Cooler Comparison Testing
Cooler Testing Methods
To best gage the quality of the system coolers under review, system CPU temperature and GPU temperature measurements were taken with both the CPU and GPU at idle and under load. To replicate idle conditions, the system was rebooted and allowed to sit idle for 10 minutes. To replicate a stress system load, Aida64 System Stability Test was used in conjunction with FurMark for 30 minutes per run. After each run, the system was shut down and allowed to rest for 10 minutes to cool down. Then the CPU cooler was removed, cleaned, and remounted to the CPU with fresh thermal paste applied. This procedure was repeated three times each for the stock and overclocking speed run. For all run FurMark was set to run at a 1280×1024 resolution and 8x MSAA in Burn-in Test mode.
Note that for all Koolance system runs, both the pump and fans were set to run at level 10. For the Glacer-240L system runs, the fans were set to run at full speed, directly connected to the PSU via fan power cables. For all runs, the ASUS Poseidon card was left to run at stock speed settings.
CPU temperature measurements were taken directly from the CPU thermistors using RealTemp (the newer Tech|Inferno edition). Because of the volatile nature of the Haswell thermistor readings, the Z87-based system temperatures were measured in a different manner. For idle temperatures, the highest recorded value was used for the run. For load temperatures, a series of three values were notated: the average (high and low) across all cores, the average (high and low) across the single highest core, and the high temperature.
GPU temperature measurements were taken directly from the GPU thermistors using TechPowerUp GPU-Z v0.7.7. For both the idle and load temperatures, the highest recorded value in the application were used for the run. Note that the temperature values are reported as deltas rather than absolute temperatures with the delta value reported calculated as GPU temperature – ambient temperature.
Note that the temperature values are reported as deltas rather than absolute temperatures with the delta value reported calculated as CPU temperature – ambient temperature. For all tests, room ambient temperature was maintained between 23-27C.
Intel Z87-based Haswell System Testing
CPU Stock Speed Testing
The CPU stock speed testing was conducted with the BIOS defaults set for the CPU (including enabling of the CPU-integrated graphics processor) and Turbo Mode disabled, equating to a 3.4GHz CPU speed, 1600MHz memory speed, and 100MHz base clock. The Intel SpeedStep functionality remained enabled for the duration of the testing to get realistic CPU idle performance conditions.
In a dual block configuration, both the Koolance and Cooler Master kits saw an increase in temperatures, most evident in the CPU temperatures. The Glacer 240L was better able to absorb the additional heat load of the added GPU with its CPU temperatures climbing by an average of 5C while the Koolance system saw an average increase of 8C in CPU temperatures over single block operation.
The increase in GPU die temperatures was less dramatic with both kits averaging an increase of 2C in the multi-block configuration. The larger 240mm radiator integrated into the Glacer 240L seemed to have better cooling absorption potential on average than the Koolance unit with its 120mm radiator.
CPU Overclocked Speed Testing
The CPU overclocked speed testing was conducted with known stable settings from a previous board review with Turbo Mode disabled, equating to a 4.68GHz CPU speed, 1780MHz memory speed, 4.0GHZ ring bus speed, and 167MHz base clock. Also, the CPU-integrated graphics processor was disabled to reduce the processor heat generation. The Intel SpeedStep functionality remained enabled for the duration of the testing to get realistic CPU idle performance conditions.
Board voltage settings were configured as follows:
- CPU Core Voltage – 1.25 + 0.005
- VCCIN Voltage – 1.90
- DRAM Voltage – 1.55
- CPU Ring Voltage – 1.125 + 0.005
- CPU SA Voltage Offset – +0.100
- CPU IO Analogue Voltage Offset – +0.100
- CPU IO Digital Voltage Offset – +0.100
- PCH 1.05 Voltage – 1.120
The overclocked Haswell processor was a better test of the cooling potential, since the CPU dumps significantly more heat into the cooling loop at higher speeds. The EXT-440CU was balanced on a knife's edge of stability with minimal CPU throttling witnessed throughout the run (2% or less according to the AIDA64 measurement tool). The unit's CPU temperatures exceeded those of the single block runs by up to 11C. The Glacer 240L performed significantly better, with its temperatures going up by a maximum of 3C under load
.
Similar to the stock speed CPU tests, the GPU temperatures did not increase significantly for either unit. The Glacer 240L saw an increase of 3C while GPU temperature on the EXT-440CU increased by 4C under load.
I really don’t agree with
I really don’t agree with this diagrams flow direction.
Should be out of rad into res and out of res to CPU or GPU…. the end of the loop should be going to the rad out of rad to pump res out to CPU/GPU
You want to go directly from
You want to go directly from radiator to CPU block so that the CPU get the absolutely coolest possible coolant. If you go from the pump or reservior, the coolant will absorb heat from the rest of the loop or from the pump before going to CPU…
That Theory is sound, but in
That Theory is sound, but in practices it dose not really matter, because if you have a good pump, you have about 0.5c drop in temp from inlet to outlet of the rad.
