Jump to content
Corsair Community

Confused about h115i Platinum cooler and fans speeds


kendallb
 Share

Recommended Posts

Hi Guys,

 

I just built my first rig in a long time and back when I used to build rigs we never had water cooling. I have had a water cooled PC before, as the one that this replaces was an Alienware Aurua R7. That PC had a 120mm liquid cooling AIO installed and part of the reason I wanted to build my own was to get some cool RGB as well as make it quieter. The Alienware would spin up and sound liek a vaccuum cleaner under full CPU load, and when you put full graphics load on it the 2080ti from Dell also spins up super load (oddly it only has a single fan?).

 

Anyway my new rig is as follows:

 

- NZXT H150 Elite

- AMD Ryzen 9 3900X

- Corsair Force Series MP600 1TB

- Corsair h115i Platinum 280mm RGB cooler mounted to the front as intake (fan RGB facing out)

- Corsair Vengeance RGB Pro 32GB memory (2x16GB)

- Stock case fans as exhast on back (12mm) and top (140mm)

- Corsair Commander Pro connected to stock case fans

- Four temp temp sensors in various areas of the case corners

- Windows 10 Pro 1909

 

What I am confused about is exactly how these fans are supposed to be controlled because the case is always super quiet when I run iCUE and the fans are set to the Quiet mode. Which is exactly what I want, but what I don't understand is that when I run the CPU at full load using CPU-Z or Cinebench, the fans never really change speed. They just sit there not changing much, while my old Dell would spin up like crazy.

 

Checking the CPU temps with Ryzen Master I can see that under full load the CPU is getting a little over 80 degrees, but according to iCUE the h115i cooler is not changing much in temp. The cooler sits about 30.30C degrees at idle and the CPU temp is in the very low 40's when doing nothing. I I leave CPU-Z running for a long time the cooler temp will rise a bit, but not much. Maybe 2-3C.

 

Given that, since the cooler temp is not going up much, the fans really dont change in speed. If I create my own custom profile for the cooler and use the quiet mode it's at 30% speed at 31 degrees and goes up to 38% at 34C, but the cooler never really gets that high. I have had it running while posting this and my cooler is now at 33.90 degrees and CPU temp is at 81.59 degrees.

 

But the thing is, if I spin up the fans to and pump to maximum, it does bring the coolant temp and CPU temps down a bit. So my question is, why is the stock fan profile not spinning the fans up much when the CPU is under load? It would seem in my old Dell the cooler fans spun up based on CPU load, but with the h115i they are tied to the cooler temp and I don't see any way to tied it to the CPU temp (the Ryzen 9 sensor does not show up in the list).

 

I now have the fans running at full speed and the cooler temp is now dropping back down to the low 32's and the CPU temp is now 77C, so clearly it makes a difference. But the stock fan profile is not doing a good job of knowing when to spin up the fans?

 

Also the case fans are not really spinning either.

 

Now yesterday I was playing around with overclocking and the CPU was getting up to 90 degrees, but once again the coolant temp was not going up a whole lot, and hence the fans were not spinning. If I manually spin them up to full speed, then the temps on the CPU go down quite a bit.

 

So, my specific questions are:

 

1. Why is there now way to tie the fan speed of the h115i to the CPU temps, so when I have it at full load they will be keeping it cool? Ryzen 9 is not an option in the drop down for a custom curve.

 

2. Why does the stock fan profile not really ramp up the speed until the cooler is at 44C, which is a temp my cooler just never seems to ever get close to? If it got that high, my CPU would be frying.

 

3. What controls the case fans I have attached to my Command Pro in the stock quiet mode profile? Are they attached to the h115 coolant temp, the temp probes I have installed or something else? They also cannot be tied to CPU temps either.

Link to comment
Share on other sites

You are working the same problem, but from opposite ends with the Commander Pro and Platinum.

 

Commander Pro - The three presets in the Performance tab use CPU temp as the control variable. This is a terrible control variable because modern CPUs are very reactive and of course blasting your front intake fans do nothing to reduce the heat created when voltage meets the bottom of the CPU at the pins in the socket. So why does it use it? It's the only control variable every system is guaranteed to have. Immediate replacements are GPU temp or the H115i Platinum Temp (coolant temperature), but of course there is no requirement a system have a GPU or certainly a specific liquid cooler. The best case fan control is ultimately going to be data obtained from the temp sensors for the Commander Pro. You stick one at the rear exhaust or on the back side of the H115i radiator to measure exit air temp. That is the number you are hoping to control with case fans.

 

H115i Platinum (or any other "i" cooler) - Fan speed based on coolant temperature. Why? That's how the cooling actually works. People have been conditioned to expect fan speed based on CPU temp from years of using air coolers. Air coolers work this way, because they don't have another option. They run faster and more often than necessary because there isn't a good way to control it otherwise. Regardless of whether you have an air cooler, a tower, an AIO, a custom loop, or some exotic system, the heat for the CPU is created when the voltage is applied at the pins. That heat must pass through the CPU to exit. This first stage of "cooling" is conductive. CPU pins -> CPU material => TIM -> cold plate of whatever cooler is attached. There is no control for this rate of heat transfer and the end result is entirely based on the level of voltage and conductive properties of the materials involved. This is why you see people bashing Intel over things like solder vs TIM applied heat plates and why so many ran to delid Coffee Lake CPUs for a -25C reduction in temperature. However, gains like that are rare, most materials are all relatively similar with the same rate of conductivity. That makes voltage the heavy hand.

 

If I give you three cooling options (a tiny bolt air fan, an AIO, and wall sized custom external cooling panel), all three will give you the exact same CPU temp when starting CPU-Z or anything else... at least for the first 5-10 seconds. It's what happens after that distinguishes one cooling type or cooler model from another. The heat transfer starts between CPU and cooling device. This is the second stage of cooling. The cooling device must get rid of that heat in some way. If it does not, then it will transfer heat back to the CPU. That conductive plate allows transfer both ways. The best way to explain that for your situation is coolant temp (H115i Temp) is the minimum possible CPU temperature. If the coolant on one side of the cold plate is 30C, then it is not possible for the CPU to be less than that. When at 0 volts, the CPU will still be 30C because the heat from the coolant is transferred back in. Of course, you don't get to be a 0 volts unless you are off -- or in the microseconds of modern CPU power management where it creates nearly the same condition. However, understanding that coolant temp is base temperature is critical to understanding the next part about the fans.

 

Water/coolant flows to the radiator, fans blow cooler air across the fins, heat is released to... whatever is on the other side. Your question was why don't the fans ramp up until later in the temperature scale. The answer is noise vs cooling. I have the perfect cooling curve for you. Fixed fan speed, 2000 rpm. That is the maximum speed of the fan on that model and insures you are always getting maximum results. Of course, no one could stand that and it is a bad trade for almost every one. Remember the coolant temp is the baseline CPU temp and the coolant temperature rise reflects the amount of heat in the cooling system. So back to the three cooler type example. The little air fan will quickly start to take on more heat into the physical radiator than it can shed. That warms the metal itself and that heat raises the baseline CPU temp. Same thing on the AIO, but the values are easy to see. +3C coolant rise = +3C CPU temp rise. The AIO can handle more wattage than the small air tower both in terms of air vs water heat capacity and has more radiator to disperse the heat. It will have a smaller cooling system rise compared to the air tower. Then there is the wall sized panel/radiator. The wattage is still the same. The CPU is still the same. When you start Prime 95 or whatever and the CPU hits 90C, it will still be same on all coolers. The difference is the others will warm slightly in time. In this extreme example, the wall sized cooler will never get 1C warmer than the first seconds. It is able to dump all the wattage transferred in on a single pass through the system. The coolant is returned to the lowest possible state before it enters the CPU block again.

 

What this means is even with a wall sized cooler, you still can't run Prime 95 at 1.80v. You still zap the CPU at the heat creation point (pins) the moment the full voltage comes on line. This is where we are all limited. We can make bigger coolers, but it still doesn't solve the heat at the pins problem. This is why extreme overclockers use liquid nitrogen. They still have to deal with a huge +70-90C spike the moment the voltage comes on, but they can lower the starting point of the CPU temp to -20C so that it can survive the increase.

 

Back to the practical application of your H115i Platinum. When you see the H115i temp change from 30 to 33C, that 3C rise is the most you can possible reduce the CPU temp at any fan speed. It would be nice to reduce it to zero, but that is not going to be sustainable with any AIO. However, you have to decide if it is worth it to run the fans at 2000 rpm to reduce the CPU temp by 1-2C. That is all the fans can do. Reduce coolant temp back to the level you started. They will never affect the heat created at the pins. You must attack that with voltage settings and whatever else is in the BIOS. That is getting to be more complicated these days as board makers try balance winning comparison benchmark based reviews vs making sure even a weak CPU boots up. The cover all solution --- more voltage --- and that means more CPU heat that cannot be conducted away fast enough.

 

The last part is about setting up your own fan curve and is sometimes the more difficult part to explain. The preset curves on the H115i Platinum are no overly helpful and the same is true of every other AIO on the market. This is because we all have different room and case temperatures. That is what governs the minimum coolant temp sitting at the desktop. Mine will sit at 20C in my cold Winter office. Then at 29C all afternoon as the Summer sun bakes the glass on this side of the building. That 9C swing is the same as a maximum load of 500W. My Winter fan curve would be blasting at the desktop with no possible hope of making the coolant below the room temperature. Meanwhile the Summer curve would never move a single RPM in Winter since the peak values are the same as my normal idle. You need to create your own temp curve and it may need to be seasonally adjusted. The idle coolant temp is both cooler, CPU, and system layout affected in addition to the room temp baseline. That means every one really does have a unique set-up. You have to learn its behavior. The included curves were likely designed to cover an average environment with an ear toward noise. I suspect inherently 'loud' coolers don't sell as well.

 

It seems like your baseline is 30C right now. Set that speed to a comfortable fan speed for desktop work. For most people, that will be between 500-700 rpm, but this is user choice. Somewhere about 3C down the line, start sloping up. Don't program your fan curves for a stress test. If you want to run a stress test, set a fixed fan rate of whatever. Some people like to run max. I typically test 140s at around 1200-1300, the most I could stand for normal use. The best way to determine your normal maximum coolant temp is to use it the way you normally would. If that's renders, do that. If its gaming or CPU number crunching, do that. Find the peak level, set a tolerable fan speed to match. Set the 2000 rpm blast about 4-5C above the highest coolant temp you see. That will be a clear and audible warning if things are out of bounds.

