An AIO cooler is a sealed box: quiet, simple, and impossible to upgrade. A custom water-cooling loop is the opposite. You choose the pump, the radiators, the tubing, and the exact blocks that sit on your CPU and GPU, and you can drain it, expand it, or rebuild it whenever your hardware changes. In September 2026, with RTX 5090 water blocks running $275 to $425 and full loop parts totaling roughly $600 to $1,000, building one is still a serious investment — even as the wider water-cooling category keeps growing, with DataInsightsMarket sizing the computer water-cooling device market at roughly $2.5 billion in 2025 and IndustryResearchBiz putting liquid cooling’s share of the overall PC cooling market at 38.0% as of October 2025. It’s also one of the few PC upgrades that actually changes how your rig sounds and performs at the same time. This tutorial walks through every step of a real CPU-plus-GPU custom loop build: planning, parts, mounting, filling, leak testing, and the tuning that happens after first boot.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Why Build a Custom Water-Cooling Loop Instead of an AIO
A 360mm all-in-one cooler costs $130 to $220 in 2026 and cools a CPU well enough for most builds. So why do enthusiasts still spend three to five times that on a custom loop? The answer is GPU heat. Modern flagship cards like the RTX 5090 push well past 500W under full load, and stock blower or air coolers on those cards get loud fast. Even AIO makers are chasing that same headroom now: Noctua’s first-ever all-in-one, the NL-LC1, is built on custom-loop engineering, backed by a 6-year warranty, and expected around June 2026, according to Igor’sLAB and Noctua — but it still can’t touch a GPU. A custom loop routes coolant through both the CPU block and a full-cover GPU block in series, pulling heat from two of the hottest components in the case into the same radiator stack. The result is lower load temperatures on both parts, a quieter system overall because fans spin slower across more radiator surface area, and a case that looks like it belongs in a build showcase rather than a parts bin.
There’s a practical case for it too. Once you own the pump, reservoir, radiators, and fittings, adding a new GPU block after your next graphics card upgrade costs $275 to $425, not another $150 to $220 for a whole new AIO. The infrastructure is reusable. That’s the trade you’re making: more money and time up front, in exchange for lower long-term cooling costs, quieter thermals under sustained load, and a system you fully understand because you built every joint yourself. The AIO side of the market isn’t sitting still either: Alphacool’s Core 2 XT, a CPU AIO derived from its own custom water-cooling lineup, launched September 3, 2026, and Fractal Design added the Radius liquid cooler to its Summit and Define families that same month as part of a four-product expansion of its cooling portfolio, per Vortez and Fractal Gaming. Ongoing hardware coverage from outlets like Tom’s Hardware and PC Gamer continues to track this tradeoff every generation, since GPU power draw keeps climbing faster than stock coolers can keep pace quietly.
The realistic delta between a well-built custom loop and a good AIO is usually 5-10°C lower sustained load temperatures on the CPU, and considerably more on the GPU, since most flagship cards ship with air coolers that were never designed to run silently at 500W+. The bigger win for most owners is acoustic, not thermal: five slow-spinning radiator fans across 640mm of total radiator length move the same air as two or three fans on a single AIO radiator, but at a much lower RPM and a noticeably quieter noise floor under sustained gaming load.
Prerequisites: Tools, Skills, and Case Requirements
Before you order a single fitting, confirm you have the right foundation. This build assumes a mid-tower or full-tower case with at least one 360mm radiator mount and one 240mm or 280mm mount, since a CPU-plus-GPU loop needs more radiator surface area than a CPU-only setup. Cases without at least two radiator locations will fight you at every step.
- A case with 360mm top or front mount, plus a second 240/280mm mount
- Phillips-head and Torx screwdrivers, plus a set of hex keys for block mounting hardware
- A tubing cutter (for hard line, a proper acrylic/PETG cutter, not scissors)
- A heat gun, if you’re bending hard tubing
- Digital calipers, for measuring tube runs before you cut
- 2-3 liters of distilled water, plus a coolant additive or premix coolant
- A funnel and a length of scrap tubing for filling
- Paper towels, isopropyl alcohol, and nitrile gloves for cleanup
- Basic comfort disassembling a PC: removing a motherboard, GPU, and case panels without help
Budget four to eight hours of active work for a first CPU-plus-GPU loop, according to build logs and current 2026 tutorials. Hard-line builds with bending and polishing run closer to a full day. Set aside a separate 12 to 24 hours afterward for leak testing before you ever power on the rest of the system. Don’t compress that timeline. Rushing the leak test is the single most common reason first loops end in a flooded motherboard.
Also confirm memory clearance before you start. A 240mm or 280mm radiator mounted at the front or top of a case sits close to the motherboard’s RAM slots, and tall heat-spreader kits can physically collide with the radiator or its fan shroud. Manufacturer spec sheets, including compatibility notes from RAM makers like G.Skill, list module height in millimeters. Measure your case’s actual clearance against that number before buying radiators, not after.
Step 1: Plan Your Loop Layout Before Buying Anything
Every custom loop is a series circuit: coolant leaves the pump, moves through each block and radiator in an order you choose, and returns to the reservoir. Draw the layout before you buy a single part. A standard CPU-plus-GPU order looks like this: reservoir to pump, pump to radiator one, radiator to CPU block, CPU block to GPU block, GPU block to radiator two, and back to the reservoir. Some builders put both radiators back to back near the pump instead, which shortens tubing runs and reduces the number of 90-degree bends.
Loop order affects performance less than most first-time builders expect. Whether the CPU or GPU block comes first only shifts coolant temperature by a degree or two at the second component, because the loop is constantly circulating and radiators are pulling heat out the whole time. What matters far more is total radiator surface area relative to your combined CPU and GPU wattage, and minimizing sharp bends that restrict flow. Sketch your case from the side, mark where the pump/reservoir combo sits (lower front or bottom of the case is standard, since gravity helps keep the reservoir primed), and note where each radiator mounts.
Step 2: Choose Radiators and Fans for Your Case
Radiators are named by fan slot count and size: a 360mm radiator holds three 120mm fans, a 280mm holds two 140mm fans, and so on. 360mm remains the reference size across the whole liquid-cooling market, not just DIY loops — Guru3D reports that CPS’s DX360 Pro ARGB Display, a 360mm all-in-one with two screens built into the pump housing, is scheduled to launch in Japan on September 25, 2026. A water-cooling calculator benchmark from mid-2026 puts real-world heat removal at roughly 8 to 12 watts per degree Celsius of coolant-to-ambient difference, per 120mm radiator slot, at medium fan speed. That number is a planning tool, not a hard spec, since airflow, ambient temperature, and fin density all shift it.
| Radiator Size | Fan Slots | Approx. Heat Capacity (quiet fan speed) | Typical Use Case |
|---|---|---|---|
| 120mm (1×120) | 1 | ~100-150W | Small-form-factor CPU-only loop |
| 240mm (2×120) | 2 | ~200-300W | CPU-only, mid-tower |
| 280mm (2×140) | 2 | ~250-350W | CPU-only or light GPU assist |
| 360mm (3×120) | 3 | ~300-450W | CPU + mid-range GPU |
| 420mm (3×140) | 3 | ~375-525W | CPU + high-end GPU (RTX 5080/5090 class) |
| 480mm (4×120) | 4 | ~400-600W | Dual-radiator builds, extreme overclocks |
For a build cooling an RTX 5090 (450W+ under load) and a modern high-core-count CPU, combined heat load often exceeds 550-650W. That’s why a 360mm plus a 240/280mm combination, rather than a single radiator, is the standard configuration for flagship-tier loops in 2026. Buy fans rated for static pressure, not airflow, since they’re pushing air through dense radiator fins rather than open space. Match fan count to radiator size exactly. A 360mm radiator with only two fans wastes the third slot’s cooling potential.