And unless you want to be able to get the last one MHz out of your OC, i would not bother to much.
The main important things you need to pay attention to is a good rad and a good pump.
Also push fan setup is more effective then a pull setup for your rad.
I would avoid the EXT-440CU pump unit as the plague, unless you install a extra rad, as its a sub par solution, the CM pomp/CPU block is a o.k. start set, but noting more then that.
I just build in a Fractal Design R4 of a friends system, this set in:
1x EK-D5 PWM Motor 12V DC PWM Pump Motor – €70,95
1x Alphacool HF Screw connector G1/4 10mm – Deep Black – €47,60
2x Akasa 140mm Apache PWM fan Black – 1300RPM – €31,90
1x EK-Ekoolant CLEAR premix 1L – €7,10
Monsoon Silver Bullet Antimicrobial G1/4 Plug – €13,00
XSPC Acrylic Dual 5.25 bay Reservoir for D5 – €64,05
Alphacool NexXxoS UT60 Full Copper 280mm – €75,95
EK-Supremacy – Acetal + Copper – €60,95
EK-FC780 GTX Ti – €90,90
Totaal 465 euro
The pump is a PWM D5, so you dont need a special controller to adjust the speed, you can just connect it and control it with your motherboard fan controller software, PWM really rocks!!!!
A nice fat rad like the NeXxos UT60 cools a lot better then the rad’s used in the article.
For comparison with a EXT-440CU and a Cooler Master Glacer 240L:
EXT-440CU – 120×210 rad 12 x 12 x 2.7 = 388cm3
240L – 120×240 rad 12 x 24 x 2.7 = 777cm3
UT60 – 140×280 rad 14 x 28 x 6 = 2352cm3
And till a certain point, size really maters with cooling!
The more rad you have the silencer your system is.
And yeah a custom build loop cost a lot more, but also works a lot better, and except for the GPU block, can be re-used on many new systems.
Quality over budget gives you better cooling, a more silent system and longer use of the parts, and above all it looks better to.
You are absolutely correct
You are absolutely correct about loop order. People usually have 0.1C between hottest and coldest point in the loop with normal flow.
And yet, I tend to disagree with push/pull thing.
The thing is, if you use a static pressure fan, it’s not really important if it’s push or pull for the cooling. The only difference is how convenient is the cleaning. And it’s way better when you use pull 🙂 So I’m ready to sacrifice 0.x C in temperature for the easy cleaning.
My setup is Res -> Pump -> VGA mosfet -> GPU -> 120 rad -> CPU -> 240 rad. Running like a charm for many years.
The most load was with overclocked GTX570 + overclocked 4770K. ~31C idle, below 60C (usually 55-56) under load @ 25-27C ambient.
Block order does not matter
Block order does not matter in a watercooling loop. The only important thing is reservoir to pump, the rest can be in any order and it will not impact cooling performance.
You are very wrong
You are very wrong there…
If the best overall performance is what matters (meaning you want the lowest possible temps on all of the components), then the block with the least flow resistance is the first after the pump and the block with the highest resistance is the last. Typically, the CPU block is the highest flow resistance – so that will be the last in the loop.
It does not really matter if the Rad is the first or the last in the loop. However, the pump is the most temperature sensitive component in the loop, so the rad should be the last component in the loop – to get the coolest possible coolant to the pump – as it will fail much sooner if the coolant is too hot. This is particularly true with systems tuned to low noise – as these systems see the highest coolant temps.
The only thing that matters
The only thing that matters is the prioritization of the components you want to cool. If you want more cooling capacity for your cpu, the cpu block goes first.
It doesn’t matter which block has more or less flow resistance because the loop acts as a system. Liquid in = liquid out.
…believe me, I thought this
…believe me, I thought this was the case too. Then, one time, I had to turn the loop around and I found out is is not the case at all 😉 Sure, temps on the CPU rose about 2C – quite logical, considering it now followed three GPUs, but overall temps were much lower. Test it for yourself if you need prof. You need to know which block has the highest flow resistance though, but like I said earlier – CPU Blocks are your usual suspect.
…believe me, if thats the
…believe me, if thats the case, then your pump is just to small!
I had with my old setup, Quad-CF 5870@950MHz and a 3930K@4.8GHz doing Furmark and OCCT.
I had a temperature drop of 0.7C over the inlet and outlet of my radiator, during normal gaming the drop was between 0.4c and 0.6c.
(From the 8 temp censors i have for my Aquaero 5 Pro, i picked two that ware the closes in temp with 0.1c difference between 20c/70F and 50c/125F, to measure the temp drop/delta)
I use a D5 pump, but the smaller DCC pump is for most setups more then enough, or any trustworthy pump that dose minimum of 400L/h (100gallon/h).
More flow means a more turbulent flow, means better cooling then a slow laminated flow.
https://www.google.com/search?q=turbulent+laminar+flow
This is just wrong.
This is just wrong. Thermodynamics dictates that it will get to a steady state, so , yeah, a non-issue which order the components are in.
Super cool to see PCPER guys!
Super cool to see PCPER guys!
Why don’t you split the
Why don’t you split the output of the Rad/Res and use a separate pump for each sub-loop? That way you can size the pump to the flow rate you need for the device it’s meant to cool.
So, to be clear, put a Y on the output of the Rad/Res and run each output of the Y to a different device to be cooled. Then put a pump after that. Finally, combine the outputs of the two pumps with another Y which will then feed into the Rad/Res.
One thing I have learned over
One thing I have learned over the years is that setting up the loop order is always hotly debated.
Keep in mind that the visuals
Keep in mind that the visuals shown display how the flow is setup in the All-in-One cooling systems used, and is not supposed to be a guide to optimal setup. As has been stated, there are many differing opinions on loop setup, order, and flow control based on what type of cooling loops you build and how much experience you've had in the hobby.
Thanks all for the feedback though, its great…
Yeah, sorry, realized that
Yeah, sorry, realized that after I posted.
Love the Poseidon card –
Love the Poseidon card – definitely my next upgrade.
Also love the combined CPU block and pump but I do worry about long term effect of vibrations.
Personally I would swap the radiators for a 200×200 rad (bigger cooling surface large fan working more slowly = less noise).
What this shows is how easy it is to build a full WC in a real small space.
Water cooling is not for everyone, it is pricey for often minimal gains but it is fun and that is important
*reservoir, not reservior
(in
*reservoir, not reservior
(in the pictures)
Glad to see some water
Glad to see some water cooling excitement again.
I’m going to throw hard earned water cooling bones into the pack just for fun.
Reservoirs are only a source of leaks (and reduce flow).
1/4″ tubing systems are junk.
Overall performance is only improved by radiator surface area and fan cfm. (draw backs are size limitation problems and noise) Everything else does nothing because any ‘improvement’ is cancelled out by other factors.
Always great to see a liquid
Always great to see a liquid cooling story on PCPER! I think you would agree, Morry, that you’re not a serious hardware enthusiast until you go with a proper cooling system — even if it’s just an AIO cooler! Love reading all your liquid cooling reviews! Also, they should have you on the podcast more. You’re very thorough!
Thanks for the support. I do
Thanks for the support. I do agree that you really aren't hardcore until you go with water cooling, but I think its more of an enthusiast's evolution as they find that air cooling just won't cut it for the o/c and temps they want out of their CPU and GPUs.
At one point, I was doing phase change cooling, but got sick of the condensation issues that ensued – killed a few graphics cards from leaks in the condeser tube going to the CPU. But my water loop is fixing to get upgraded in the very near future – multiple rads, pumps, and blocks on all my GPUs…
The order of the loop isn’t
The order of the loop isn’t important to temps. Over time the entire loop will normalize to a set temp range based on the amount of heat your components generate and how much your water-cooling components dissipate heat. It’s far more important to worry about your order in terms of connection points, tube length, the number of hard turns in your loop, fans, radiator sizes, aesthetics, etc.
There’s a point-counterpoint argument for all the components you get…fast pumps = lots of noise but since you move more water through the rads you may get slightly better overall temps. Low speed high CFM fans might get you a nice quiet system but may not be able to handle cooling the water as well as you’d like…so you get more fans or make them run faster…and that means more PSU power and noise…and so on.
The key is finding the right balance to meet your personal needs. I like a quiet system and will sacrifice on the temp side.
Whether it’s the air through
Whether it’s the air through the radiator or the water being pumped through, the real output of your fan/pump is dependent on the total resistance it’s pushing against. If you want that high airflow/coolant flow you better have the pressure to back it up against the resistance you have in it.
As others have said… I
As others have said… I don’t understand why people don’t split the output of the radiator and have one feed for the CPU and one for the GPU. That will give you the lowest flow restriction and the GPU’s coolant won’t be pre-heated.
I like to mention too.. If you’ve got the space, build a plenum between the fan’s output and the radiator. I like to take junk fans, cut everything out except for the body, and use it as a a plenum. Stack two if I’ve got the space. You kill an axial fan’s flow when you slap it right on to a radiator. A sirocco (squirrel cage) fan, it doesn’t matter.
1:A faster/stronger pump is
1:A faster/stronger pump is not always the answer,it really depends on what you are cooling [1 or more devices].A super fast pump can actually raise temps due to the friction of the pump.
My old system had cpu,chipset,mosfet and 2 gpus on one loop.My pump always ran on low for silence,if i cranked the pump up to screaming lvs the overall temp would remain relatively the same [maybe 1-2 c difference under a full load]
2:Every case /system is different .What works well for one wont necessarily work well for another.
3: block order can be anyway thats convenient for your build.
4: a few degrees higher or lower will make little if any difference on system performance.
As others have said… I