 

It's not as complex as it sounds. The more difficult task you are currently faced with is why the instant on CPU temp is so high. That is not usually a cooler issue unless it doesn't make good physical contact. If that were true, use would be problematic for everything, including opening this browser. Unfortunately, some MB makers set things so high that it almost looks like that. When the Intel Kaby Lake series came out, several of them told users to stop running synthetic benchmarks. Their CPUs were not designed for it [sic]. That's kind of where we are in terms of setting up the BIOS these days.

 

It is possible to tie some of these fan functions to CPU temp, although I strongly suggest you do not. Besides the reasons explained above, I do not think iCUE 3,23 can read your AMD 3000 series CPU temp. They are waiting for CPUID to update the SDK, I believe that has happened, and it should be in next month's release. I still suggest you not use them. The end result is fans that ramp up and down like crazy while providing no additional cooling benefit.

Edited by c-attack
Link to comment
Share on other sites

Thanks for the detailed info! I guess I should sit down and make a reasonable fan curve for my AIO cooler and like you said, probably set the max fan speeds I would see in normal use to the max I can stand, then full speed when something is clearly wrong.

 

Anyway as for the Command Pro which controls the case fans, given that iCUE 3.23 does not support reading the Ryzen 3000 series CPU temp (it's not available as an option), what exactly is it using to program the fans? Honestly it seems as though the case fans just sit at idle all the time no matter what the CPU is doing, so I am thinking it is trying to tie it to CPU temps, but the sensor is not there so they just sit idle.

 

Since I do have the four temp sensors installed, I need to document where I put them and which port they are in. One of them is on the bottom of the case at the rear close to the GPU, the other is next to the rear exhaust fan, the other is right behind the AIO cooler inside the case and the other I think I stuck down by the hard disk caddy (have to check). Sounds like the one I should use is the one behind the AIO cooler fans and make a custom fan curve based on that.

 

I would assume that if I did not have the Commander Pro (I got it as I have some LED fans to replace the stock case fans and some RGB strips to install :) ), you would hook the case fans to the motherboard fan channels, and those would probably be tied to CPU temps?

Link to comment
Share on other sites

Also how do the pump profiles work? It appears it's not possible to set your own profile there, so what would people normally set it to? Honestly I put the pump on extreme mode vs quiet mode and can't really tell the difference? Not sure I can even hear it, even over slow 450rpm fan speeds?
Link to comment
Share on other sites

Anyway as for the Command Pro which controls the case fans, given that iCUE 3.23 does not support reading the Ryzen 3000 series CPU temp (it's not available as an option), what exactly is it using to program the fans? Honestly it seems as though the case fans just sit at idle all the time no matter what the CPU is doing, so I am thinking it is trying to tie it to CPU temps, but the sensor is not there so they just sit idle.

 

I would assume that if I did not have the Commander Pro (I got it as I have some LED fans to replace the stock case fans and some RGB strips to install :) ), you would hook the case fans to the motherboard fan channels, and those would probably be tied to CPU temps?

 

 

Excellent question and we were discussing this the other day. Short answer is I don't know either. However, what you are reporting confirms my theory that is sees a null value and interprets the CPU temp as 0C or some other nominal value. When it is tied to CPU in the default state (Quiet mode), if you open a folder on the desktop, you will hear all of your fans ramp up instantly. I have a 6 and 9 fan pair of systems. It's annoying any time I create or import a new profile. First thing I need to do is immediately set the fans. If you are not hearing that, then there is no value on the other end.

 

 

 

Also how do the pump profiles work?

 

This part is a little confusing. The three presets (Quiet/Balanced/Extreme) each have different meanings on different devices. On the C-Pro they are tied to CPU temp and invisible. I have no idea what they are. The three shape tools above the right corner of the graph do not correspond to that. Those are copies of the water cooling Q/B/Ex presets. If you use it, notice the temps all max out at 40C, typical of coolant temp. Those are the same as the ones in the H115i Platinum device. There are some minor tweaks for smaller coolers or those with higher/lower speed fans. For the pump, those three presets represent a Low/Medium/High Fixed speed. They have tweaked the Platinum pump speeds several times, but they are approximately 1800-1900 for Quiet, 2400-2500 Balanced, Extreme 2800-3000. The pump will stay in the mode selected.

 

Flow rate is not a critical element of cooling on an AIO systems with relatively short length and one block and radiator. It can be more important on multi-component custom systems, but even then I don't run more than 2000 rpm on my 2x360, CPU+VRM+GPU build with way too many right angles. You would think more trips to the block and radiator to release heat in a given unit of time would be better, except that less time in the block means less heat picked up and less time in the radiator means less heat released. In a best case scenario, you can pick up some tenths here or there, but it is extremely hardware specific. You can go too slow, but not on the Platinum series. You can park on Quiet and leave it. Certainly experiment with the different speeds for load and see how things turn out. My H115i Pro never left the balanced state, which was 2160 rpm. If you can't hear the pump on Extreme, fantastic. That means you got a good one. Most people will hear the hum pretty clearly, but not over moderate fan speed.

Link to comment
Share on other sites

You are working the same problem,

 

...

 

 

Just wanted to say thanks c-attack for this incredibly detailed work. I think I knew most of the stuff you covered, but it was a real pleasure to see it so well laid out.

 

(Its also brought home to me that I need to go and skill up on how to use iCue better !!!)

Link to comment
Share on other sites

H115i Platinum (or any other "i" cooler) - Fan speed based on coolant temperature. Why? That's how the cooling actually works. People have been conditioned to expect fan speed based on CPU temp from years of using air coolers. Air coolers work this way, because they don't have another option. They run faster and more often than necessary because there isn't a good way to control it otherwise. Regardless of whether you have an air cooler, a tower, an AIO, a custom loop, or some exotic system, the heat for the CPU is created when the voltage is applied at the pins. That heat must pass through the CPU to exit. This first stage of "cooling" is conductive. CPU pins -> CPU material => TIM -> cold plate of whatever cooler is attached. There is no control for this rate of heat transfer and the end result is entirely based on the level of voltage and conductive properties of the materials involved.

 

I want to come back to this, as the more I test this and play with it, the more I think this method is not optimal and could be improved. More on that later.

 

Firstly however I am surprised to not see a much more dramatic increase in coolant temperature when the CPU gets hot. Although I think part of what you are saying is that the measurement of the CPU temp I see in Ryzen Master is not necessarily the measurement of the temperature at the CPU cooling plate that bolts to the h115i cooling head? Is that correct? Because if the cooling head has 80C directly attached to it, I would expect the coolant temperature to rise rapidly, but it is not. What would be considered as reasonable rise in coolant temperature in practice? I am only really seeing 3-4C total rise as mine never seems to go over 34C. Is that normal? Or should I consider tossing the stock thermal paste and redoing it with something better?

 

Some might consider a low coolant temperature to be a good thing, but I think its the opposite? If the coolant temperature is not rising much when the CPU is under load, then it is not doing a very good job at all of moving the heat away from the CPU and into the radiator. In order to properly dissipate the head from the CPU to the air, the temperature needs to transfer to the coolant, which is pumped the radiator and then released to the air via the fans blowing cooler air over the radiator. The hotter the radiator coolant is relative to the ambient temperature, the more heat is released and hence the more efficient the device is cooling.

 

So why exactly is the coolant not going up all that much in temperature? From my testing I have noticed that the CPU temp starts to rise relatively quickly under load from it's initial load value, but once the coolant finally starts to rise and the fans kick in, then the temps actually go down. Which is because there is better head transfer away from the CPU once the coolant is hotter. I guess this all depends on where exactly the coolant temperature is measured as well (in the head block, at the intake of the radiator or at the outlet of the radiator?).

 

Anyway getting back to the controlling the fans based on coolant temperatures, ignoring the fact that the coolant temp does not seem to rise much, if we base fan speeds on coolant temperature, because it takes so long to react to CPU load, the entire process is extremely delayed. The fans won't spin up for a long time after CPU load starts, and as mentioned above CPU temps will actually go up quite a bit until the coolant and fans start working, and then it comes down again.

 

If we control the fan speed based on CPU temperature, then the fans will be asked to ramp up as soon as there is load on the CPU and already be providing better cooling capability for when the coolant does start to rise in temp, long before the point it does if we run the fans based on coolant temperature.

 

I know it has been stated by you and others in other forums I have read, that modern CPU's see drastic temperature spikes all the time, and hence that's why basing the fans on CPU temp is considered a bad idea. You can clearly see that if you just leave Ryzen Master open and monitor the temps while you do stuff so I think that argument is sold. However, at least with the fans I have in my case controlled via iCUE, they actually take a while to ramp up to speed, so would inherently act to smooth out the CPU temperature spikes naturally.

 

However what I don't get, it why we don't simply have the option to control the fan speeds based on average CPU temperatures? The easiest way to smooth out the spikes you see in CPU temps when you say, open windows explorer or launch chrome, would be to smooth it out and use the average temperature over a period of time, like 3s, 5s, 10s or whatever. That would eliminate all the spikiness from the CPU temperatures, and would allow you to ensure the fans are running and doing their job in advance of when the cooling is needed. You could fine tune the time window used to average the CPU temperature out to adjust how quickly the fan speeds ramp up, so small spikes won't budge them much at all, but prolonged CPU load could be producing your 'max' acceptable fans speed much more quickly than coolant temperature would. And it would likely ramp them down a lot quicker also, as what I have seen in my testing is that once the fans finally do get up to 1200-1300 rpms, when the load goes away they keep spinning at that speed for a long time when in reality they could be running slower as the load is gone.

 

Also is stands to reason that perhaps average CPU temperature might be a better control variable as a default out of the box profile, since it's a lot less likely to be affected by ambient room temperature than coolant temperature is?

 

Alas I can't test any of my theories here since the Ryzen temperature is not a control variable currently in iCUE :(

 

Who would we ask to add the option to use average CPU temperature rather than instant CPU temperature in iCUE? It would be simple to add that as a feature :)

Link to comment
Share on other sites

A few traps to avoid getting entangled in. First, the temperature of your CPU represents the amount of energy available to be transmitted into the cooler. It does not. The CPU temp is the current physical state of the little silicon materials inside the CPU. How much energy there is usually represented in watts, and those watts have a specific relationship to the amount of energy required to raise a given unit of water by 1C. Your CPU is not a perfect conductor nor will it ever be. It will always retain heat. The example I used give is this: I ask you to hold out your hand. I produce a blow torch. You are understandable nervous. I explain, "Don't worry, I am going to put a block of ice in the palm your hand. When I touch the flame to your skin, it won't be so bad because the ice will cool the other side." Horrific and I won't get any takers for that offer. Remember, cooling is not cooling. It is transferring the heat somewhere else. With the cooler, we are transferring as much heat as we can out of the CPU. That rate is fixed and unalterable, but it must also go through the CPU. Your decisions revolve around how to manage the heat removal from the cooler, not the CPU. I know that sounds slightly evasive, but it is critical to understanding why the fans don't need to react to the CPU temp changes.

 

Perhaps the better way to illustrate this is to take a variable load stress test (like AIDA, OCCT, or Intel XTU) and continually stop and start, stop and start. Your CPU temps will ping pong all over. The coolant will never change. Your are lighting the flame, but turning it off before heat builds up. The silicon feels it because it is thermally sensitive, but there isn't any meaningful wattage for more than a moment. Heat build up requires time.

 

Now look at the problem from the other side. Take a fixed CPU load test like that of CPU-Z or any Linpack. Set your fans to 500 rpm fixed. Initiate the test. Record the before CPU temp and that of 100% load 5 seconds later. Now do the same thing with the fans set to 1000/1500/2000 rpm, whichever you like. The before and after results will be the same. Fan speed does not prevent CPU temp spikes. Nothing can do that on the cooler end. The only difference the fans can make is to prevent heat from accumulating in the coolant system. That change is always slow. Even with your fans set to the minimum, it will take several minutes to raise the coolant temp by 3-5C. It's those 3-5C you are fighting to keep down and I am afraid that is the limit to our control over the process. Bigger coolers with more radiator surface area get you closer to zero. If I gave you a little 120mm radiator and slow fan, the coolant temp change would be 10-15C. Not so small anymore and you have raised your CPU temp minimum by 15C. People with smaller coolers can usually see this plainly. My normal idle is 25C, I go play games for 2 hours, then when I quit my idle CPU temp is 40C. How come? They have raised the temperature of the entire case (or at least the coolant) to 40C. GPUs tend to help with that as well.

 

I often lead with "it's annoying to have your fans ramp up and down with CPU temp" because it tends to be more persuasive than the scientific explanation (and a fair bit shorter), but remember running the fans up and down will not change the stage 1 cooling of the CPU -- physical conduction from CPU->TIM->cold plate. Most people are a little surprised to learn we have so little control over the process. If there was a better way, I am pretty sure the industry would push things. It's not just home consumers who are limited by these rules and this is a roadblock to future development. Somewhere out there someone is playing around with exotic elements trying to scrape an extra 1% of of the transference, but that is never going to be a practical application. Ultimately, there is more success in actually trying to cool the side of the CPU to below the room temperature, either with liquid nitrogen or some kind of heat exchange system that chills the water externally to below room temperature.

 

I believe 3000 series temp monitoring is coming i n the next iCUE release. If you really needed to test it now, you can always move the radiator fans over to the motherboard and give it CPU temp control curves. Hopefully your board has some kind of fan delay/hysteresis to make it less annoying. Even then, you are quite literally spinning your wheels for no reason. We are conditioned to think CPU load must equal fan response, but it is about as effective as running a table fan in another room and hoping it will cool you off.

Edited by c-attack
Link to comment
Share on other sites

Now look at the problem from the other side. Take a fixed CPU load test like that of CPU-Z or any Linpack. Set your fans to 500 rpm fixed. Initiate the test. Record the before CPU temp and that of 100% load 5 seconds later. Now do the same thing with the fans set to 1000/1500/2000 rpm, whichever you like. The before and after results will be the same. Fan speed does not prevent CPU temp spikes. Nothing can do that on the cooler end. The only difference the fans can make is to prevent heat from accumulating in the coolant system. That change is always slow.

 

Right, I get that CPU temps won't change with different fans speeds if the coolant is 28C. But the difference is how much of an impact the fans can make once the coolant starts to rise, and more importantly how quickly it can react. In my testing it's clear that the coolant changes temperature very slowly over time. If I run a test with my optimal fan profile, the CPU temp will get up to a high value long before the coolant goes up in temp enough to spin up the fans, then when they do finally spin up, the CPU temp stabilizes and starts to come down quickly. Then when the coolant finally goes down again in temp, the fans slow down and then the temps on the CPU go back up again. If I leave the stress test running for 10 mins I can clearly see the fans going up and down in speed and the CPU temp going up and down due to the extremely slow reaction time of the coolant temperature to CPU temp changes.

 

But what is interesting to me is that although the coolant temp does not seem to change much when I crank the fans on, the CPU temp on the other hand *does*.

 

I ran a test where I had the cooler fans running at 300rpm and my case fans at their lowest settings. Then I measured temps and run CPU-Z stress testing.

 

@ 0 min

CPU: 64.3C

Coolant: 28.3C

 

@ 10 min

CPU: 92C

Coolant: 46.3C

 

Then I cranked the fans on maximum for both the cooler and the case and kept the stress test running. CPU temp started dropping almost immediately.

 

@ 1 min

CPU: 80C

Cooland: 45.2C

 

@ 5 min

CPU: 63.7C

Coolant 31.2C

 

So clearly having the fans active is very effective at bringing down the CPU temps, and what I found very interesting is that after 1 min the CPU temp had dropped 12 degrees, but the coolant has only gone down about 1C. As I watched it, the CPU temps seemed to come down a lot faster than the coolant temps. Also interestingly after 5 mins the CPU temp was actually lower than when I first started the test, yet the coolant temperature was 3C higher.

 

So the point I was trying to make earlier, is that using coolant temp as the fan control variable means it is always going to be very delayed in responding to CPU temps, both going up and down. Over a prolonged period of time the CPU temp is going to zig-zag up and down as the fans speed up and slow down in a delayed response to the coolant temps.

 

But what if you just controlled the fan based on average CPU temps over a period of time, like say 15-30s. After prolonged CPU load, the fans would happily ramp up to 1200-1300 RPM and just stay there keeping things at a nice constant cooling and not too loud.

 

I suppose you could set a fan profile to have the fans just sit at 1200-1300 rpm or so once the coolant goes over a non-ambient temperature (say 32C in my case) and then it would remain constant but then you are stuck with the unfortunate consequence of having to change the profile for summer and winter based on the ambient temperature of your room.

 

I suspect the CPU temperature ranges, being so much broader, would be a better control variable. Ie: temps below an average of 50C would mean low fans speeds, and then ramp it up to 1200rpm @ around 80C and go full bore if it gets to 90C. I suspect that fan profile would work in both summer and winter without needing changes.

 

Unfortunately I cannot test my theory because a) Ryzen CPU temp is not an available control variable yet and b) iCUE would need to be changed to support temperature averaging.

Link to comment
Share on other sites

 

But what is interesting to me is that although the coolant temp does not seem to change much when I crank the fans on, the CPU temp on the other hand *does*.

 

 

So clearly having the fans active is very effective at bringing down the CPU temps, and what I found very interesting is that after 1 min the CPU temp had dropped 12 degrees, but the coolant has only gone down about 1C. As I watched it, the CPU temps seemed to come down a lot faster than the coolant temps. Also interestingly after 5 mins the CPU temp was actually lower than when I first started the test, yet the coolant temperature was 3C higher.

 

That can't happen. With a flat, linear test like CPU-Z it should XX degrees when you start and then creep up +1C for each 1C of coolant rise, or -1 and -1C. CPU temp should never jump in bunches like that (for that specific test) until you kill the power or the CPU throttles. I don't have an explanation other than some kind of CPU temp reading issue from AMD software (I am assuming that is what your were using) or you are looking at socket temp core temps. AMD is a little different with their nomenclature. I am not sure that was exactly the controlled test I had in mind, but since I did not elaborate further with instructions, I won't pick it apart.

 

 

So the point I was trying to make earlier, is that using coolant temp as the fan control variable means it is always going to be very delayed in responding to CPU temps, both going up and down. Over a prolonged period of time the CPU temp is going to zig-zag up and down as the fans speed up and slow down in a delayed response to the coolant temps.

 

That's right, but the fans remove heat from the water system, not blow air at the underside of the CPU.

 

 

But what if you just controlled the fan based on average CPU temps over a period of time, like say 15-30s. After prolonged CPU load, the fans would happily ramp up to 1200-1300 RPM and just stay there keeping things at a nice constant cooling and not too loud.

 

This is the same as what you are doing now, except you have changed the variable to one not directly related to the fans. "It's hot in my room. I am going to program the ceiling fan to change speed based on the air temp. Except I decide to use the outdoor temperature instead." It may be 100F outside, but if its only 62F in the room, I really don't need the fan to speed increase speed. The heat outside will eventually warm the house, but not yet. You are making the assumption that if it is hot outside, then it must be hot inside. If the CPU is hot, then the coolant must be getting hot as well.

 

 

 

I suppose you could set a fan profile to have the fans just sit at 1200-1300 rpm or so once the coolant goes over a non-ambient temperature (say 32C in my case) and then it would remain constant but then you are stuck with the unfortunate consequence of having to change the profile for summer and winter based on the ambient temperature of your room.

 

Yes and this another way to do it. You don't have to make constant adjustments. Everyone has a noise vs performance stop point. Set a relatively flat curve for desktop work. Set up another "work" profile for whatever load things you do. In that profile, set the fans to the highest speed you can tolerate. Fixed. They stay there. You are removing the most heat from the coolant at all times, without going insane from the noise. I have done this with systems in the past. I might do that if I were running close to the edge of my limits and single degrees mattered. It would take longer to reach your normal "balance" coolant temperature for load. This works for people who know they are about to enter an extended period of load. I am going to start a render now. I am going to game now, etc.

