Hi all,
I have tested some different AIO coolers for my Threadripper 3960x build and want to share the results so that others can at least profit from my experiences.
I built a machine using an Asrock Taichi TRX40 motherboard in a Cooler Master H500m case and have tried several AIO coolers for it. The tests were not rigorously scientific but should give at least some idea on the efficiency of these coolers on the Threadripper.
I have tested the following coolers:
Of these coolers only the first has a proper base plate the size of the Threadripper, the others have a round plate covering only part of the IHS. I have tested the coolers at load under Linux by running mprime95 -t and looking at how the temperatures behaved.
I used Conductonaut cooling liquid metal cooling paste, and the BIOS has the default overclock setting for the TRX40.
ML360 results
This is a really nice cooler with a full size Threadripper plate - ready to install.
I set the pump to full speed in the BIOS, and the fans to "silent".
When running the rest the CPU temperature rises quickly, then stabilizes around 75C with occasional peaks to 80C and seldomly 90C. A this temperature the fans are loud but not excessively so. The CPU runs constantly at a clock of 4050MHz which seems to be the base overclock speed set by the bios.
At (mostly) idle, only Chrome running at 15% cpu, the CPU temperature moves between 48 and 58 degrees around every minute. This leads to a very irritating issue: the fan speed changes up and down in more or less the same frequency and that is very noticeably. The ML360 has no liquid temperature sensor, so the pump and the fans are directly connected to the motherboard CPU fan and pump headers. The BIOS can only use either the CPU temperature or the mobo temperature to control fan speed, and by default it uses CPU temperature. As this temperature changes quite a bit over a small amount of time the fans do the same.
Corsair H150i results
This cooler is a problem because to use it with a Threadripper you need to buy a special bracket which is not included with this expensive cooler. It's bad enough that you have to do that, but it's also about 20 euro's and it took three weeks to get it as no shop had it in stock here (in the Netherlands).
The cooler can be controlled from Linux more or less by OpenCorsairLink, a command line utility which works quite well (do remember to run it as root though). I used that to set the pump to max performance.
This is by far the quietest cooler of the three, both at idle and at load. But to be fair none of the others is loud at idle, so even though I aimed to have the least sound possible all three would be fine at idle, but the corsair would sound-wise win at load.
At idle the Corsair ran slightly higher on average than the ML360, between 50 and 60C. At load however it quickly moved to 80C and then to around 90C in the next few minutes. The temperature after that continued to rise as the liquid temperature went up slowly with fans running at max. After about half an hour the temperature got to around 100C and was spiking higher even, and the CPU speed throttled to 3500MHz before that. At that point I stopped the test.
The Corsair has a water temperature sensor and bases its fan speed on that temperature. This leads to very gradual sound increase because that water temp rises and falls relatively slowly, so from an acoustic point of view that is perfect. As said the corsair was the most silent of the three, but I think it had trouble with keeping the liquid cool enough at load.
Kraken X72 results
This one seems to have a slightly bigger base plate than the Corsair. It comes with everything needed to install on a Threadripper which is at least something for its price - this is the most expensive of the lot. Like the Corsair there is an open source project to control and get information on the cooler under Linux: Liquidctl - thanks a lot to all those people doing the work that these companies do not.
Idle temperatures are much like the Corsair, between 50C and 60C. But it performed better at load, going to around 87C in the first 20 minutes with slowly rising water temperatures. After 20 minutes the CPU temperature jumped to 94C and the CPU speed dropped to 3900MHz (from 4050). At this point both water temperature (around 48C) and CPU temperature stayed stable.
Like the Corsair this has a water temperature sensor too, and as such has the same gradual fan speed/noise response as the Corsair, but in general it is just a bit louder (mostly at high speeds).
Conclusions
The ML360 is by far the best performing cooler here. I think a lot of that has to do with the threadripper size base plate; even though the other coolers' base plate would mostly cover the Threadripper die positions I think the CPU has definite trouble getting rid of its heat with the round plates.
The ML360 has no internal control logic and no water temperature sensor, so it depends on the usual motherboard fan/pump connectors to handle its performance. But because the CPU temperature changes in a completely different way that the water temperature this leads to irritating and not very useful behaviour of the fans.
This is the cheapest of the coolers, btw, probably because that lack of internal logic.
The Corsair was quiet but performed badly under load. I also found it very expensive for what it does, and having to buy that stupid metal bracket for an extra 20 is just ridiculous.
The Kraken had reasonable performance but it was still way worse than the ML360. But at least it was able to keep the temperature high but under control.
It's a shame that there are no other (reliable) Threadripper size coolers on the market, because these results do seem to indicate that the base plate has a large influence on performance. Let's hope that more of them become available soon. For now I will go back to the ML360, even with the highly irritating fan sound variation..
