From Cold Plate to Immersion: Why the Pump Is the Key Component in Liquid Cooling for Computing Power
In the era of large AI models, computing power demand is surging exponentially. For instance, a single NVIDIA H100 GPU consumes 700 watts, while the B200 has exceeded 1,000 watts. An 8-GPU server boasts a total power consumption of nearly 10 kW, and a fully loaded GPU cluster cabinet achieves a power density ranging from 40 to 80 kW. Traditional air cooling has hit its physical limits: air’s heat capacity is merely one-fourth that of water, and its thermal conductivity trails water by two orders of magnitude. Air cooling solutions cease to function once the chip heat flux density surpasses 100 W/cm².
Liquid cooling has become the only viable solution for high-density computing power. What lies at the core of a liquid cooling system? It is not the cold plate, nor the CDU, nor the piping — it is the pump. The pump serves as the "heart" of the liquid cooling system. Without precise, reliable and durable fluid delivery, even the most sophisticated cold plate designs and high-efficiency coolants cannot deliver their full performance. The performance of the pump directly determines heat dissipation efficiency, operational stability and total lifecycle cost.
▼ Five Core Technical Barriers of Liquid Cooling Pumps
In a liquid cooling system, the pump’s function is far more than merely "circulating liquid". It has to overcome five major technical challenges simultaneously:
| Technical Barrier | Core Requirement | Consequence of Noncompliance |
| Precise Flow Control | Flow fluctuation ≤ ±1% | GPU junction temperature drift → Cluster computing power loss |
| Ultra-Low Pulsation | Pressure/flow pulsation < 1% | Micro-vibration → TIM degradation, solder joint fatigue |
| Chemical Compatibility | All wetted components resistant to coolants | Pump body corrosion → Coolant contamination across the entire cabinet |
| Long-Term Reliability | MTBF ≥ 50,000 hours | Unscheduled downtime → Hundreds of millions in business losses |
| High Energy Efficiency | Fluid delivery efficiency > 90% | Deteriorated PUE → Loss of liquid cooling economic benefits |
▼ Cold-Plate GPU Servers: The Optimal Application Scenario for Precision Magnetic Gear Pumps
The Essence of Cold-Plate Liquid Cooling
Cold-plate liquid cooling is currently the mainstream solution for AI training servers. Liquid cooling cold plates are mounted directly against GPU chip surfaces; coolant flows through internal microchannels to carry away heat via forced convection.
Key parameter: Microchannels inside cold plates feature characteristic dimensions of only 0.2–0.5 mm. At this scale:
Insufficient flow velocity → Local boiling → Chip burnout
Pressure pulsation → Cold plate micro-vibration → TIM layer degradation
Uneven flow distribution → Temperature differentials across multiple GPUs → "Bucket effect" limiting overall cluster performance
Magnetic Gear Pumps vs. Centrifugal Pumps vs. Diaphragm Pumps
| Comparison Dimension | Precision Magnetic Gear Pump | Centrifugal Pump | Diaphragm Pump |
| Flow Pulsation | <1% (ultra-low) | 3–8% | Periodic pulsation |
| Flow Accuracy | ±0.5% | ±5–10% | ±3–5% |
| Back Pressure Adaptability | Rigid (constant flow rate) | Flexible (flow declines under backpressure) | Moderate |
| Low-Flow Stability | Excellent | Poor | Moderate |
| Service Life & Reliability | Extremely High | High | Diaphragm fatigue risk |
| Maintenance Complexity | Maintenance-free for 3 years (only 2 rotating parts) | Moderate | High (regular diaphragm replacement required) |
JONSN Precision Magnetic Gear Pump Solutions
With years of expertise in precision magnetic gear pump technology, JONSN delivers fully validated solutions for cold-plate liquid cooling scenarios, supported by three core strengths: extended service life, compact footprint, and 3-year maintenance-free operation.
| Technical Feature | JONSN Solution | Industry Value |
| Manufacturing Precision | Mesh clearance ≤ 5 μm, surface roughness Ra ≤ 0.2 μm | Volumetric efficiency ≥ 95% |
| Pulsation Control | Multi-tooth meshing + optional dampers, pulsation < 0.3% | Near-zero risk of micro-vibration |
| Flow Accuracy | ±0.5% with closed-loop PID control | GPU junction temperature fluctuation < ±1°C |
| Material Options | Standard 316L stainless steel / optional Hastelloy | Fully compatible with mainstream coolants |
| Drive Type | Magnetic drive (zero leakage) / mechanical seal variant | Eliminates leakage risks from dynamic seals |
| Design Service Life | ≥ 50,000 hours (3 years maintenance-free) | Matches the hardware refresh cycle of servers |
▼ Three Core Advantages of JONSN Precision Magnetic Gear Pumps
Extended Service Life – 50,000-hour design life matching the full lifecycle of GPU servers
JONSN precision magnetic gear pumps adopt high-hardness gear materials and ultra-precision machining (mesh clearance ≤ 5 μm, Ra ≤ 0.2 μm). Magnetic coupling drive eliminates mechanical friction, delivering continuous runtime exceeding 50,000 hours. With data centers operating 8,760 hours per year, the pumps support GPU servers for over 5 years of full lifecycle operation, eliminating mid-term pump replacement needs.
