supplier for hydrostatic, shock/vibration and thermal-cycling test reports and wetted-material compatibility evidence.
• Choose in-rack CDUs at a U height that fits an EIA-310 19 in or ORv3 rack (4U, 5U); for in-row units check rack-size form (42U, 48U).

Who should read it

Mechanical engineers who design or buy CDUs, liquid cooling integrators and test laboratories.

Document contents

1. Introduction
2. Compliance with Open Compute Project Tenets
3. CDU Development Guidelines
4. General Liquid to Liquid CDU Requirements
5. Basic L-L CDU Design
6. Mechanical Design
7. Material Selection
8. Pump Selection
9. Cooling Loop Integration
10. Sensors & Controllers
11. Quality Control
12. Surface Treatment
13. Cleaning
14. Fluid Servicing
15. Filling & Flushing the System
16. Filter Servicing
17. Test Methodology
18. Mechanical Testing
19. Performance Testing & Reporting
20. Thermal Performance
21. TCS Pressure Head
22. FWS Flow Impedance
23. Recommendation on Performance Reporting
24. Reliability Testing
25. Hydrostatic Pressure Testing
26. Chemical Compatibility Testing
27. Dynamic Testing
28. Thermal Cycling
29. Typical Regulatory Requirements
30. Conclusion
31. Glossary
32. References

Related standards

• ASHRAE TC 9.9 Liquid Cooling Guidelines
• IEC 62368-1
• ASME B31.1
• ASME B31.3
• OCP Project Deschutes CDU specification

Read the document

Publication details

Authors

Wondimu, B.; Kung, E.; Zhou, D. (Intel); Zakaib, B. (CoolIT Systems); Chu, C. (Meta)

Publisher

Open Compute Project Foundation

Year

2024

Document number

Rev 1.0

Document language

English

Source type

Industry body

Access

Open licence — downloadable

Licence

CC BY-SA 4.0 — share with attribution under the same licence

Licence evidence

www.opencompute.org

Citation