I have tested some different AIO coolers for my Threadripper 3960x build and want to share the results so that others can at least profit from my experiences.
I built a machine using an Asrock Taichi TRX40 motherboard in a Cooler Master H500m case and have tried several AIO coolers for it. The tests were not rigorously scientific but should give at least some idea on the efficiency of these coolers on the Threadripper.
I have tested the following coolers:
- Cooler Master ML360 TR4 edition
- Corsair H150i Pro RGB
- NZXT Kraken X72
Of these coolers only the first has a proper base plate the size of the Threadripper, the others have a round plate covering only part of the IHS. I have tested the coolers at load under Linux by running mprime95 -t and looking at how the temperatures behaved.
I used Conductonaut cooling liquid metal cooling paste, and the BIOS has the default overclock setting for the TRX40.
ML360 results
This is a really nice cooler with a full size Threadripper plate - ready to install.
I set the pump to full speed in the BIOS, and the fans to "silent".
When running the rest the CPU temperature rises quickly, then stabilizes around 75C with occasional peaks to 80C and seldomly 90C. A this temperature the fans are loud but not excessively so. The CPU runs constantly at a clock of 4050MHz which seems to be the base overclock speed set by the bios.
At (mostly) idle, only Chrome running at 15% cpu, the CPU temperature moves between 48 and 58 degrees around every minute. This leads to a very irritating issue: the fan speed changes up and down in more or less the same frequency and that is very noticeably. The ML360 has no liquid temperature sensor, so the pump and the fans are directly connected to the motherboard CPU fan and pump headers. The BIOS can only use either the CPU temperature or the mobo temperature to control fan speed, and by default it uses CPU temperature. As this temperature changes quite a bit over a small amount of time the fans do the same.
Corsair H150i results
This cooler is a problem because to use it with a Threadripper you need to buy a special bracket which is not included with this expensive cooler. It's bad enough that you have to do that, but it's also about 20 euro's and it took three weeks to get it as no shop had it in stock here (in the Netherlands).
The cooler can be controlled from Linux more or less by OpenCorsairLink, a command line utility which works quite well (do remember to run it as root though). I used that to set the pump to max performance.
This is by far the quietest cooler of the three, both at idle and at load. But to be fair none of the others is loud at idle, so even though I aimed to have the least sound possible all three would be fine at idle, but the corsair would sound-wise win at load.
At idle the Corsair ran slightly higher on average than the ML360, between 50 and 60C. At load however it quickly moved to 80C and then to around 90C in the next few minutes. The temperature after that continued to rise as the liquid temperature went up slowly with fans running at max. After about half an hour the temperature got to around 100C and was spiking higher even, and the CPU speed throttled to 3500MHz before that. At that point I stopped the test.
The Corsair has a water temperature sensor and bases its fan speed on that temperature. This leads to very gradual sound increase because that water temp rises and falls relatively slowly, so from an acoustic point of view that is perfect. As said the corsair was the most silent of the three, but I think it had trouble with keeping the liquid cool enough at load.
Kraken X72 results
This one seems to have a slightly bigger base plate than the Corsair. It comes with everything needed to install on a Threadripper which is at least something for its price - this is the most expensive of the lot. Like the Corsair there is an open source project to control and get information on the cooler under Linux: Liquidctl - thanks a lot to all those people doing the work that these companies do not.
Idle temperatures are much like the Corsair, between 50C and 60C. But it performed better at load, going to around 87C in the first 20 minutes with slowly rising water temperatures. After 20 minutes the CPU temperature jumped to 94C and the CPU speed dropped to 3900MHz (from 4050). At this point both water temperature (around 48C) and CPU temperature stayed stable.
Like the Corsair this has a water temperature sensor too, and as such has the same gradual fan speed/noise response as the Corsair, but in general it is just a bit louder (mostly at high speeds).
Conclusions
The ML360 is by far the best performing cooler here. I think a lot of that has to do with the threadripper size base plate; even though the other coolers' base plate would mostly cover the Threadripper die positions I think the CPU has definite trouble getting rid of its heat with the round plates.
The ML360 has no internal control logic and no water temperature sensor, so it depends on the usual motherboard fan/pump connectors to handle its performance. But because the CPU temperature changes in a completely different way that the water temperature this leads to irritating and not very useful behaviour of the fans.
This is the cheapest of the coolers, btw, probably because that lack of internal logic.
The Corsair was quiet but performed badly under load. I also found it very expensive for what it does, and having to buy that stupid metal bracket for an extra 20 is just ridiculous.
The Kraken had reasonable performance but it was still way worse than the ML360. But at least it was able to keep the temperature high but under control.
It's a shame that there are no other (reliable) Threadripper size coolers on the market, because these results do seem to indicate that the base plate has a large influence on performance. Let's hope that more of them become available soon. For now I will go back to the ML360, even with the highly irritating fan sound variation..