 

 

I suspect the CPU temperature ranges, being so much broader, would be a better control variable. Ie: temps below an average of 50C would mean low fans speeds, and then ramp it up to 1200rpm @ around 80C and go full bore if it gets to 90C. I suspect that fan profile would work in both summer and winter without needing changes.

 

I think I have probably failed to make the point through all of this and I am not sure what else I can say. The above would work, you would not overheat your CPU (cooling system fails), your fans would run more often and faster than necessary to get the same level of cooling as someone else running much slower and quieter fans. So we are back to "it's louder for no performance gain".

Link to comment
Share on other sites

That can't happen. With a flat, linear test like CPU-Z it should XX degrees when you start and then creep up +1C for each 1C of coolant rise, or -1 and -1C. CPU temp should never jump in bunches like that (for that specific test) until you kill the power or the CPU throttles. I don't have an explanation other than some kind of CPU temp reading issue from AMD software (I am assuming that is what your were using) or you are looking at socket temp core temps. AMD is a little different with their nomenclature. I am not sure that was exactly the controlled test I had in mind, but since I did not elaborate further with instructions, I won't pick it apart.

 

Well I think this is the crux of the problem. I am using the Ryzen Master tool to measure the CPU temperature, and it's what I have been confused about from the beginning. I don't see how the CPU temp can rise at such a rapid rate when the coolant temp is not rising much, or that it can drop much more than coolant temp.

 

Maybe the way Ryzen measure the CPU temp is that it's not at the coolant plate at all, so there is a big difference between what the Ryzen measures and what the coolant temperature actually is. It's clear that the cooling system is working.

 

Oh well, for other reasons I have a Core i9 9900K system coming on Monday so when I get that put together I will be able to do some comparison testing. This is my first AMD rig in like 15 years so I am not just not that familiar with how their stuff works.

 

Oddly as a completely separate aside to this original post, the whole reason I have a Core i9 9900K coming is that the reason I got the Ryzen 9 3900X was to have both a box I can game with, and also do software development on when I am at home. It replaces an Aurora R7 with Core i7 8700K. When I first got it all set up, no matter what I did it actually compiled slower than my 8700K (33s full builds on Ryzen vs 29s on R7). I could not figure it out for the life of me as I fully expected it to be much faster given I also have an MP600 Gen4 drive in there. So I eventually figured Ryzen sucks and ordered the Core i9 9900K to replace it.

 

Then somehow it all magically got faster. Now I can do full builds in 19s? Go figure. I have never been able to get it to run slow again, and the only thing I changed is that it seemed to get fast the first time I ran Ryzen Master. Makes me think maybe it applied a firmware update to the CPU or something. Super odd.

 

Anyway long story short, the 9900K arrives tomorrow (I hope!) so I plan to build that rig anyway so I can compare the two and return the slowest one :)

Link to comment
Share on other sites

  • 3 weeks later...

Hi I have same CPU with 280mm Corsair AIO and having similar huge discrepancy of each temperature. May I ask your room temperature?

 

FWIW, I’m getting cpuz 15min stress test temperature like Coolant:28C, Room:15C, CPU:80, Fan RPM:1200, Pump: 2700rpm.

 

[EDIT] temp corrected

Edited by wzrd
Link to comment
Share on other sites

Hi I have same CPU with 280mm Corsair AIO and having similar huge discrepancy of each temperature. May I ask your room temperature?

 

FWIW, I’m getting cpuz 15min stress test temperature like Coolant:28C, Room:12C, CPU:80, Fan RPM:1200, Pump: 2700rpm.

 

Sure, my room is usually around 70 degrees Fahrenheit , so probably 21-22C. I have since changed cases to ones that provide a lot better cooling but I still use the 280mm AIO. I have also found it depends on the particular AIO in question as my gaming rig has the RGB Platinum with RGB fans which uses the new Cool-IT pump. My work rig has the RGB without RGB fans, which uses the Asetek pump. They do behave differently and my Cool-IT pump with identical CPU tends to run a little cooler.

 

My coolant temps on my gaming rig now hover around 28C under load, maybe getting into the low 30's while my CPU temps now sit at or below 80C. With a case that allowed less airflow, my CPU temps were getting into the 85's (and performance was degraded). Coolant temps on that case would get to 35C or so. I actually tested the MSI Gundr 100 case, which has absolutely horrendous airflow and with the front and top panel on, even with the fans going full tilt my CPU would hit 95 degrees and the coolant would be well into the 40's, at one point getting close to 50. Yikes. That case was super cool without the covers on but naturally looks like garbage without the covers on so it went back to Best Buy.

 

My old case was the NZXT h510 Elite (for home) and h510 regular (for work). The elite was actually worse for cooling as even though the designs are similar, for the RGB fan layout in that case NZXT pushed them a lot closer to the font glass, which choked them out a bit. It still worked decent, but was not great. The non Elite h510 case without the glass front panel allows the fans to breathe more as they sit back further from the front panel.

 

Last week I rebuilt it all in the Phanteks P600S case, which with the covers off (which still looks good), has amazing cooling. So now I get the temps I want.

 

As for the discrepancy between the coolant temp and the CPU temps, it's just how it works. My coolant temp seems to run lower on the Cool-IT pump version compared to the Asetek one for some reason. And CPU temps stay a little lower.

