Specifications
Product Family
Stäubli Universal Quick Disconnect UQD (hand-mate) and UQDB (blind-mate)
Sizes
UQD02, UQD04, UQD06 and UQD08
Nominal Diameter
DN03, DN06, DN07 and DN10 mm
Maximum Allowable Pressure
16 bar (PS)
Temperature Range
-20 °C to +150 °C (TS)
Shut-Off
Double shut-off (flat-face, non-spill on both halves)
Seal Material
Ethylene-Propylene (EPDM), JE code
Connection Thread (example)
UNF 9/16-18 in on UQDB04.5414/UN/JE
Key Dimensions (UQDB04)
25.4 mm body, 35.4 mm total length, 10 mm flats, 24 mm hex
Blind-Mate Compensation
Guide geometry with floating compensation; maximum misalignment of 3 mm (radial ±2.5–5 mm, angle ±2.7°)
OCP Standard
Interface dimensions and performance to the Open Compute Project Universal Quick Disconnect specification
Max Fluid Loss (OCP)
0.020 ml (UQD02), 0.025 ml (UQD04), 0.035 ml (UQD06), 0.070 ml (UQD08) per connection/disconnection at 0 psi
Drip Rate
Lower than the OCP drip requirement
Working / Burst Pressure (OCP)
100 psi maximum working pressure; 300 psi minimum burst pressure
Flow Rating (OCP)
≥ 0.55 GPM (UQD02), ≥ 1.7 GPM (UQD04), ≥ 3.0 GPM (UQD06), ≥ 4.7 GPM (UQD08)
Cv Value, Water (OCP)
0.25 (UQD02), 0.8 (UQD04), 1.6 (UQD06), 2.5 (UQD08)
Service Life
5,000 connection/disconnection cycles; 10-year service life; 5-year storage life
Operating / Transport Temp (OCP)
17 °C to 65 °C operating; -40 °C to 75 °C transport
Locking Feedback
Visual, tactile and/or audible confirmation of full connection
Fluids
Glycol water and dielectric liquid cooling fluids
Qualification
UQD and UQDB range approved by NVIDIA for AI liquid cooling deployments
Applications
High-performance computers, AI data centers, internet/network switches
Overview
Every liquid-cooled GPU rack depends on a coupling that can be opened without dumping coolant on the servers, and the Stäubli UQD/UQDB range is the reference implementation of that part. The couplings are flat-face, double shut-off, non-spill devices built to the Open Compute Project Universal Quick Disconnect interface specification, which means they are dimensionally and hydraulically interchangeable with other OCP-compliant couplings from Stäubli, Parker or CPC.
The distinction that matters most in an NVL72-class rack is hand-mate versus blind-mate. Standard UQD couplings are mated by an operator; UQDB adds guide geometry, a floating compensation mechanism and an expanded self-alignment window so a GPU tray or sled can be slid into position and connect to the manifold automatically. The published tolerance is a maximum misalignment of 3 mm, with a radial self-alignment window of ±2.5–5 mm and angular tolerance of ±2.7°. This is the enabling technology behind trays that connect on insertion without anyone reaching in to align a coupling.
Four sizes — UQD02, UQD04, UQD06 and UQD08, at DN03 through DN10 — carry the same pressure and temperature envelope: 16 bar maximum allowable pressure and -20 °C to +150 °C. Their flow capacity scales with size, from 0.55 GPM at UQD02 to 4.7 GPM at UQD08, with Cv values of 0.25 to 2.5. Fluid loss per connection is capped at 0.020 ml on the smallest size and 0.070 ml on the largest. The range is approved by NVIDIA for AI liquid cooling.
Key Benefits
Flat-face, double shut-off, OCP-compliant and interchangeable with other OCP UQD couplings. Blind-mate UQDB self-aligns to ±2.5–5 mm radial and ±2.7° angular, with 3 mm maximum misalignment, enabling tray insertion without manual alignment. 16 bar, -20 to +150 °C, 5,000 mating cycles and fluid loss as low as 0.020 ml per connection.
Applications
NVIDIA NVL72-class GPU tray and sled connections, CDU branch and manifold connections, server line and test setups, tray-level blind-mate slide-in racks, and any direct-to-chip loop where a coupling must be opened without spilling coolant onto live electronics.
Request a Quote — STÄUBLI UQD & UQDB — UNIVERSAL QUICK DISCONNECTS FOR AI LIQUID COOLING
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →