AIO vs Custom Water: Showdown at 280mm

Joined
Apr 9, 2021
Messages
943
In a previous thread, I was wondering exactly what the difference in performance is between an AIO and a custom loop if the radiator is the same size and the same fans are used. I said I would build a custom loop just to answer that question, and that's what I did. I pulled the AIO from my favorite test rig and replaced it with a custom loop using top-shelf components. I spent some time trying different mounting pressures and orientations on the blocks for both setups, and I think these results represent the best performance both cooling systems are capable of.

Test Rig:
Case: Coolermaster HAF XB EVO
Motherboard: ASUS X99-A USB 3.1
CPU: Intel i7 5960X. Cores at 4.2GHz at 1.175V, Cache at 4.0GHz at 1.1V
Memory: 8x8GB G.Skill Trident-Z at 3200 14-14-14-28, 1.35V
VGA: 2 GTX 980TI reference cards
Storage: Samsung 950 Pro NVMe, Crucial MX500 1TB SATA, Seagate 4TB HDD, 2 optical drives
Case Fans: Noctua A20 200mm (intake), Coolermaster SF120 120mm (exhaust), 2X Arctic P8 80mm (exhaust)
PSU: Corsair RM1000x

EVGA CLC 280 with better fans:
Radiator: 280mm x 27mm aluminium, 20fpi
Fans: Arctic P14 PWM in push/pull (exhaust)
Pump: The sticker on the rad says it's 12V, 250mA. 100% speed (2800RPM)
Reservoir: LOL
Block: Copper, 132 micro-fins
Tubing: Some kind of rubber sleeved with nylon
Coolant: AIO mixed-metal magic potion
TIM: Arctic MX-5

Custom 280mm Water:
Radiator: Koolance HX-CU1402V, 280mm x 30mm copper, 30fpi split-fin
Fans: Arctic P14 PWM in push/pull (exhaust)
Pump: Alphacool VPP755 V3 at 60% PWM (3500RPM - zero gain from running it any faster)
Reservoir: Alphacool Eisstation VPP, 240mL
Block: Alphacool Eisblock XPX, Nickel-plated copper, 84 micro-fins
Tubing: 10mm ID EPDM rubber
Coolant: 99.9975% distilled water, 0.0025% dimethyl benzyl ammonium chloride (biocide)
TIM: Arctic MX-5

Test Software:
AIDA64 Extreme stability test, FPU only mode
15 minute warm-up, followed by 3-minute log in HWiNFO64 of min/max/average

Results:

280mm AIO280mm CustomDifference
Ambient Temp23.123.10
Coolant Temp*32.0*30.0*-2.0C*
Core 058.357.5-0.8C
Core 159.957.9-1.0C
Core 266.664.6-2.0C
Core 363.460.4-3.0C
Core 465.163.0-2.1C
Core 562.560.9-1.6C
Core 664.361.9-2.6C
Core 761.058.4-2.6C
Core Average62.760.6-2.1C

*The method for measuring coolant temp was not the same. I'm using an Alphacool inline temp sensor to measure coolant temp after the radiator and before the CPU in the custom loop. For the AIO, the coolant temp is from the built-in sensor, and I have no idea whether it's before the CPU, after the CPU, or somewhere in between.

It looks to me like the AIO performed very well for what it costs. $70 plus a $40 fan upgrade and it only lost by 2C. Of course, the custom loop would totally bury it with more rad space, but that wasn't the point of this test. In terms of noise, there's not much to report. For some reason, the fans sounded slightly less annoying to me at full-speed on the custom loop's radiator, but that's probably just an idiosyncrasy of these Arctic fans. The pumps on both systems were very quiet and were a non-factor.

Raw Data:

AIO AIDA FPU.png


custom loop AIDA FPU .png
 
Last edited:
Just out of curiosity, why did you use the average instead of the maximum temps? I would think that your test was designed to find the maximum temps during a looped run. I don't think it changes the deltas much between the two setups.

Overall though, great info. (y)
 
Just out of curiosity, why did you use the average instead of the maximum temps? I would think that your test was designed to find the maximum temps during a looped run. I don't think it changes the deltas much between the two setups.

Overall though, great info. (y)

The maximum values are in the images I attached. 66C for the custom loop vs 68C for the AIO.

Speaking of images...

IMG_20220308_075745618.jpg

IMG_20220308_075932898.jpg

IMG_20220308_080323946.jpg

IMG_20220308_080302633.jpg
 
The maximum values are in the images I attached. 66C for the custom loop vs 68C for the AIO.

I'm talking about above in the OP where you listed the temps for each core for both setups. Comparing that to the pictures of AIDA, you used the average temps of the cores. I was wondering why you did that instead of using the max temps per core. The max temps for Core 0 for example are 3.5C higher for both setups. The answer could be as simple as the fans were set to ramp up under a load instead of at a constant speed throughout the test.
 
I'm talking about above in the OP where you listed the temps for each core for both setups. Comparing that to the pictures of AIDA, you used the average temps of the cores. I was wondering why you did that instead of using the max temps per core. The max temps for Core 0 for example are 3.5C higher for both setups. The answer could be as simple as the fans were set to ramp up under a load instead of at a constant speed throughout the test.

The reason why I like to look at the averages is because I feel like taking the mean of 180 readings over a 3 minute period and then rounding it to the nearest tenth is more precise than a single max integer value. On this platform, temperature readings are integers.

Core 0 always seems to have the most variation. I guess it's because Windows background processes must like that core.
 
yup that is about what i saw going from alu aios on the cpu(120) and gpu(240) to a loop w/ copper rads(120+240), ~3c max. the copper is really the only thing making a difference
 
yup that is about what i saw going from alu aios on the cpu(120) and gpu(240) to a loop w/ copper rads(120+240), ~3c max. the copper is really the only thing making a difference

That could well be the case. The difference always seems to be right there in the water temp in any test I do. So yeah, I guess I'd have to conclude that the AIO cold plate is very well optimized for the comparatively tiny amount of water flow that the pump provides. The plate itself is very thin which is good for heat transfer, and the fins are so incredibly fine, nothing in the custom market really comes close, even the Optimus blocks which are the best in that regard. I'm slightly tempted to pull apart the AIO pump/block, yank the impeller out of it, and see what happens when I blast water through it with a real pump. It might be good, or it might just blow apart.
 
I'm slightly tempted to pull apart the AIO pump/block, yank the impeller out of it, and see what happens when I blast water through it with a real pump. It might be good, or it might just blow apart.
works fine for me. im using an unaltered pump/block from a corsair aio with a g12 and it flows just fine, got a little whirlpool in the res after it, 2 rads and the cpu block.
 
works fine for me. im using an unaltered pump/block from a corsair aio with a g12 and it flows just fine, got a little whirlpool in the res after it, 2 rads and the cpu block.
That's interesting. Do you have power running to the AIO pump?
 
That's interesting. Do you have power running to the AIO pump?
nope, just tucked the wire to where i could plug it in if needed but didnt need to. the impellers are magnetic and float so it doesnt take much to spin them, its not like the water needs to turn a motor. id think they only add a tiny resistance like one of those inline flow indicators, if even that, as those are actually on an "axel".
edit: you do have to mod the g12 for 5000/6000 and 3000 series cards though, you just cut the mounting bracket in half, and widen two screw holes in the pump mount.
 
I'd say the biggest benefit to custom loops is being able to decide exactly what type of radiator you want for your specific use. Would be neat to compare a GTR rad with Noctua 3k rpm industrials with a standard AIO of the same size.
 
nope, just tucked the wire to where i could plug it in if needed but didnt need to. the impellers are magnetic and float so it doesnt take much to spin them, its not like the water needs to turn a motor. id think they only add a tiny resistance like one of those inline flow indicators, if even that, as those are actually on an "axel".
edit: you do have to mod the g12 for 5000/6000 and 3000 series cards though, you just cut the mounting bracket in half, and widen two screw holes in the pump mount.

Yeah, that makes total sense. I'll have to give it a shot one of these days just to see how it does. Maybe I'll do it when I tear this thing down to add another radiator. I might use one of these to bolt a 360 on top of the case. I figure that should knock off another 3 degrees.

51DBwjsse6L._AC_SL1280_.jpg
 
What was the fan speed?

Nvm, I see it now. Set to 1600 RPM, which in push/pull would play into the strength of a high FPI radiator like the Koolance. It would be interesting to see the results with a low FPI radiator and slower fans.
 
What was the fan speed?

Nvm, I see it now. Set to 1600 RPM, which in push/pull would play into the strength of a high FPI radiator like the Koolance. It would be interesting to see the results with a low FPI radiator and slower fans.

Probably higher temps but it might still beat the AIO. I tried locking the fans on the Koolance to 1200rpm, and the temps only went up by 1.5 degrees, which means the custom loop still wins even with 400rpm less fan speed than the AIO. Aluminum radiators really are hot garbage.
 
Probably higher temps but it might still beat the AIO. I tried locking the fans on the Koolance to 1200rpm, and the temps only went up by 1.5 degrees, which means the custom loop still wins even with 400rpm less fan speed than the AIO. Aluminum radiators really are hot garbage.

The difference probably would have been greater with low FPI radiators. It's been a long time, but in testing long ago, I think 15-20 FPI radiators passed 8 FPI radiators in performance at around the 800-1000 RPM mark in push/pull. 30 FPI radiators would probably require 1200-1400 RPM fans to match 8 FPI radiators before pulling away at higher fan speeds. That's why most custom watercooling radiators these days are in the 8-15 FPI range as most watercoolers are primarily concerned about low speed fan performance. It's also worth noting that 140mm fans generally have lower static pressure than their 120mm counterparts at the same RPM, so the numbers above will probably need to bump up by about 200 RPM or so.

At 1600 RPM, those Koolance radiators were just starting to get out of the range where high FPI is a detriment. The performance difference would most likely be more pronounced with lower speed fans and a low FPI radiator used in the custom loop.
 
The difference probably would have been greater with low FPI radiators. It's been a long time, but in testing long ago, I think 15-20 FPI radiators passed 8 FPI radiators in performance at around the 800-1000 RPM mark in push/pull. 30 FPI radiators would probably require 1200-1400 RPM fans to match 8 FPI radiators before pulling away at higher fan speeds. That's why most custom watercooling radiators these days are in the 8-15 FPI range as most watercoolers are primarily concerned about low speed fan performance. It's also worth noting that 140mm fans generally have lower static pressure than their 120mm counterparts at the same RPM, so the numbers above will probably need to bump up by about 200 RPM or so.

At 1600 RPM, those Koolance radiators were just starting to get out of the range where high FPI is a detriment. The performance difference would most likely be more pronounced with lower speed fans and a low FPI radiator used in the custom loop.
Eh, nowadays FPI can be misleading. Overall radiator design (and fin design) can affect the performance curve just as much as raw FPI.
 
Eh, nowadays FPI can be misleading. Overall radiator design (and fin design) can affect the performance curve just as much as raw FPI.

Yes, I was just making a generalization, of course other things like thickness and flow path also come into play. Also, the Koolance radiator used in this test is a pretty old design if I'm not mistaken. I don't recall Koolance updating their radiators ever since they were introduced, which I believe was in the mid 2000s.

Point being, the radiator used in the custom loop comparison wasn't the optimal choice in terms of FPI for the fan speeds being used. On top of that, since you brought up design, I'm pretty sure it's an old and outdated design as well. A newer design appropriate for the fan speeds being used would almost certainly have resulted in a larger performance delta.

I miss Martin's Liquid Lab....
 
Yes, I was just making a generalization, of course other things like thickness and flow path also come into play. Also, the Koolance radiator used in this test is a pretty old design if I'm not mistaken. I don't recall Koolance updating their radiators ever since they were introduced, which I believe was in the mid 2000s.

Point being, the radiator used in the custom loop comparison wasn't the optimal choice in terms of FPI for the fan speeds being used. On top of that, since you brought up design, I'm pretty sure it's an old and outdated design as well. A newer design appropriate for the fan speeds being used would almost certainly have resulted in a larger performance delta.

I miss Martin's Liquid Lab....
Ah, is it? Nowadays it feels like everything has been around the block a few times :D

I missed out on MLL, though I did like ThermalBench. Seems impossible to find anything nowadays with liquid delta charts, restriction, etc.

Probably something I'd do on the side if I had "FU" money just for fun.
 
Point being, the radiator used in the custom loop comparison wasn't the optimal choice in terms of FPI for the fan speeds being used. On top of that, since you brought up design, I'm pretty sure it's an old and outdated design as well. A newer design appropriate for the fan speeds being used would almost certainly have resulted in a larger performance delta.

Old does not necessarily mean inferior. On the xtremerigs radiator round-up, the Koolance left every other slim radiator in the dust at 1300RPM and above in push/pull, including the HW Labs GTS. Data is for the 360mm version, but it should carry over I think.

HX-CU1020V-Slims-PC-6.4-PP-Avg-1300.png

HX-CU1020V-Slims-PC-6.4-PP-Avg-1850.png


This is how it stacked up against all radiators, thick included (which I didn't use ofc due to a lack of space)

pp1300-1 (1).png
pp1850-819x1024.png
 
Last edited:
If the resistance of the radiator you used is similar to the 360 version then you have about 2.5 PSI worth of pressure drop from the radiator and CPU block alone at 1gpm. My system has around 2 PSI at 1GPM from radiator and block, but I barely manage 0.5gpm at 60% with a d5 so you are probably at 0.4 or so GPM with your setup when running 60% pump. My guess is that you top out at 0.8-0.9gpm with pump at full speed.

Not sure how the XPX block reacts to flowrate, but you probably are leaving somewhere between 0.5 and 1.5 degrees on the table by not having a DDC pump at full speed (most from CPU block and some from radiator). Some waterblocks gain a lot and some barely gain anything, but I doubt you would gain less than 0.5 degrees by going to 1GPM. There are also waterblocks that should perform a bit better than the XPX so I do believe if you had optimal flow rate and the best waterblock then the delta would be 1-3 degrees larger.
 
Last edited:
If the resistance of the radiator you used is similar to the 360 version then you have about 2.5 PSI worth of pressure drop from the radiator and CPU block alone. My system has around 2 PSI at 1GPM from radiator and block, but I barely manage 0.5gpm at 60% with a d5 so you are probably at 0.4 or so GPM with your setup when running 60% pump. My guess is that you top out at 0.8-0.9gpm with pump at full speed.

Not sure how the XPX block reacts to flowrate, but you probably are leaving somewhere between 0.5 and 1.5 degrees on the table by not having a DDC pump at full speed (most from CPU block and some from radiator). Some waterblocks gain a lot and some barely gain anything, but I doubt you would gain less than 0.5 degrees by going to 1GPM.

The 280mm version is a lot less restrictive than the 360. Like I said in the OP, I don't gain anything going from 60% to 100% on the pump. I saw a difference of maybe 0.2C, but that's well within the margin of error, and my reservoir turns into a fountain with the pump at full speed, so I don't run it that high.

hx-cu1020.gif


hx-cu140.gif
 
Huge thanks for testing this and sharing data. As much as I love custom loops, it's hard to ignore that the current crop of AIOs is really quite good. Considering the price and the simplicity, it's hard to justify a custom loop these days unless you're doing a multi-radiator setup or you just like having the best solution possible at any cost.

It would be interesting to see one of the more ambitious and well-funded reviewers or YT channels do an all-out optimization challenge to see how much they can beat an AIO with a unicorn combination of radiator/fans/pump/flowrate. I expect there are a couple more degrees that could be squeezed out of a custom loop with enough testing (and $$$) but it's not going to be much.
 
Huge thanks for testing this and sharing data. As much as I love custom loops, it's hard to ignore that the current crop of AIOs is really quite good. Considering the price and the simplicity, it's hard to justify a custom loop these days unless you're doing a multi-radiator setup or you just like having the best solution possible at any cost.

It would be interesting to see one of the more ambitious and well-funded reviewers or YT channels do an all-out optimization challenge to see how much they can beat an AIO with a unicorn combination of radiator/fans/pump/flowrate. I expect there are a couple more degrees that could be squeezed out of a custom loop with enough testing (and $$$) but it's not going to be much.

Custom loops have never been justifiable from a price/performance perspective for CPUs. The best tower coolers were always within a few degrees of custom loops at typically 1/3 of the price or less. Many 240 AIOs struggle to beat the best of tower coolers.

Custom loops were somewhat justifiable in the days before AIOs for GPUs were a thing, which is obviously no longer the case today.

The limiting factor in an unlimited custom cooling setup will be the heat transfer rate between the CPU die and the coolant. You can increase flow rate by adding pumps or using stronger pumps to the point where the delta T between block inlet and outlet is negligible, though you'll probably run into pressure limitations first. Enough radiators and/or high speed fans will have the coolant at ambient temperature by the time it returns to the CPU block. The only "unicorn" factor, as it were, is the design of the CPU block itself. That is pretty easy to test on its own.

You can actually calculate the theoretical savings with the data provided. The custom loop has the coolant temperature at 7 degrees above ambient, so there's a potential for lowering the temperature by 7 degrees with enough radiators and fans. The 5960x can pull over 350 watts. At a flow rate of 0.5 GPM, the delta T will be about 2.5 C. Increasing the flow rate can theoretically lower core temperatures by another 2.5 C, but in practice this doesn't actually happen. This is because the rate of heat transfer is based on the temperature difference between the core and the coolant. When you have a 25-30 C delta, a 2 C change between inlet and outlet lowers heat transfer rate at the hottest part by 8% at the worst, which means averaged across the block is actually around 4% (assuming linear loss, which is reasonable). With a 30 C delta, getting rid of the inlet/outlet delta nets you a 1.2 C change at best, but you would need to be running well over 10 GPM. Doubling the flow rate to 1 GPM lowers heat transfer by 4% at worst, translating to 2% on average. Theoretically this would lead to 0.6 C change at best and would typically be lower. Disclaimer: I am fudging the numbers a bit for the sake of simplifying things, and by no means should you use this as an example in your heat transfer class. This should, however, be a reasonable approximation. I'm also doing this at 3 AM so.... yeah.
TL;DR: A unicorn custom setup would improve temps by about 8-9 C over OP's current setup, assuming the CPU block used is the best for OP's CPU on the market. This unicorn setup will need flow rates of well over 10 GPM and a very large number of radiators and/or high speed fans (think Mo-Ra3 420 with push/pull 2000 RPM fans or more).

Watercooling is a prime example of exponential deminishing returns. Doubling the radiator setup to 2x280 radiators would net 3.5 C at best. Doubling that to 4x280 would get another 1.75 C for a grand total of 5.25 C. 8x280 will get a total of 6.125 C. And so on and so forth until you have an infinitely large radiator setup that gets the coolant down to ambient temperatures.
 
Custom loops have never been justifiable from a price/performance perspective for CPUs. The best tower coolers were always within a few degrees of custom loops at typically 1/3 of the price or less. Many 240 AIOs struggle to beat the best of tower coolers.

Custom loops were somewhat justifiable in the days before AIOs for GPUs were a thing, which is obviously no longer the case today.

The limiting factor in an unlimited custom cooling setup will be the heat transfer rate between the CPU die and the coolant. You can increase flow rate by adding pumps or using stronger pumps to the point where the delta T between block inlet and outlet is negligible, though you'll probably run into pressure limitations first. Enough radiators and/or high speed fans will have the coolant at ambient temperature by the time it returns to the CPU block. The only "unicorn" factor, as it were, is the design of the CPU block itself. That is pretty easy to test on its own.

You can actually calculate the theoretical savings with the data provided. The custom loop has the coolant temperature at 7 degrees above ambient, so there's a potential for lowering the temperature by 7 degrees with enough radiators and fans. The 5960x can pull over 350 watts. At a flow rate of 0.5 GPM, the delta T will be about 2.5 C. Increasing the flow rate can theoretically lower core temperatures by another 2.5 C, but in practice this doesn't actually happen. This is because the rate of heat transfer is based on the temperature difference between the core and the coolant. When you have a 25-30 C delta, a 2 C change between inlet and outlet lowers heat transfer rate at the hottest part by 8% at the worst, which means averaged across the block is actually around 4% (assuming linear loss, which is reasonable). With a 30 C delta, getting rid of the inlet/outlet delta nets you a 1.2 C change at best, but you would need to be running well over 10 GPM. Doubling the flow rate to 1 GPM lowers heat transfer by 4% at worst, translating to 2% on average. Theoretically this would lead to 0.6 C change at best and would typically be lower. Disclaimer: I am fudging the numbers a bit for the sake of simplifying things, and by no means should you use this as an example in your heat transfer class. This should, however, be a reasonable approximation. I'm also doing this at 3 AM so.... yeah.
TL;DR: A unicorn custom setup would improve temps by about 8-9 C over OP's current setup, assuming the CPU block used is the best for OP's CPU on the market. This unicorn setup will need flow rates of well over 10 GPM and a very large number of radiators and/or high speed fans (think Mo-Ra3 420 with push/pull 2000 RPM fans or more).

Watercooling is a prime example of exponential deminishing returns. Doubling the radiator setup to 2x280 radiators would net 3.5 C at best. Doubling that to 4x280 would get another 1.75 C for a grand total of 5.25 C. 8x280 will get a total of 6.125 C. And so on and so forth until you have an infinitely large radiator setup that gets the coolant down to ambient temperatures.
Custom loops are just as much about noise as they are about cooling performance. It is essentially small, low temps and low noise where you have to pick 2 (ignoring cost). When your goal is to have low temps at noise levels where an AIOs and air towers could barely cool 100w then custom loop is the way to go. Same for if you want a cool GPU at low noise. Granted you must go big and expensive, but if low temps and low noise is the goal then there aren't many other alternatives. Granted having temps that an AIO or Air cooler can achive with 50+ bdA noise at around 30-31dbA is not a goal for most people. Price to performance ratio is subjective and depends on what you value. If you want a very cool and quiet system at full load then the performance of an AIO or an air tower is not satisfactory (even the waterpump at usable speed on most, if not all AIOs is too loud for my requierments), which means your price to performance ratio makes them infinitely expensive for the performance they provide.


The cooling engine of the waterblock has a larger impact than the GPM unless the waterblock has a suboptimal cooling engine (e.g. EK velocity on AMD 3000 and 5000 CPUs gain a lot while AM4 optimized waterblocks have small gains from 0.5gpm to 1gpm). Very few waterblocks have any meaningful improvment above 1.5gpm and most have small improvements above 1gpm. There are however several degrees difference between the good and the mediocre waterblocks.
 
Some great posts this morning. I have some more data I can share since I've continued to try various things with this loop over the past month. The first thing I did after collecting the original data was to experiment with the formula for my homebrew coolant. After some trial and error, I ended up with the following:

Previous:
Distilled water with 25ppm dimethyl benzyl ammonium chloride (surfectant and biocide)
Current:
Distilled water with 100ppm dimethyl benzyl ammonium chloride (let's call it DMBAC) and 200ppm sodium bicarbonate (pH buffer to improve longevity)

This actually improved temps by a consistent and repeatable 0.5C, which increased the custom loop's lead from 2.0C to 2.5C. DMBAC is a very potent surfactant like dish soap or water wetter, and it has the side benefit of killing bacteria, algae, and fungus. The best concentration was 100ppm - more than that resulted in foaming. This has to do with staying just below critical micelle concentration where surface tension stops decreasing appreciably, but foaming does not occur.

I know 0.5C doesn't sound like much, because it isn't much, but it will affect all future data for other tests like fan speed, pump speed, radiator size, cooling engine, etc, so I've documented it.

Here is the data for the coolant change. The cores fluctuate a bit from run to run, but the average is always about 0.5C better with the new coolant.

Old FormulaNew FormulaDifference
Core 0 to Air Delta34.433.6-0.8C
Core 1 to Air Delta34.834.0-0.8C
Core 2 to Air Delta41.541.0-0.5C
Core 3 to Air Delta37.637.3-0.3C
Core 4 to Air Delta39.939.0-0.9C
Core 5 to Air Delta37.837.0-0.8C
Core 6 to Air Delta38.838.3-0.5C
Core 7 to Air Delta35.335.4+0.1C
Core Average to Air Delta41.541.0-0.5C

thermals 3-11-22 0924 AIDA FPU (15m).png


Then, I ran a test with the fan speed at 1200RPM instead of 1600RPM. It still beats the AIO by 1.2C with much less noise.


thermals 3-14-22 1555 AIDA FPU (15m).png


Then I swapped from Arctic P14s to Phanteks F140XP. No significant change to temps or overall loudness, but much smoother noise profile to my ears.
thermals 3-29-22 0632 AIDA FPU (15m).png


Then I added a Bykski 120mm x 40mm radiator behind the case's exhaust fan, and bumped the RPM on that fan from a max of 1200 to a max of 1500. Approximately 1.5C improvement, noise increase not noticeable.

thermals 4-06-22 1520 AIDA FPU (15m).png


I'm debating the next modification. I was going to add an externally mounted 360mm radiator, but I might do a 1080 instead. Eventually, I will upgrade my GPU and add it to the loop, so I want to have enough capacity to maintain or exceed the current level of quietness and performance with an additional 300+ watts or so of max heat. I might also ditch the internal rads altogether since at that point they might be adding more complexity and restriction than they're worth.
 
  • Like
Reactions: Tsumi
like this
Custom loops are just as much about noise as they are about cooling performance. It is essentially small, low temps and low noise where you have to pick 2 (ignoring cost). When your goal is to have low temps at noise levels where an AIOs and air towers could barely cool 100w then custom loop is the way to go. Same for if you want a cool GPU at low noise. Granted you must go big and expensive, but if low temps and low noise is the goal then there aren't many other alternatives. Granted having temps that an AIO or Air cooler can achive with 50+ bdA noise at around 30-31dbA is not a goal for most people. Price to performance ratio is subjective and depends on what you value. If you want a very cool and quiet system at full load then the performance of an AIO or an air tower is not satisfactory (even the waterpump at usable speed on most, if not all AIOs is too loud for my requierments), which means your price to performance ratio makes them infinitely expensive for the performance they provide.


The cooling engine of the waterblock has a larger impact than the GPM unless the waterblock has a suboptimal cooling engine (e.g. EK velocity on AMD 3000 and 5000 CPUs gain a lot while AM4 optimized waterblocks have small gains from 0.5gpm to 1gpm). Very few waterblocks have any meaningful improvment above 1.5gpm and most have small improvements above 1gpm. There are however several degrees difference between the good and the mediocre waterblocks.

Price/performance for me means strictly cooling potential and disregarding noise as long as noise levels remained reasonable (i.e. not using 2500+ RPM fans). I also remember in the past that the top air coolers like the NH-D14 were usually only 5-7 degrees behind custom loops using 360 mm radiators at similar noise levels (can't find those articles now, been way too long). Most people wouldn't care about how noisy their CPU air cooler is when GPU coolers tend to be much noisier.

Some great posts this morning. I have some more data I can share since I've continued to try various things with this loop over the past month. The first thing I did after collecting the original data was to experiment with the formula for my homebrew coolant. After some trial and error, I ended up with the following:

Previous:
Distilled water with 25ppm dimethyl benzyl ammonium chloride (surfectant and biocide)
Current:
Distilled water with 100ppm dimethyl benzyl ammonium chloride (let's call it DMBAC) and 200ppm sodium bicarbonate (pH buffer to improve longevity)

This actually improved temps by a consistent and repeatable 0.5C, which increased the custom loop's lead from 2.0C to 2.5C. DMBAC is a very potent surfactant like dish soap or water wetter, and it has the side benefit of killing bacteria, algae, and fungus. The best concentration was 100ppm - more than that resulted in foaming. This has to do with staying just below critical micelle concentration where surface tension stops decreasing appreciably, but foaming does not occur.

I know 0.5C doesn't sound like much, because it isn't much, but it will affect all future data for other tests like fan speed, pump speed, radiator size, cooling engine, etc, so I've documented it.

Here is the data for the coolant change. The cores fluctuate a bit from run to run, but the average is always about 0.5C better with the new coolant.

Old FormulaNew FormulaDifference
Core 0 to Air Delta34.433.6-0.8C
Core 1 to Air Delta34.834.0-0.8C
Core 2 to Air Delta41.541.0-0.5C
Core 3 to Air Delta37.637.3-0.3C
Core 4 to Air Delta39.939.0-0.9C
Core 5 to Air Delta37.837.0-0.8C
Core 6 to Air Delta38.838.3-0.5C
Core 7 to Air Delta35.335.4+0.1C
Core Average to Air Delta41.541.0-0.5C

View attachment 461550

Then, I ran a test with the fan speed at 1200RPM instead of 1600RPM. It still beats the AIO by 1.2C with much less noise.


View attachment 461552

Then I swapped from Arctic P14s to Phanteks F140XP. No significant change to temps or overall loudness, but much smoother noise profile to my ears.
View attachment 461553

Then I added a Bykski 120mm x 40mm radiator behind the case's exhaust fan, and bumped the RPM on that fan from a max of 1200 to a max of 1500. Approximately 1.5C improvement, noise increase not noticeable.

View attachment 461554

I'm debating the next modification. I was going to add an externally mounted 360mm radiator, but I might do a 1080 instead. Eventually, I will upgrade my GPU and add it to the loop, so I want to have enough capacity to maintain or exceed the current level of quietness and performance with an additional 300+ watts or so of max heat. I might also ditch the internal rads altogether since at that point they might be adding more complexity and restriction than they're worth.

Great info!
 
Price/performance for me means strictly cooling potential and disregarding noise as long as noise levels remained reasonable (i.e. not using 2500+ RPM fans). I also remember in the past that the top air coolers like the NH-D14 were usually only 5-7 degrees behind custom loops using 360 mm radiators at similar noise levels (can't find those articles now, been way too long). Most people wouldn't care about how noisy their CPU air cooler is when GPU coolers tend to be much noisier.
You are just proving my point in that price to performance ratio is subjective. Noise is a high priority for me along with good cooling while price isn't that much of a priority while price is very important to you.

You are aware that there are bigger rads than 360s and you can run more than one? Not sure why the GPU would be that loud if your GPU is in a loop with plenty of capacity so you can run high quality fans on low speed which means there is very little noise. The 30-31dbA that I mentioned included a GPU drawing 350w. If you have ever measured a computer with a proper DB meter then you would know that 30dbA is quite low (close to the noise floor of most rooms) and the jump to 35dbA or higher is massive.
 
I don't think I've ever seen anyone justify custom loops as a price/performance solution lol. Much like buying high end GPU/CPU etc, the higher you go, the more you pay for ever decreasing overall gains in performance. Sure AIOs are usually good enough, but sometimes you want better because you just do. There's also the hobby side of it. It's a lot more hands on than just slapping AIOs all over the place, and some of us are tinkerers.
 
I never did custom wc for the performance. I just though it was cool and good hobby for the past 10 years. That said I am done with WC when I do my next major overhaul which will be a while. I don't have the passion for PC building anymore. I am perfectly fine plopping down on the couch with a PS5.
 
You are just proving my point in that price to performance ratio is subjective. Noise is a high priority for me along with good cooling while price isn't that much of a priority while price is very important to you.

You are aware that there are bigger rads than 360s and you can run more than one? Not sure why the GPU would be that loud if your GPU is in a loop with plenty of capacity so you can run high quality fans on low speed which means there is very little noise. The 30-31dbA that I mentioned included a GPU drawing 350w. If you have ever measured a computer with a proper DB meter then you would know that 30dbA is quite low (close to the noise floor of most rooms) and the jump to 35dbA or higher is massive.

I don't consider noise as part of the performance metric. It is in its own metric entirely: noise levels, or more specifically, noise to performance ratio. How well it cools is the performance metric. You balance out how important each metric is to you. As also said, the hobby/fun factor is also a metric that can be considered, and once again it shouldn't be lumped in with any of the other metrics. Ease of care and maintenance is also another metric.

If I valued price/performance, I wouldn't be running a custom watercooling setup with 480 and 280 radiators.

A single 360 radiator was more than enough to cool 200 watt CPUs of the time of that comparison where adding additional radiator space didn't amount to much. The NH-D14 typically trailed performance by about 5-7 degrees at the same noise level while being less than 1/3 the cost of a custom setup.

You can get a large AIO that's about half the cost of a custom setup and performs with 2-3 degrees as nicely shown by OP. Two large AIOs (or purchase a GPU with AIO already installed) and you'll get performance almost equal to that of a custom setup without the hassle of setup and maintenance, if you consider those things to be a hassle.

I'll say it again, custom watercooling has never been and never will be about price/performance. It will be about noise/cooling, peak performance, the unique/cool factor, or some combination of the three.
 

JSHamlet234

Thanks for all the great data!

Peeps going on about how AIO/CLC is half or less the cost of custom loop are not looking at whole picture. To me AIOs are pre-assembled and filled components with threaded fittings, fill port, etc. that can be repaired if needed. Granted they are lower end of custom loop components, but at least they are component systems. CLCs on the other hand are factory assembled systems with no provision for servicing or repairing with pump flowrate at bottom end of what custom loop pumps flow. CLCs cooling hi-wattage components often fail in 2 years, sometimes less, but an AIO or custom loop can last many, many years when serviced annually and even longer by replacing components (pump is always first to fail). Upgrade waterblock to newer mounts and AIO / custom loops can last almost forever.

Basic CLC is $80-120 last a couple years, then fail or loose performance and have to be junked and replaced with something else.

AIO / custom loops start at $90 but most are $150-180 .. twice what a CLC costs but will at least as long before needing repair and if serviced will last easily twice as long. Additional components increase life to almost indefinitely.

I would much rather have AIO / custom loop .. and if not running super high heat components good air coolers. ;)
 

JSHamlet234

Thanks for all the great data!

Peeps going on about how AIO/CLC is half or less the cost of custom loop are not looking at whole picture. To me AIOs are pre-assembled and filled components with threaded fittings, fill port, etc. that can be repaired if needed. Granted they are lower end of custom loop components, but at least they are component systems. CLCs on the other hand are factory assembled systems with no provision for servicing or repairing with pump flowrate at bottom end of what custom loop pumps flow. CLCs cooling hi-wattage components often fail in 2 years, sometimes less, but an AIO or custom loop can last many, many years when serviced annually and even longer by replacing components (pump is always first to fail). Upgrade waterblock to newer mounts and AIO / custom loops can last almost forever.

Basic CLC is $80-120 last a couple years, then fail or loose performance and have to be junked and replaced with something else.

AIO / custom loops start at $90 but most are $150-180 .. twice what a CLC costs but will at least as long before needing repair and if serviced will last easily twice as long. Additional components increase life to almost indefinitely.

I would much rather have AIO / custom loop .. and if not running super high heat components good air coolers. ;)
Wait, what? Aren't AIOs the exact same thing as CLC? Back in "the day" last time I checked all of the Asetek based designs were called AIOs. Has the definition of AIO changed on me?
 
Wait, what? Aren't AIOs the exact same thing as CLC? Back in "the day" last time I checked all of the Asetek based designs were called AIOs. Has the definition of AIO changed on me?
To me "AIO" is "All in One", "CLC" is "Closed Loop Cooler".

AIO is parent group and CLC are a sub-group of AIO.

AIOs are a individual components (same as custom loops use) assembled into a loop and filled before shipping to consumer. AIOs have threaded fittings, a fill port, etc. It can be drained, flushed and refilled and if a component goes bad it can be replaced. Granted, the components used in AIOs are pretty much lower level, low cost custom loop components. But at least the system can be sericed and as components wear out they can be replaced. Loop can be expanded to include more radiator area and/or more waterblocks (but will likely need a more powerful pump to perform well. AIO pumps are usually higher spec than CLC pumps, but not as good as D5, most common pump in custom loops. AlphaCool Eisbaer series and Swiftech Drive X3 AIO are examples of what I call AIO and not CLC.

CLCs are factory assembled units that cannot be disassembled, usually without a fill port. When something goes bad (low coolant level and/or pump failure) it is thrown away and replaced with a new system.

Sorry, I should have explained it to start with. While I'm not the only one who uses AIO and CLC to explain difference, it's terminology not used by everyone. ;)
 
Wait, what? Aren't AIOs the exact same thing as CLC? Back in "the day" last time I checked all of the Asetek based designs were called AIOs. Has the definition of AIO changed on me?
AIO is pre-built and pre-filled. Most AIOs are closed systems, but a few are made with with changeable components. E.g. Alphacool Eisbear extreme is built from custom loop components and can be upgrade while most AIOs are built by asetek, cool-it etc. and are not meant to be serviced.
 
To me "AIO" is "All in One", "CLC" is "Closed Loop Cooler".

AIO is parent group and CLC are a sub-group of AIO.

AIOs are a individual components (same as custom loops use) assembled into a loop and filled before shipping to consumer. AIOs have threaded fittings, a fill port, etc. It can be drained, flushed and refilled and if a component goes bad it can be replaced. Granted, the components used in AIOs are pretty much lower level, low cost custom loop components. But at least the system can be sericed and as components wear out they can be replaced. Loop can be expanded to include more radiator area and/or more waterblocks (but will likely need a more powerful pump to perform well. AIO pumps are usually higher spec than CLC pumps, but not as good as D5, most common pump in custom loops. AlphaCool Eisbaer series and Swiftech Drive X3 AIO are examples of what I call AIO and not CLC.

CLCs are factory assembled units that cannot be disassembled, usually without a fill port. When something goes bad (low coolant level and/or pump failure) it is thrown away and replaced with a new system.

Sorry, I should have explained it to start with. While I'm not the only one who uses AIO and CLC to explain difference, it's terminology not used by everyone. ;)

To add to that, AIOs using servicable components are a relatively new addition to the market. A lot of the older folks have gotten used to using CLC and AIOs interchangeably. I would probably call them prebuilt custom systems.
 
To add to that, AIOs using servicable components are a relatively new addition to the market. A lot of the older folks have gotten used to using CLC and AIOs interchangeably. I would probably call them prebuilt custom systems.
Swiftech started selling repairable AIOs many years ago. I can't remember exactly when but I remember them about the same time Asetek's CLCs hit the market. Asetek got a cease-and-desist order against Swiftech in 2013 over patent on CLC. Asetek called them LCLC for Low-Cost Liquid Cooling way back in 2007.
 
Last edited:
All I know is literally everywhere has called the standard crimped hose sealed units AIOs forever.

Correct or not, that's what people are going to think when someone says AIO.
 
All I know is literally everywhere has called the standard crimped hose sealed units AIOs forever.

Correct or not, that's what people are going to think when someone says AIO.
Some think everything is "AIO", some think of "CLC" as a sub-group of give .. and some think CLCs are just a way to get cooling like custom water cooling loops have for a fraction the price because they are "water cooling". ;)

If you understand that CLCs are only marginally the same thing as a quality custom loop in quality and performance, you can understand why we differentiate between AIOs and CLCs.
 
Last edited:
Some think everything is "AIO", some think of "CLC" as a sub-group of give .. and some think CLCs are just a way to get cooling like custom water cooling loops have for a fraction the price because they are "water cooling". ;)
I wrap AIO and CLCs into the latter. It is watercooling simplified. Though nowadays you can grab cheap starter custom kits from China on Amazon/Ebay for like $110 or so. No better than how I first started tinkering around with the hobby when I didn't have much money. Cheap plastic tubing from home depot, cheap plastic sharkbites off ebay, 2nd hand blocks off ebay, etc.
 
Back
Top