Step 3: Pick a Pump and Reservoir Combo
Two pump architectures dominate the DIY market: D5 and DDC. D5 pumps typically spec around 1,000 to 1,500 liters per hour at max speed, and cost $90 to $150 as standalone units or pump/reservoir combos. DDC pumps are more compact, generally rated 700 to 1,000 liters per hour with higher head pressure, and run $70 to $120. For a two-radiator, two-block loop with a handful of 90-degree bends, a D5 is the safer default: it moves more volume at lower RPM, which means less pump whine and more headroom if you add a third radiator later.
Buy a pump/reservoir combo unit rather than sourcing them separately for your first build. Combo units bolt the pump directly to the bottom of the reservoir, which eliminates one length of tubing, one pair of fittings, and one potential leak point. Mount it low in the case, ideally below the highest point in the loop, so gravity keeps the pump inlet flooded and you avoid pump cavitation (a rattling, grinding sound that happens when air gets sucked into the impeller). Component makers such as Arctic, better known for CPU coolers and case fans, also publish flow-rate and noise-level spec sheets worth cross-referencing if you’re unsure whether a given pump can push coolant through a dense, two-radiator loop without excessive noise at higher RPM.
Step 4: Select CPU and GPU Water Blocks
Block selection is where compatibility actually matters, and where beginners lose the most time. A CPU block has to match your socket (AM5 or LGA1851/1700, depending on platform), and a GPU block has to match your exact card’s PCB, not just its GPU chip, because the block’s coolant channels line up with the specific VRM and memory layout of that board. A block designed for a Founders Edition RTX 5090 will not bolt onto an ASUS ROG Astral RTX 5090, because the two cards use different PCBs.
Current retail pricing for RTX 5090 full-cover blocks runs from roughly $275 to $425 depending on brand and finish, according to a May 2026 buyer’s guide. Specific examples: the EK-Quantum Vector³ Suprim RTX 5090 (nickel/plexi) lists at $344.99, and the Corsair Hydro X Series XG5 RGB 5090 Astral block lists at $389.99 direct from Corsair. Budget-tier options from Alphacool and Barrow show up closer to $250 to $300 on marketplaces like Newegg. Confirm your card’s exact model number against the block manufacturer’s compatibility list before ordering. There’s no universal fit here, and returns on used thermal-paste-covered blocks are rarely accepted.
Step 5: Decide Between Soft Tubing and Hard Line
This is the single biggest decision that shapes how hard your build will be. Soft tubing (PVC or EPDM) flexes by hand, forgives measurement mistakes, and lets you re-route a run in seconds if something doesn’t line up. Hard tubing (PETG or acrylic) has to be cut to an exact length, heated and bent with a heat gun for every angle, and replaced entirely if a bend goes wrong. The payoff is a cleaner, sharper look with perfectly straight runs and crisp corners.
| Factor | Soft Tubing (PVC/EPDM) | Hard Tubing (PETG/Acrylic) |
|---|---|---|
| Common sizes | 3/8″ ID x 5/8″ OD, or 1/2″ ID x 3/4″ OD | 10mm, 12mm, or 14mm OD rigid tube |
| Bending method | By hand, no tools needed | Heat gun + silicone bending insert |
| Mistake tolerance | High, reroute and retry freely | Low, a bad bend means a wasted tube |
| Added build time | Roughly 1-2 hours | Roughly 3-5 hours, plus practice bends |
| Cost per tube | Lower, sold by the foot | Higher, plus wasted practice material |
| Best for | First-time builders, budget builds | Show builds, second or third loop attempts |
Why most first loops should use soft tubing
Hard-line builds look sharper, but the learning curve is steep enough that a large share of first attempts end with a cracked tube or a fitting that won’t seat because the cut end isn’t perfectly square. Soft tubing lets you focus your first build on the part that actually matters for safety: leak-free fitting connections. Once you’ve done one successful soft-tube loop and understand how your case’s clearances work, a hard-line rebuild is a much lower-risk second project.