Ultra-Compact Footprint – Space-saving magnetic coupling design for valuable data center floor space
Magnetic coupling replaces traditional mechanical seal shaft assemblies, removing auxiliary components such as seal glands and flushing piping. The unit volume is only one-third that of equivalent centrifugal pumps. For high-density 1U/2U server chassis, JONSN pumps can be flexibly integrated near CDUs or cold plate manifolds, simplifying piping layouts and reducing flow resistance losses.
3-Year Maintenance-Free Operation – Zero leakage magnetic drive cuts 3-year maintenance costs entirely
JONSN precision magnetic gear pumps contain just two rotating components (driving and driven gears). Magnetic coupling enables contactless torque transmission, eliminating mechanical seal wear and fluid leakage risks. Three years of maintenance-free operation is a structural guarantee: no seal replacements, no lubricant refills, and no routine inspections required, drastically cutting data center labor overhead and total lifecycle costs.
JONSN has been approved as a qualified supplier by NAURA Technology Group, established in-depth cooperation with HZC, and signed supply agreements with Advanced Micro-Fabrication Equipment (AMEC). Recognition from leading semiconductor manufacturers validates JONSN’s leading supporting capacity and benchmark standing in high-end precision fluid equipment.
▼ Immersion Cooling Data Centers: Pumping Challenges for High-Volume Circulation
Two Immersion Cooling Technical Routes
Immersion cooling submerges entire servers directly in dielectric coolant, split into two mainstream routes:
| Single-Phase Immersion Cooling | Two-Phase Immersion Cooling | |
| Operating Principle | Heat dissipation via liquid convection | Heat dissipation via boiling on chip surfaces (latent heat of vaporization) |
| Coolant Media | Synthetic hydrocarbons / silicone oil / fluorinated fluids | Low-boiling-point fluorinated fluids (~50°C boiling point) |
| Coolant Cost | Tens to hundreds of RMB per liter | Hundreds to thousands of RMB per liter |
| Circulation Driving Mechanism | Pump-driven primary circulation | Passive thermosiphon + auxiliary pump circulation |
| Heat Dissipation Capacity | ~40 kW per cabinet | >100 kW per cabinet |
The Role of Pumps in Immersion Cooling Systems
Single-Phase Immersion Cooling – Pumps form the core circulation power
Each cabinet requires a circulation flow rate of 50–200 L/min to maintain vertical temperature differentials within the tank below 3°C. A mere 5% reduction in flow rate can raise top-layer GPU junction temperatures by 8–10°C. Bypass filter pumps are also required to sustain coolant cleanliness.
Two-Phase Immersion Cooling – Indispensable auxiliary pumping
Primary circulation relies on passive phase-change heat transfer, yet condenser water circulation pumps, precision fluorinated fluid makeup pumps, and emergency circulation pumps remain mandatory.
Special Pump Requirements for Immersion Cooling
| Requirement | Technical Description | Recommended Solution |
| Full Media Compatibility | Fluorinated fluids cause severe swelling in standard rubber materials | All-metal wetted components + PTFE/PEEK seals |
| High Flow Capacity | 50–200 L/min per cabinet | Centrifugal pumps for main circulation + gear pumps for precision loops |
| Zero Leakage | Fluorinated fluids carry high GWP; even minor leaks incur environmental and economic losses | Magnetic drive pumps are mandatory |
| Low NPSH Design | Coolants operate near saturation point, prone to cavitation | Enlarged pump inlets + low net positive suction head design |
| Modular Quick Swap | Tank maintenance requires full coolant drainage (over 24 hours downtime) | External tank mounting + N+1 redundant design + quick-disconnect fittings |
Recommended Tiered Pumping Architecture
Primary Pumps (Main Circulation): High-flow stainless steel centrifugal pumps for bulk circulation, N+1 redundant configuration, variable frequency speed regulation.
Secondary Pumps (Precision Loops): JONSN MR series miniature precision magnetic gear pumps, used for hot-spot flow enhancement, precise bypass filtration control, and quantitative coolant makeup.
Tertiary Auxiliary Pumps: Condenser water circulation, emergency fluid makeup, and system priming & draining.
▼ Conclusion: Pumps Are the Core of the Liquid Cooling Era
When cooling precision tightens to ±1°C, when coolants cost thousands of RMB per liter, and when every minor pressure pulsation threatens GPU service life — pumps transcend ordinary fluid equipment to become strategic core components of liquid cooling systems.
| Dimension | Cold-Plate GPU Servers | Immersion Cooling Data Centers |
| Core Pump Role | Precision flow control (critical heart component) | High-volume circulation + precise fine-tuning |
| Optimal Pump Type | JONSN precision magnetic gear pumps | Centrifugal pumps (main circulation) + gear pumps (precision regulation) |
| Key Performance Metrics | Near-zero pulsation, ±0.5% flow accuracy | High flow rate, zero leakage, low NPSH |
| Consequence of Pump Failure | GPU throttling → Computing power loss | Thermal runaway → Full cluster outage |
Selecting the right pump defines the reliability baseline of your entire liquid cooling system. JONSN precision magnetic gear pumps deliver long service life, compact dimensions, and three years of maintenance-free operation, redefining industry standards for core liquid cooling components through structural design advantages.
