I did set up my secondary system with a 5800x and used an AIO while making sure everything is working before setting up my loop. I decided to do some thermal testing to see what all the comments about AIOs loosing out to air coolers are about. I have a Noctua NH-U12A around that is for a future nearly silent build which I tested against the Corsair h150i Pro XT. I also did the measurements on my loop once it was finished. For a lot of people this will most likely be TL;DR, but I do hope some find it interesting.
Mounting:
Case used was the Fractal Design Meshify 2 with stock fans. It has 2x140mm in the front and 1x140mm at the rear. Fans are OK and are quite quiet but do not move as much air at the same RPM as Noctua NF-A14 or other high quality fans.
The AIO was top mounted and the NH-U12A was mounted normally with stock case fans. For my loop I did a front mount so the front case fans were removed. The results from the loop aren't directly comparable for the given noise level as there are no front case fans, but it gives an indication. Drawbacks to having no case fans is that it has to pull air through the mesh compared to having case fans creating positive pressure, but it can also run the fans higher for a given noise level.
I applied a thin spread of NT-H2 paste and tried to get the mounting as optimal as possible. For the Pro XT I pulled the cooler as far to the bottom as possible before tightening the screws. There is some wiggle room and the fin stack doesn't cover the entire heat spreader so getting the mount as offset as possible should give better temps. The NH-U12A only mounts one way so just tightened all screws in a crisscross pattern with aprox 2 revolutions on each screw until they stopped. The mounts where checked upon removal and showed a very thin spread of paste across the entire IHS.
Noise and fan setup:
To avoid brute force i decided to limit it to a certain noise level to 38dbA (quite loud) using a db-meter to measure as phone apps are very inaccurate. The Pro XT will surpass 50dbA if the fans are allowed to run at full speed. I placed the computer under the desk and the db meter on top of the desk and adjusted fan speed until the target was hit. This is approximately equivalent to measuring 1m from the front of the case. The loop had slightly higher fan speeds due to having 2 case fans less. I also ran the PRO XT on auto to see a more realistic use case, but didn't measure noise as it is variable and. I also ran a 32dbA result for my loop.
Testing: I ran Cinebench r23 on a loop and made sure the cooler had hit stable state before taking a 2+ min measurement and reading the average temp and also checking that the average watt had low variance. A 1 watt difference is about 0.3-0.4 degrees Celsius and the measured package power was within .8 watts difference over a 2 minute period and the average run to run variance with the same watt is less then 0.2 degrees given the same ambient. Ambient was at 22.1-22.4. Overall I would estimate the margin of error to be around +/- 1 degree. The exception is the PRO XT that would have to be mounted the same way every time due to large tolerances in how it can be mounted. The effective boost is just an indication on how the PB behaves with different temps, it can vary slightly based on different parameters under the same load and die-average temps.
Results
Pro XT (38dbA):
Radiator fans: 1520RPM
Pump speed: extreme (2700rpm)
Die average temp: 73.1
Effective boost clock: 4600mhz
NH-U12A (38dbA):
Cooler fans: 2110rpm
Die average temp: 78.4
Effective boost clock: 4576mhz
Pro XT Auto:
Radiator fans: extreme profile (approx 1300rpm average and most likely around 35-36dbA)
Pump speed: extreme (2700rpm)
Die average temp: 77.2
Effective boost clock: 4580mhz
Custom loop (38dbA front mounted):
Radiator fans: 1780rpm
Pump speed: 4800rpm
Die average temp: 68.6
Effective boost clock: 4615mhz
Custom loop (32dbA front mounted):
Radiator fans: 1190rpm
Pump speed: 3000rpm
Die average temp: 73.7
Effective boost clock: 4596
Conclusion
First off i would say that 38db(A) is quite loud. Unless you use a DB meter you will not get accurate measurements. I do try to limit my systems to 35dbA or less under load and around 30dbA or less on idle.
The difference in absolute cooling capacity, at a fixed noise level, between the AIO and the air cooler is quite significant. It does however shrink to insignificant once you start using corsairs auto fan curves. With the type of fan curves I would be running then the NH-U12A and PRO XT are almost within margin of error in cooling and fairly close when it comes to noise as I would be running 100% fan speed on the air cooler once temps go above 75 degrees and fans on AIO at extreme. There is a good chance that I would set the AIO pump speed to quiet (2000rpm) for low noise idle, as I find the pump to be slightly annoying at extreme and balanced so they would be approx. equal in cooling capacity in a real world scenario. I didn't test balanced, but it should be 3-4 degrees worse than extreme in this scenario so it would then lose out to the air cooler.
The larger Noctua NH-D15 should be somewhere between 1-3 degrees better than the NH-U12A, and would beat the Pro XT on extreme profiles, but I haven't owned one or noise normalized it so it is guesswork from my side.
The Corsair h150i PRO XT manages fan speed based on water temp and it lets water temp rise. When you set it to extreme mode it will wait until the water temp is quite high before it spins up the fans. This leads to higher temps than the unit is capable of so making it step up the fans slightly faster would give it an a small advantage.