Step 6: Calculate Tubing Length and Fitting Count
Measure every run in your sketch from Step 1 using a piece of string or your calipers along the actual case interior, not straight-line distances on paper. Add roughly 10% extra to each measured length to account for the tube insertion depth on each fitting (typically 15-20mm per end). Every connection point, whether it’s a block port, radiator port, or 90-degree adapter, needs one compression fitting for soft tubing or one fitting pair for hard line.
Tubing length worked example (soft tubing, 1/2" ID x 3/4" OD):
Run 1: Reservoir -> Pump inlet = 4 in (fixed, combo unit)
Run 2: Pump -> Radiator 1 (360mm, top) = 11 in
Run 3: Radiator 1 -> CPU block = 9 in
Run 4: CPU block -> GPU block = 6 in
Run 5: GPU block -> Radiator 2 (280mm) = 10 in
Run 6: Radiator 2 -> Reservoir = 8 in
--------------------------------------------------
Subtotal = 48 in
Insertion depth buffer (10%) = ~5 in
--------------------------------------------------
TOTAL TUBING NEEDED = ~53 in (round up to 5 ft)
Fittings needed: 12x straight compression fittings (G1/4")
2x 90-degree rotary adapters (tight case clearance)
1x drain valve fitting (bottom of loop, lowest point)Almost every block, radiator, pump, and reservoir on the market in 2026 uses the G1/4″ BSPP thread standard, which is why fittings from EK, Alphacool, Barrow, and Bykski are interchangeable across brands. Buy two or three extra straight fittings beyond your count. Cross-threading one during install is common enough that spares save a delayed build.
Step 7: Mount Radiators, Pump/Reservoir, and Fans in the Case
Remove the motherboard tray’s side panel access and, if your case allows it, pull the motherboard out entirely for this stage. It’s far easier to mount hardware in an empty case than to work around installed components. Mount fans to each radiator first, then mount the fan-and-radiator assembly to the case as one unit, orienting fans to intake or exhaust based on your case’s overall airflow direction (top and front radiators are commonly set to intake fresh air through the fins, though case-specific airflow modeling can change that call).
Install the pump/reservoir combo using its included mounting bracket, positioned low in the case and, where possible, below the CPU block so the pump primes easily during fill. Route power cables (fan headers, pump SATA or 4-pin power, and RGB headers if applicable) now, before tubing blocks access to the connectors. Leave the motherboard out of the case for the next step.
Step 8: Install the CPU Block and GPU Block
With the motherboard on a flat surface outside the case, clean the CPU integrated heat spreader with isopropyl alcohol, apply a pea-sized dot of thermal paste, and mount the CPU block using its socket-specific backplate and standoffs, tightening the four corner screws in a diagonal star pattern to even pressure. Reinstall the motherboard into the case once the block is seated.
For the GPU, remove the card’s stock cooler shroud entirely, which exposes the bare PCB and voids the manufacturer warranty on air-cooling hardware (this is expected and standard for custom loop builds). Clean the GPU die and every VRM/memory pad location, apply thermal paste to the die and fresh thermal pads to the VRM and memory contact points (stock pads are rarely reusable once removed), then mount the water block using the screw pattern specified in that block’s manual. GPU blocks typically require more torque precision than CPU blocks since they’re covering a much larger, uneven surface with dozens of contact points. Install the GPU into its PCIe slot only after the block is fully seated and inspected for gaps.
If you’ve never fully disassembled a graphics card before, spend ten minutes looking at a teardown guide for your specific model first. Repair reference sites like iFixit document screw counts, hidden fasteners under warranty stickers, and ribbon-cable connectors that are easy to tear if you pull the shroud at the wrong angle. Photograph each stage as you go. A GPU has dozens of small screws in slightly different lengths, and reassembly (on the rare loop where you later swap the block back out) goes much faster with a visual reference.
Step 9: Cut and Route Tubing, or Bend Hard Line
For soft tubing, cut each length from your Step 6 measurements using a dedicated tubing cutter for a clean, square edge, since a ragged cut won’t seal against the compression fitting’s O-ring. Push each tube end fully onto its fitting until it seats against the shoulder, then hand-tighten the compression collar. For hard line, cut PETG or acrylic tube slightly long, heat the bend section evenly with a heat gun while rotating the tube, insert a silicone bending stick if you have one to prevent kinking, and bend in one smooth motion. Expect to waste two or three practice tubes before your first clean 90-degree bend.
Route every run so it clears fan blades, RAM heat spreaders, and case cable channels. Test-fit each tube dry, without fittings tightened, before committing to a final cut. A run that looks fine on paper often clips a RAM stick or a case standoff once it’s physically in the loop.
Step 10: Connect Fittings and Torque-Check Every Joint
Go around the entire loop a second time and hand-tighten every compression fitting collar until it stops turning, then give it a final quarter-turn with light finger pressure, not a wrench. Overtightening compression fittings cracks the plastic collar or deforms the O-ring, which causes the exact leak you’re trying to prevent. Confirm every port on every block, radiator, and the pump/reservoir has a fitting or a plug screwed in. An open, unused port is the most common cause of a first-fill flood, because it’s easy to simply forget one exists.
Install a drain valve fitting at the lowest point in your loop now, even if you don’t think you’ll need it soon. Draining a loop without one means disconnecting tubing over a towel and hoping for the best, which gets messy fast on a maintenance cycle six months from now.
Step 11: Fill the Loop and Bleed Trapped Air
Before filling, disconnect your motherboard’s 24-pin and CPU power connectors, or use a PSU jumper/paperclip trick to power only the pump without booting the rest of the system (some pump/reservoir combos and PSUs support this directly through a dedicated fill mode; check your specific hardware). You want the pump spinning during fill without the CPU, GPU, or motherboard powered on.
Pour distilled water mixed with your chosen additive or premix coolant into the reservoir through a funnel, topping off as the level drops while air moves through the loop. Pulse the pump on for 5-10 seconds at a time using the PSU switch, then off, refilling the reservoir each time, repeating until coolant flows visibly through every tube section with no large air pockets. Gently rock or tilt the case (with the pump off) to help trapped air bubbles work their way toward the reservoir. Full priming typically takes 15-30 minutes of this on-off cycling before the loop runs quiet and steady.
Step 12: Run a 24-Hour Leak Test With the Rest of the System Powered Off
This step is not optional, and skipping or shortening it is the single most expensive mistake in custom loop building. Line the bottom of the case, and the area under every fitting and block, with paper towels. With the pump still running on its isolated power circuit and the motherboard, CPU, and GPU fully disconnected from power, let the system run for a minimum of 12 hours, with 24 hours recommended for a first build, per current 2026 build guides. Check every fitting every 30-60 minutes for the first few hours, then every few hours after that. Any damp spot on a paper towel means a leak at that joint, not a system-wide failure. Drain, retighten or reseat that specific fitting, and restart the test from the beginning.
Do not reconnect motherboard, CPU, or GPU power until the leak test finishes clean for its full duration. Water and live 24-pin motherboard power are the two things a leak test exists to keep apart, and a single missed drip on a connector can turn a $150 tubing mistake into a $2,000+ motherboard-and-GPU replacement.
Step 13: First Boot, Fan Curves, and Pump Speed Tuning
Once the leak test passes clean, reconnect all power, close up the case, and boot into BIOS first rather than straight into the OS. Confirm CPU temperature readings look sane at idle (typically 30-40°C) before doing anything else. If the reading looks wrong or absent, the CPU block may not be seated correctly, and you should shut down and recheck it before proceeding.
In your motherboard’s fan control panel or a utility like Corsair iCUE or Aquacomputer Aquasuite, set radiator fans to a curve tied to coolant or CPU temperature rather than a flat speed, and set the pump to a fixed moderate speed (roughly 60-70% for a D5) rather than temperature-controlled, since pumps should run consistently to maintain steady flow regardless of thermal load.
