Liquid-to-Liquid Coolant Distribution Unit
SmartSense CDU is a high-performance, rack-mounted liquid-to-liquid modular Coolant Distribution Unit designed to meet the rigorous demands of GPU-dense, high-power racks. Capable of cooling up to 300kW per rack or across neighboring racks, it effectively manages heat in power-intensive environments, ensuring your data center infrastructure remains optimized for next-generation workloads.
Cooling Capacity | 192kW @ 5°C approach 301kW @ 7°C approach |
Power Consumption | 2,000W (Typ) 3,400W (Max) |
Fluid Temp. Range | 15°C to 70°C |
Flow Rate | 288 LPM (Typ) 400 LPM (Max) |
Scalable Liquid Cooling for Power-Dense Racks
SmartSense CDU delivers efficient, reliable cooling for high-performance AI workloads. With low-speed redundant pumps, warm inlet compatibility, and advanced filtration, it integrates seamlessly into modern data centers, accelerating AI time-to-solution.
SmartSense CDU Features
- Each 6U SmartSense CDU delivers 300 kW of cooling, with scalable configurations supporting up to 1.8 MW across six units per rack.
- SmartSense CDU scales from 2 to 6 CDUs per rack, enabling operators to add capacity incrementally while preserving rack space.
- Ultra-efficient, redundant pumps operate at low speeds (3600 RPM) for quiet, efficient, and reliable performance.
- Maximize GPU utilization with warm inlet temperatures up to 60°C, run at peak performance longer, accelerate computation-intensive processes and reduce time-to-solution for AI models.
- Versatile rack design compatible with 42-inch+ (1070 mm) deep racks in the 19-inch or ORV3 form factors, enabling seamless integration into modern data centers.
SmartSense CDU Features
- 3.4 kW of cooling per SmartSense 6U unit, with scalability up to 2.1 MW in row-level configurations.
- Ultra-efficient, redundant pumps operate at low speeds (3600 RPM) for quiet, efficient, and reliable performance.
- Maximize GPU utilization with warm inlet temperatures up to 60°C, run at peak performance longer, accelerate computation-intensive processes and reduce time-to-solution for AI models.
- Automatic pump switching ensures uninterrupted operation in case of pump failure or error.
- Advanced cross-connected filtration delivers low-loss, ultra-clean fluid distribution for enhanced system efficiency.
- Versatile rack design compatible with 42-inch+ (1070 mm) deep racks in the 19-inch or ORV3 form factors, enabling seamless integration into modern data centers.
- Ultra-low thermal loading for flow ensures more uniform device temperatures and extended device lifetimes (<1 kW/LPM).
We Keep AI Cool at Scale
As a Flex company, JetCool now delivers complete liquid-cooled rack solutions—designed, built, and scaled within Flex’s global manufacturing ecosystem. This integration combines JetCool’s patented microconvective cooling with Flex’s extensive global manufacturing capabilities, IT and power infrastructure expertise for seamless deployment at scale. That same strategy extends to our modular CDU that scales from 2 to 6 units per rack (up to 1.8 MW), giving operators the flexibility to start small, expand fast, and stay efficient—all without compromising rack space.
Order Now Scale Cooling for Next-Gen Racks up to 1.8MW
SmartSense CDU delivers a rack-level liquid cooling solution seamlessly supporting both single and multiple racks with scalable configurations supporting up to 1.8 MW across six units per standard rack.
Discover our Product Portfolio
We’ve designed and developed a variety of liquid cooling solutions to help enterprises in their transition to liquid cooling. From liquid cold plates to integrated die cooling, our product catalog supports a broad range of needs.
Discover our Product Portfolio
We’ve designed and developed a variety of liquid cooling solutions to help enterprises in their transition to liquid cooling. From liquid cold plates to integrated die cooling, our product catalog supports a broad range of needs.
NVIDIA H100 Cold Plate
Experience our fully-sealed, low-profile liquid cooling module, optimizing cooling and density for NVIDIA H100 PCIe cards.
SmartLid
SmartLid targets processor hot spots directly with fluid-to-die to densify compute performance and tackle the industry’s most powerful devices.
NVIDIA GB200
Experience our cold plate designed for the NVIDIA GB200 ‘Blackwell’ processor, engineered to tackle workloads exceeding 3kW.