Custom loop is there for reference as that is what I am using on the system, but it is set up slightly different. The main advantage comes when trying to keep the system quiet as the custom loop achieves almost the same cooling performance as the AIO at 1/4 of the sound intensity, but it is also a bit better when running higher noise levels..
Mounting:
Case used was the Fractal Design Meshify 2 with stock fans. It has 2x140mm in the front and 1x140mm at the rear. Fans are OK and are quite quiet but do not move as much air at the same RPM as Noctua NF-A14 or other high quality fans.
The AIO was top mounted and the NH-U12A was mounted normally with stock case fans. For my loop I did a front mount so the front case fans were removed. The results from the loop aren't directly comparable for the given noise level as there are no front case fans, but it gives an indication. Drawbacks to having no case fans is that it has to pull air through the mesh compared to having case fans creating positive pressure, but it can also run the fans higher for a given noise level.
I applied a thin spread of NT-H2 paste and tried to get the mounting as optimal as possible. For the Pro XT I pulled the cooler as far to the bottom as possible before tightening the screws. There is some wiggle room and the fin stack doesn't cover the entire heat spreader so getting the mount as offset as possible should give better temps. The NH-U12A only mounts one way so just tightened all screws in a crisscross pattern with aprox 2 revolutions on each screw until they stopped. The mounts where checked upon removal and showed a very thin spread of paste across the entire IHS.
Noise and fan setup:
To avoid brute force i decided to limit it to a certain noise level to 38dbA (quite loud) using a db-meter to measure as phone apps are very inaccurate. The Pro XT will surpass 50dbA if the fans are allowed to run at full speed. I placed the computer under the desk and the db meter on top of the desk and adjusted fan speed until the target was hit. This is approximately equivalent to measuring 1m from the front of the case. The loop had slightly higher fan speeds due to having 2 case fans less. I also ran the PRO XT on auto to see a more realistic use case, but didn't measure noise as it is variable and. I also ran a 32dbA result for my loop.
Testing: I ran Cinebench r23 on a loop and made sure the cooler had hit stable state before taking a 2+ min measurement and reading the average temp and also checking that the average watt had low variance. A 1 watt difference is about 0.3-0.4 degrees Celsius and the measured package power was within .8 watts difference over a 2 minute period and the average run to run variance with the same watt is less then 0.2 degrees given the same ambient. Ambient was at 22.1-22.4. Overall I would estimate the margin of error to be around +/- 1 degree. The exception is the PRO XT that would have to be mounted the same way every time due to large tolerances in how it can be mounted. The effective boost is just an indication on how the PB behaves with different temps, it can vary slightly based on different parameters under the same load and die-average temps.
Results
Pro XT (38dbA):
Radiator fans: 1520RPM
Pump speed: extreme (2700rpm)
Die average temp: 73.1
Effective boost clock: 4600mhz
NH-U12A (38dbA):
Cooler fans: 2110rpm
Die average temp: 78.4
Effective boost clock: 4576mhz
Pro XT Auto:
Radiator fans: extreme profile (approx 1300rpm average and most likely around 35-36dbA)
Pump speed: extreme (2700rpm)
Die average temp: 77.2
Effective boost clock: 4580mhz
Custom loop (38dbA front mounted):
Radiator fans: 1780rpm
Pump speed: 4800rpm
Die average temp: 68.6
Effective boost clock: 4615mhz
Custom loop (32dbA front mounted):
Radiator fans: 1190rpm
Pump speed: 3000rpm
Die average temp: 73.7
Effective boost clock: 4596
Conclusion
First off i would say that 38db(A) is quite loud. Unless you use a DB meter you will not get accurate measurements. I do try to limit my systems to 35dbA or less under load and around 30dbA or less on idle.
The difference in absolute cooling capacity, at a fixed noise level, between the AIO and the air cooler is quite significant. It does however shrink to insignificant once you start using corsairs auto fan curves. With the type of fan curves I would be running then the NH-U12A and PRO XT are almost within margin of error in cooling and fairly close when it comes to noise as I would be running 100% fan speed on the air cooler once temps go above 75 degrees and fans on AIO at extreme. There is a good chance that I would set the AIO pump speed to quiet (2000rpm) for low noise idle, as I find the pump to be slightly annoying at extreme and balanced so they would be approx. equal in cooling capacity in a real world scenario. I didn't test balanced, but it should be 3-4 degrees worse than extreme in this scenario so it would then lose out to the air cooler.
The larger Noctua NH-D15 should be somewhere between 1-3 degrees better than the NH-U12A, and would beat the Pro XT on extreme profiles, but I haven't owned one or noise normalized it so it is guesswork from my side.
The Corsair h150i PRO XT manages fan speed based on water temp and it lets water temp rise. When you set it to extreme mode it will wait until the water temp is quite high before it spins up the fans. This leads to higher temps than the unit is capable of so making it step up the fans slightly faster would give it an a small advantage.
Custom loop is there for reference as that is what I am using on the system, but it is set up slightly different. The main advantage comes when trying to keep the system quiet as the custom loop achieves almost the same cooling performance as the AIO at 1/4 of the sound intensity, but it is also a bit better when running higher noise levels..
Last edited: