Custom Water-Cooling Loop Order: Does Radiator Placement Matter?

THERMAL BLUEPRINT / CUSTOM LOOP FIELD GUIDE


Custom Water-Cooling Loop


Order:


Does Radiator Placement


Matter?


CMS Field


Recommended Metadata


SEO Meta Title:


Does Water-Cooling Loop Order Matter? Radiators & Fans


Meta Description:


Learn how custom water-cooling loop order, radiator placement and push/pull


fans affect coolant temperatures, airflow, noise and build cost.


URL Slug:


custom-water-cooling-loop-order


Primary Keyword:


custom water-cooling loop order


Secondary Keywords:


radiator placement, PC water cooling, push/pull fans, custom loop cooling, RTX


5090 water cooling


Category:


PC Cooling


Subcategory:


Custom Water-Cooling


Content Type:


Technical Guide


Where


radiator placement


helps, when loop order barely matters, and how to choose fans


without wasting money.


•


Updated 22 October 2026


•


RTX 5090 + Ryzen 9 9950X3D


•


Research-based; no lab test of this exact


build


575 W


RTX 5090 total graphics


power


170 W


Ryzen 9 9950X3D default


TDP


2 × 360


Example radiator layout, in


millimetres


6 → 12?


Start with six fans; justify the


second bank


Quick answer


Heat budget


Loop route


Coolant math


Radiators


Fans


Setup


Prices


Verdict


Links

01 / HEAT BUDGET


Start with the ratings, then measure the real load


COMPONENT


VERIFIED REFERENCE POINTS


WHAT THEY MEAN FOR THE LOOP


NVIDIA GeForce RTX


5090


575 W total graphics power; 32 GB


GDDR7; available 30 January 2025; launch


starting price $1,999.


The GPU dominates the nameplate heat budget.


Actual power varies with the workload, power


limit and add-in-card model.


AMD Ryzen 9 9950X3D


170 W default TDP; 16 cores / 32 threads;


up to 5.7 GHz; launched 12 March 2025;


$699 launch MSRP.


AMD recommends liquid cooling for optimal


performance and does not include a boxed cooler.


“Up to 5.7 GHz” is not an all-core guarantee.


The two headline figures total 745 W, but this is a rough planning reference, not a measured heat load.


GPU total graphics power (TGP) and CPU default thermal design power (TDP) are different ratings;


neither tells you exactly how much heat enters the coolant under a particular workload. Size and tune the


loop around your actual CPU+GPU workload, acceptable coolant temperature and noise target.


Practical priority:


adequate radiator area and uncontaminated airflow come before elaborate component


ordering. A well-fed pump and a loop you can drain safely matter more than chasing a theoretical last degree.


02 / THE ROUTE MAP


What the inter-component radiator layout does


In this example, each water block has a radiator immediately upstream. The top radiator therefore rejects


heat between the GPU and CPU instead of leaving the two blocks back-to-back.


01


Reservoir


feeds pump


→


02


Bottom 360


mm radiator


→


03


GPU water


block


→


04


Top 360 mm


radiator


→


05


CPU water


block


→


06


Return to


reservoir


The XD5 example is a pump/reservoir combo. With separate parts, connect the reservoir outlet directly to the pump inlet so the pump


remains supplied with coolant.


Thermal role


The middle radiator can reduce the


local coolant rise carried directly


from the GPU outlet to the CPU


inlet, with the largest potential effect


at low flow.


Routing role


GPU → top radiator → CPU may


produce simple vertical runs in some


cases. Port positions decide whether


it actually saves bends or fittings.


Fixed workload


Both radiators still have to reject the


combined heat. Moving one between


the blocks changes the sequence, not


the heat generated.


This is a sound option, not a universal “best” order. A CPU-first layout can also prioritise cooler coolant


for the processor.


EK’s loop-order guidance


likewise emphasises a reservoir-fed pump and otherwise


flexible component ordering.


03 / PHYSICS, NOT MYTH


Estimate coolant rise from heat and flow


For a steady heat load, the rate of heat transfer into the coolant follows


P = ṁc


p


ΔT


. Rearranged, the


temperature rise across a block is proportional to heat power and inversely proportional to mass flow.


ΔT = P ÷ (ṁ × c


p


) • c


p


≈ 4,184 J/(kg·°C) for water near room temperature


The table below uses a simplified assumption: 575 W enters water-like coolant, with 1 litre treated as


approximately 1 kg. It estimates the temperature difference between GPU-block inlet and outlet—not the


GPU core temperature.


APPROX.


FLOW


FLOW IN


L/H


CALCULATED COOLANT


RISE AT 575 W


INTERPRETATION


0.30 GPM


68 L/h


7.3°C


Low flow can create a substantial local rise.


0.60 GPM


136 L/h


3.6°C


Doubling flow roughly halves the idealised rise.


0.75 GPM


170 L/h


2.9°C


A mid-flow illustration, not a universal pump-PWM


setting.


1.00 GPM


227 L/h


2.2°C


Higher flow reduces, but does not erase, local differences.


These are calculations, not RTX 5090 benchmarks.


Actual board power, the fraction of heat entering the


coolant, coolant properties and measurement uncertainty will change the result. They do not prove a matching


drop in CPU core temperature.


For real-world context,


Corsair’s RTX 4090 demonstration


reported roughly a 5°C coolant rise at 60 L/h


and 2°C at 170 L/h. That is useful evidence of flow dependence, but it is not a test of the RTX 5090 or this


dual-radiator route.


What a middle radiator can do


Lower the coolant temperature entering the downstream


block relative to a direct GPU → CPU connection,


assuming the radiator is rejecting heat effectively.


What it cannot promise


A universal 5°C CPU improvement, a fixed boost clock,


or identical results in every case. Those claims require


testing the complete build.


Do not translate pump PWM directly into GPM.


A D5 at 30% does not deliver one universal flow rate:


blocks, radiators, tubing and fittings alter hydraulic restriction. If the buying decision depends on low-flow


behaviour, use a flow meter rather than a PWM-to-GPM guess.


04 / RADIATOR PLAN


Buy heat-rejection capacity before extra fan count


Two 360 mm radiators provide more heat-exchange area than one, but there is no universal watt rating for


a “360”. Heat rejection depends on coolant-to-air temperature difference, fin design, airflow, fan speed,


ambient temperature and recirculation inside the case. A radiator that is adequate at one noise level may be


inadequate at another.


360 MM EXAMPLE


THICKN


ESS


NOMINAL


PUSH/PULL


STACK*


VERIFIED


CONSTRUCTION


BEST-FIT USE


Corsair XR5 360


NEO


30 mm


80 mm


Copper core; 45-micron


fins


Space-conscious builds; begin single-sided


unless clearance and testing favour


push/pull.


Corsair XR7 360 V2


55 mm


105 mm


Copper/brass; 25-micron


fins; fan mounts on both


sides


Larger cases where the extra depth and


fittings clear nearby hardware.


EK-Quantum


Surface S360


30 mm


80 mm


20-FPI copper fins; sixteen


2 mm copper tubes


Single-sided slim builds. EK says


push/pull does not bring a proportional


performance increase on its S Series.


*Radiator + two 25 mm fans only. Brackets, port fittings, tube bends, cables and screw heads need additional space. An 80 mm


or 105 mm gap alone is not a compatibility guarantee.


Default airflow: bottom intake, top exhaust


The bottom radiator receives room air and the top radiator


exhausts. This is a sensible starting point for many


showcase cases, although the top radiator may see


warmer internal air.


Optional: both radiators as intake


Both can receive room air, but the case then needs a clear


exhaust path. Without one, warm air can recirculate and


raise motherboard, memory and internal air temperatures.


Choose between those arrangements by measuring coolant, component and internal air temperatures.


Neither is automatically superior in every chassis or climate.


05 / FAN LAB


Push/pull is an adjustment; reverse blades are a visibility


choice


Will twelve fans double airflow?


No. A second fan bank does not double CFM, and it does


not guarantee lower noise. It can improve radiator


performance, but the gain depends on the radiator, speeds


and operating point.


In Corsair’s 500 W RTX 4090 test, adding push/pull at


1,000 RPM reduced coolant temperature from 44°C to


43°C. That single test is not a universal prediction; it is a


useful warning against dramatic promises.


Why use reverse-blade fans?


Reverse fans invert airflow relative to a standard design,


allowing the more attractive rotor/lighting face to remain


visible in locations where a standard intake would show


its rear struts.


That can clean up a bottom intake or a dual-sided radiator


display. It is not, by itself, a thermal upgrade. Corsair’s


orientation guide explains the visual purpose.


Verified fan specifications


FAN


SIZE


SPEED


MAX


AIRFLOW


MAX STATIC


PRESSURE


MANUFACTURE


R SOUND RANGE


iCUE LINK LX120-R RGB


(reverse)


120 × 25 mm


400–2,400 RPM


66.7 CFM


4.72 mm-H₂O


10–37.7 dBA


iCUE LINK LX120 RGB


(standard)


120 × 25 mm


400–2,400 RPM


69.9 CFM


5.22 mm-H₂O


10–36 dBA


These are manufacturer specifications for individual fans, not measured noise from a twelve-fan PC. System acoustics also


depend on fan curves, turbulence, case panels and the listener’s environment.


A clean-face push/pull layout


RADIATOR


AIR


DIRECTIO


N


OUTER / HIDDEN BANK


INTERIOR / VISIBLE BANK


Bottom 360


mm


Intake,


upward


3 standard fans below the radiator, pushing


up


3 reverse fans above it, pulling up


Top 360 mm


Exhaust,


upward


3 reverse fans above the radiator, pulling air


out through the roof


3 standard fans below it, pushing air up and


out


Both banks in each row move air in the same direction while keeping the clean rotor face toward the


intended viewing side. Confirm the arrows on the physical fan frames, the top-panel sightlines and


connector clearance before mounting.


Judge push/pull at equal noise and equal workload—not merely at equal RPM. Twelve


motors are not automatically quieter than six.