Sample fan curve (radiator fans, temp-based, set in BIOS or iCUE/Aquasuite):
Coolant Temp (C) Fan Duty (%)
------------------ -------------
30 30
35 40
40 55
45 70
50 85
55+ 100
Pump speed: fixed at 65% (D5) -- do not tie pump RPM to temperatureRun a stress test (Cinebench for CPU, a GPU-bound game or a synthetic GPU load for the graphics card) for 20-30 minutes while watching temps in HWiNFO64 or your monitoring tool of choice. A healthy loop should stabilize well below the thermal throttle point for both CPU and GPU. If temps climb steadily instead of plateauing, revisit your radiator sizing from Step 2 or check for an airlock still trapped somewhere in the loop.
Common Pitfalls That Ruin a First Custom Loop
- Buying a GPU block for the wrong PCB revision. Matching the GPU chip isn’t enough. Confirm the exact card model against the manufacturer’s compatibility chart, since a Founders Edition and an AIB card rarely share a block.
- Skipping or shortening the leak test. A slow drip can take 6-8 hours to become visible. A 2-hour test misses exactly the failures that matter most.
- Overtightening compression fittings. Cracked collars and squeezed O-rings cause more leaks than they prevent. Hand-tight plus a light quarter-turn is the ceiling, not a starting point.
- Leaving an unused port unplugged. Every reservoir, pump top, and radiator has spare ports for future expansion. Forgetting to cap one is an easy, common mistake during a first build.
- Mounting the pump above the highest point in the loop. This starves the pump inlet and causes cavitation noise, plus a harder priming process during fill.
- Ignoring galvanic corrosion risk by mixing metals. Nickel-plated and bare copper blocks in the same loop, combined with plain tap water instead of distilled water and a corrosion inhibitor, can cause metal degradation over months.
- Cutting hard-line tubing without a dry test-fit first. A bend that’s off by even a few millimeters won’t seat against the next fitting, wasting an entire tube.
Troubleshooting Custom Water-Cooling Problems
Even a carefully built loop runs into issues in its first weeks. Here’s how to diagnose the most common ones.
| Symptom | Likely Cause | Fix |
|---|---|---|
| Gurgling or rattling noise | Air trapped near the pump inlet (cavitation) | Tilt the case gently, run pump longer to work bubbles toward the reservoir |
| Damp spot under a fitting | Under-tightened compression fitting or bad O-ring seat | Drain that section, reseat tube fully, retighten to hand-tight plus a quarter-turn |
| CPU temps normal, GPU temps high | GPU block not fully seated, or thermal pads too thin/thick | Reseat block, replace VRM/memory pads with correct thickness for your card |
| Coolant level drops slowly over weeks with no visible leak | Evaporation through tubing walls (normal, especially with soft tubing) | Top off reservoir monthly with distilled water, no additive needed for top-offs |
| Discolored or cloudy coolant after a few months | Algae growth, or reaction between coolant and metal/tubing plasticizer | Drain, flush with distilled water, refill with fresh coolant plus biocide |
| Pump seems to run but flow indicator shows nothing | Airlock somewhere in the loop, often at the highest point | Power-cycle the pump repeatedly, tilt case to move the air pocket toward reservoir |
| Fittings show white/green residue after months | Galvanic corrosion from mixed metals or hard water minerals | Drain and inspect, replace corroded fitting, switch to distilled water plus inhibitor |
| System won’t boot after reassembly | Loose RAM/GPU seating disturbed during block install, or a power cable left disconnected | Reseat RAM and GPU, retrace every power cable back to the PSU |
Advanced Tips: Monitoring, Coolant Choice, and Maintenance Schedule
Automated temperature logging
Once your loop is stable, a few upgrades make ownership easier. First, log temperatures automatically instead of eyeballing HWiNFO64 during stress tests. A lightweight PowerShell script pulling from LibreHardwareMonitor’s WMI interface can record coolant and component temps to a CSV file for later review, which is especially useful during your first 24-hour leak test and again after any maintenance drain.
# PowerShell: log CPU/coolant temps every 60s via LibreHardwareMonitor WMI
# Requires LibreHardwareMonitor running with WMI provider enabled
$logFile = "C:\LoopLogs\leaktest_$(Get-Date -Format 'yyyyMMdd_HHmm').csv"
"Timestamp,SensorName,Value" | Out-File $logFile
while ($true) {
$sensors = Get-WmiObject -Namespace "root\LibreHardwareMonitor" -Class Sensor |
Where-Object { $_.SensorType -eq "Temperature" }
foreach ($s in $sensors) {
"$(Get-Date -Format o),$($s.Name),$($s.Value)" | Out-File $logFile -Append
}
Start-Sleep -Seconds 60
}For builders who want a permanent dashboard rather than a one-off log, a small self-hosted Grafana and InfluxDB stack can graph coolant temperature, pump RPM, and fan duty in real time on a second monitor or phone. This is overkill for a single loop, but genuinely useful if you’re running a loop cooling a home server or an always-on workstation.
# docker-compose.yml -- minimal InfluxDB + Grafana stack for loop monitoring
version: "3.8"
services:
influxdb:
image: influxdb:2.7
ports: ["8086:8086"]
volumes: ["./influxdb-data:/var/lib/influxdb2"]
grafana:
image: grafana/grafana:latest
ports: ["3000:3000"]
depends_on: [influxdb]
volumes: ["./grafana-data:/var/lib/grafana"]Linux builders can get the same visibility without WMI, using lm-sensors and a simple loop written straight into a cron-scheduled shell script. This is a smaller footprint than a full Grafana stack and works well for a quick leak-test log on a SteamOS or Linux desktop build.
#!/bin/bash
# leak_test_log.sh -- log coolant/CPU temps every 60s during a leak test
# Requires: lm-sensors (sudo sensors-detect), and a coolant temp probe
# exposed via a compatible motherboard header or USB sensor
LOGFILE="$HOME/loop_logs/leaktest_$(date +%Y%m%d_%H%M).csv"
mkdir -p "$HOME/loop_logs"
echo "timestamp,sensor,value_c" > "$LOGFILE"
while true; do
sensors -u | awk -v ts="$(date -Iseconds)" '
/_input/ { gsub("_input",""); print ts","$1","$2 }' >> "$LOGFILE"
sleep 60
doneCoolant choice and refresh schedule
On coolant, premixed options from EK (CryoFuel), Corsair (Hydro X), and Alphacool (Eiswasser) run roughly $10 to $20 per liter, and a CPU-plus-GPU loop with two radiators typically needs 1 to 2 liters total. Distilled water plus a separate biocide and corrosion inhibitor costs less but requires more careful mixing. Either approach works long-term; premix is simpler for a first build, distilled-water-plus-additive is cheaper if you’re maintaining multiple loops. Plan a full drain, flush, and refill every 6 to 12 months regardless of which coolant you choose, since even clean loops accumulate microscopic buildup over time that gradually reduces flow.
Loop Kits vs Building From Individual Parts
Every major brand in this space, including EK, Alphacool, and Corsair, sells starter kits that bundle a pump/reservoir combo, one or two radiators, fittings, and tubing into a single box for roughly $250 to $400. That’s cheaper than buying each part separately, and it removes the compatibility guesswork from Steps 2 and 3, since the manufacturer has already confirmed the pump can push coolant through that specific radiator combination. For a first CPU-only loop, a kit is the lower-risk choice.
The tradeoff shows up once a GPU block enters the picture. Kits are sized for a single radiator and a CPU block, so adding a 450W+ GPU on top usually means buying a second radiator and extra tubing anyway, at which point you’ve paid the kit premium and then paid full price for the added-on parts. If you already know you want CPU-plus-GPU cooling from day one, buying radiators, a pump/reservoir combo, and fittings individually usually costs less overall, even though it puts more compatibility research on you upfront. Check the pump’s rated flow against the combined restriction of every block and radiator in your plan (each manufacturer publishes a flow-restriction curve), since an underpowered pump paired with two dense GPU and CPU blocks in series is a common cause of disappointing temperatures on a first individually-sourced build.
Complete Parts List: A Working 360mm + 280mm CPU/GPU Loop
Here’s a realistic full parts list for the exact build described in this tutorial, cooling a modern high-core CPU and an RTX 5090-class GPU, based on September 2026 pricing.
| Component | Example Product | Approx. Price (Sept. 2026) |
|---|---|---|
| GPU water block | EK-Quantum Vector³ Suprim RTX 5090 (nickel/plexi) | $344.99 |
| CPU water block | Any AM5/LGA1851 nickel block from EK, Alphacool, or Corsair | $70-120 |
| Pump/reservoir combo | D5-based combo unit, 1,000-1,500 LPH | $100-150 |
| Radiator 1 | 360mm (3×120), high fin-density | $80-120 |
| Radiator 2 | 280mm (2×140) | $80-120 |
| Fans (5 total) | Static-pressure rated, PWM | $70-100 |
| Soft tubing (5 ft) | 1/2″ ID x 3/4″ OD PVC/EPDM | $15-25 |
| Fittings (14-16 pieces) | G1/4″ compression + rotary adapters | $60-90 |
| Coolant (1.5L) | Premix or distilled water + additive | $15-30 |
| Estimated total | ~$835-1,000 |
That lands inside the roughly $600 to $1,000 range typical of a full CPU-plus-GPU custom loop in 2026, versus $130 to $220 for a high-end 360mm AIO that only covers the CPU. The gap is real, and it’s the reason custom loops remain a project for builders who want both the performance and the process, not just the lowest possible cooling cost.
Frequently Asked Questions
Is a custom water-cooling loop actually worth it over an AIO in 2026?
If you only need CPU cooling, a 360mm AIO at $130-220 covers most builds without the risk or labor of a custom loop. A custom loop earns its cost when you’re also cooling a 450W+ flagship GPU and want quieter sustained performance across both components, or when you plan to reuse the pump, radiators, and reservoir across future GPU upgrades.
How much does a full custom loop cost?
A CPU-plus-GPU loop with two radiators typically totals $600 to $1,000 in September 2026, with the GPU block alone accounting for $275-425 of that on flagship cards like the RTX 5090.
How long does the leak test need to run?
A minimum of 12 hours, with 24 hours recommended for a first build. Check every fitting periodically throughout, since slow leaks can take several hours to show up on a paper towel liner.
Can I mix nickel and bare copper blocks in the same loop?
It’s not recommended long-term. Mixing metals in the presence of tap water or a low-quality coolant accelerates galvanic corrosion. Stick to one finish type across your blocks, radiators, and fittings, and always use distilled water with a corrosion inhibitor or a manufacturer premix.
Should a first-timer use soft tubing or hard line?
Soft tubing. It tolerates measurement mistakes, requires no heat gun, and lets you focus your first build on leak-free fittings rather than perfect bends. Hard line is a strong second-build project once you understand your case’s clearances.
How often does a custom loop need maintenance?
Plan a full drain, flush, and refill every 6 to 12 months. Top off evaporated coolant with plain distilled water monthly between full maintenance cycles; no additive is needed for small top-offs.
What happens if I find a small leak during the test?
Drain that section of the loop, dry the area completely, reseat the tube fully into the fitting, and retighten to hand-tight plus a light quarter-turn. Restart the full leak test duration from zero rather than assuming the fix worked after a short recheck.
Do I need to cool both the CPU and GPU, or can I do just one?
A CPU-only loop is a valid, simpler starting project and needs only one radiator in most cases. Adding the GPU later just means draining the loop, installing the GPU block in the existing tubing path, and rerunning the leak test. Many builders start CPU-only and add GPU cooling on their second pass once they’re comfortable with the process.
Related Coverage
- DualSense Edge Stick Module Swap: 12 Steps, $19.99 [2026]
- How to Test RAM for Gaming Crashes: MemTest86 in 12 Steps [2026]
- How to Set Up 4K 120Hz VRR on PS5, Xbox, Switch 2: 12 Steps [2026]
- How to Overclock a GPU: RTX 5080/5090 & RX 9070 XT in 12 Steps [2026]
- Fix Stick Drift on PS5, Xbox, Switch, Deck: 12 Steps [2026]
