Server racks for liquid-cooled servers in data centers

For high-density data centers, liquid-cooled server racks such as Flex OCP ORv3, Supermicro rack-scale solutions, and direct-to-chip or immersion cooling systems are recommended for efficient thermal management and scalability.Key Liquid Cooling Methods1. Direct-to-Chip (DLC) Cooling: Coolant flows through cold plates mounted directly on CPUs, GPUs, and DIMMs, providing targeted heat removal. DLC is highly efficient for high-density racks, supporting server densities up to 80 kW per rack and beyond, and is suitable for AI and HPC workloads . 2. Immersion Cooling: Servers are fully or partially submerged in dielectric fluid, allowing complete heat coverage across all components. This method is ideal for extreme-density deployments exceeding 100 kW per rack . 3. Rear-Door Heat Exchangers: Liquid-to-the-rack cooling uses chilled water coils in rear doors to remove heat from air exiting the rack, suitable for moderate-density racks and hybrid cooling setups .Recommended Rack SolutionsFlex Liquid-Cooled Racks:Supports up to 120 kW per rack with single-phase and two-phase liquid cooling.Integrates JetCool SmartPlate™ direct-to-chip technology for high-power GPUs and AI accelerators.Customizable for AI and HPC workloads with high thermal headroom . Supermicro Rack-Scale Solutions:End-to-end liquid cooling including servers, racks, networking, and in-rack CDUs.Supports densities up to 80 kW per rack with redundant hot-swappable pumps.Offers both direct liquid cooling and immersion options, reducing PUE and TCO while enabling rapid deployment of AI infrastructure . Hybrid Cooling Approaches:Combining air cooling for low-density racks with liquid cooling for high-density racks optimizes cost and efficiency.Incremental deployment allows retrofitting existing data centers without full redesign .Deployment ConsiderationsConduct a site audit and CFD analysis to evaluate airflow and existing cooling capacity.Size piping, manifolds, and coolant distribution units (CDUs) appropriately for the chosen liquid cooling method.Plan for future scalability, especially for AI/HPC workloads with high TDP CPUs and GPUs.Ensure collaboration between IT, facilities, and power teams to integrate liquid cooling with existing infrastructure .ConclusionFor modern data centers handling AI, HPC, or high-density workloads, direct-to-chip and immersion liquid-cooled racks from vendors like Flex and Supermicro provide the most efficient and scalable solutions. Hybrid approaches can optimize cost and flexibility, while careful planning ensures reliable operation and future-proofing.