06 / CONTROL & ASSEMBLY


Use coolant as the control signal, then build for


serviceability


The


Corsair iCUE LINK XD5 RGB ELITE


combines a Xylem D5 PWM pump, a 440 ml reservoir and an


integrated coolant-temperature sensor; an iCUE LINK System Hub is required. Where supported by your


control setup, coolant temperature is a sensible input for radiator fans because liquid warms more slowly


than a boosting CPU. It avoids turning every brief core-temperature spike into a fan-speed change.


iCUE LINK can reduce individual fan PWM and RGB cable runs through daisy-chaining, but “single


cable” does not eliminate power or channel limits. The iCUE LINK System Hub supports up to 24


compatible devices across two channels, with up to 12 devices per channel. Confirm the planned fan-and-


pump allocation, compatibility and power requirements before assembly.


✓


Match the blocks exactly.


Confirm AM5 support for the CPU block and the precise graphics-card PCB—not just


“RTX 5090”—for the GPU block.


✓


Measure the full stack.


Include radiator, fans, brackets, fittings, tubing and cable bends in the clearance drawing.


✓


Count fittings from the final sketch.


Each straight tube run normally needs two end fittings; rotary elbows, drain


hardware and fill ports are additional decisions.


✓


Plan filling and draining.


A low-point drain can help, but no fixed percentage of coolant is guaranteed to drain


without tilting or opening the loop.


✓


Bleed and leak-test methodically.


Follow the component manufacturers’ instructions and keep the rest of the PC


unpowered during the initial leak test.


✓


Validate under your real workload.


Log coolant temperature, CPU/GPU behaviour and fan speeds. Add a flow


reading if loop-order decisions rely on low-flow operation.


07 / PRICE REALITY


Separate launch prices from today’s shopping basket


Hardware prices move with stock, promotions, taxes and region. These reference points were checked on


22 October 2026; they are not quotes for a future purchase.


ITEM


REFERENCE PRICE


HOW TO USE IT


RTX 5090


$1,999 US launch starting price; NVIDIA’s India


product page displayed “Starting at Rs.


2,39,000”.


Partner cards and water-block-compatible models


can cost differently. Check the exact card and


current seller.


Ryzen 9 9950X3D


$699 US launch MSRP.


This is historical context, not a current US or India


retail price.


Optional second


six-fan bank


One LX120 and one LX120-R triple kit totalled


$188.98 at checked US sale prices, or $279.98 at


the displayed regular prices, before tax.


This is the fan cost alone. Expansion packs,


fittings, shipping and Indian import costs can


change the total.


Use the live US and India links below to price the


complete


loop. The central cooling decision is not


whether twelve fans are affordable in isolation; it is whether their measured benefit justifies their cost after


radiator capacity and case fit are already sorted.


08 / BUILD VERDICT


Our recommendation: route with intent, add fans with


evidence


Recommended starting point


For a high-heat showcase PC with


room for two 360 mm radiators, start


with one bank of six suitable fans,


bottom intake/top exhaust, and a


reservoir-fed D5 with coolant-based


control. Use the inter-component


route if it improves tube geometry or


gives a measured benefit in your


low-flow setup.


Add six more fans when…


You have verified the full 80–105


mm-plus stack clearance, want the


dual-sided visual treatment, and


either accept the cost for aesthetics or


testing shows a worthwhile benefit at


your noise target.


Skip this configuration


when…


The case is tight, the actual heat load


is modest, or the purchase depends


on a guaranteed 5°C CPU saving. A


simpler sequential route with single-


sided fans is easier to service and


usually the better value.


Bottom line:


a radiator between the GPU and CPU is a legitimate routing choice, not a thermal cheat


code. Spend first on radiator area, airflow and a controllable pump; treat push/pull and reverse-blade


symmetry as optional refinements.


09 / QUICK ANSWERS


Frequently asked questions


Should the CPU or GPU come first?


There is no universal winner. At low flow, putting the CPU directly after a radiator can prioritise lower coolant


temperature for the processor. A GPU-first route with another radiator between the blocks can also supply cooled


coolant to the CPU. Keep the reservoir before the pump, then choose the route that fits and services best.


Does a radiator between the GPU and CPU lower temperatures?


It can lower the coolant temperature entering the downstream CPU compared with a direct GPU-to-CPU connection.


The effect depends on flow and radiator performance. It does not guarantee the same reduction in CPU core


temperature, and the total heat load remains unchanged.


Is push/pull quieter than six fans?


Not automatically. Push/pull may deliver slightly better cooling at the same RPM, but it adds motors and possible


turbulence. Compare configurations at equal noise and workload to find out whether lower RPM offsets those extra


noise sources.


Do reverse-blade fans improve cooling?


“Reverse” describes airflow direction and viewing orientation, not a performance tier. Compare airflow, static


pressure, noise and price for the specific standard and reverse models.


Is there one ideal coolant-temperature fan curve?


No. Ambient temperature, heat load, radiator area and noise preference differ. Use coolant temperature as the control


input where supported, then tune the curve against stable CPU+GPU workloads rather than copying a universal 28–


40°C preset.