 

Now as for how to control the AIO, since your temps are going to vary depending entirely on room temp, if you set your fans based on coolant temps you pretty much need to make your own custom curve to get it where you want. I still truly believe the optimal setup for AIO cooling fans is to run them from the average CPU temps, probably an average of the last 20 seconds or so. I always find that my temps will climb, then later on the coolant finally heats up, then the fans finally kick on, and then even though the coolant temp remains relatively stable, the CPU temps start to come down as the airflow over the radiator starts to have an impact. But for me that always seems to happen too darn late.

 

But for the moment that's how it all works and its a whole lot better than having the fans spin up every time you open Chrome! But I am convinced there is a better solution if Corsair would allow us to set temps based on average CPU temps, not instant CPU temps (or motherboards for that matter; they always spin the fans up instantly). As has been pointed out by others in this thread, CPU temps rise and fall instantly based on load, and once they hit a high value under sustained load they will climb over time as the CPU and cooling apparatus gets heat soaked, so ideally our cooler needs to move heat away. But AIO's don't really function at all without fans. You get minimal cooling without any airflow, so unless the fans run, it won't cool the CPU much at all even with the pump running at maximum speed.

Link to comment
Share on other sites

Also while iCUE recently got updated and can now read the Ryzen CPU temperature for the 3900X, it's not every useful to run fans against because the max temp on the custom fan profile is 60 degrees. One option is to plug the fans into a Commander Pro to get a wider temp curve (up to 100 degrees).

 

I actually ended up doing just that for my gaming rig as I changed out all my fans two days ago for Silent Wings 3 fans to see how much of an impact that has on performance (I want more airflow for the same loudness level :) ). Unfortunately the RGB Platinum cooler does not seem to be able to properly interface to the Silent Wings 3 fans and while they are PWM fans, 95% makes them run about 50% speed, 40% stalls them out and the last 50% of speed seems to happen from 95% to 100% which is super annoying when trying to program a curve. But since I have a Commander Pro in there anyway to run all my case fans, I plugged them into that and that controls them nicely and I can set specific RPM values (not percentage values like the cooler requires for some reason).

Link to comment
Share on other sites

Sure, my room is usually around 70 degrees Fahrenheit , so probably 21-22C. I have since changed cases to ones that provide a lot better cooling but I still use the 280mm AIO. I have also found it depends on the particular AIO in question as my gaming rig has the RGB Platinum with RGB fans which uses the new Cool-IT pump. My work rig has the RGB without RGB fans, which uses the Asetek pump. They do behave differently and my Cool-IT pump with identical CPU tends to run a little cooler.

Thanks for the reply. That's nice finding. I don't have experience with cool-it aio.

 

My coolant temps on my gaming rig now hover around 28C under load, maybe getting into the low 30's while my CPU temps now sit at or below 80C.

That must be expected result for these AIOs as you described.

 

With a case that allowed less airflow, my CPU temps were getting into the 85's (and performance was degraded). Coolant temps on that case would get to 35C or so. I actually tested the MSI Gundr 100 case, which has absolutely horrendous airflow and with the front and top panel on, even with the fans going full tilt my CPU would hit 95 degrees and the coolant would be well into the 40's, at one point getting close to 50. Yikes. That case was super cool without the covers on but naturally looks like garbage without the covers on so it went back to Best Buy.

 

My old case was the NZXT h510 Elite (for home) and h510 regular (for work). The elite was actually worse for cooling as even though the designs are similar, for the RGB fan layout in that case NZXT pushed them a lot closer to the font glass, which choked them out a bit. It still worked decent, but was not great. The non Elite h510 case without the glass front panel allows the fans to breathe more as they sit back further from the front panel.

 

Last week I rebuilt it all in the Phanteks P600S case, which with the covers off (which still looks good), has amazing cooling. So now I get the temps I want. [\quote]

I tried to check temp with radiator being took out of the case. Holy, you actually compared with aio installed in the cases, that would be a hell of a work.

 

As for the discrepancy between the coolant temp and the CPU temps, it's just how it works. My coolant temp seems to run lower on the Cool-IT pump version compared to the Asetek one for some reason. And CPU temps stay a little lower.

 

Now as for how to control the AIO, since your temps are going to vary depending entirely on room temp, if you set your fans based on coolant temps you pretty much need to make your own custom curve to get it where you want. I still truly believe the optimal setup for AIO cooling fans is to run them from the average CPU temps, probably an average of the last 20 seconds or so. I always find that my temps will climb, then later on the coolant finally heats up, then the fans finally kick on, and then even though the coolant temp remains relatively stable, the CPU temps start to come down as the airflow over the radiator starts to have an impact. But for me that always seems to happen too darn late.

Looks like I need to adjust the fan curve. I tested how pump speed affect under full load with cpuz.

 

Pump speed

  • 1000rpm
    CPU:88C, Coolant:32C
  • 2000rpm
    CPU:80C, Coolant:28C
  • 2700rpm
    CPU:79C, Coolant:27C

(FANs are all 1100rpm)

 

But for the moment that's how it all works and its a whole lot better than having the fans spin up every time you open Chrome! But I am convinced there is a better solution if Corsair would allow us to set temps based on average CPU temps, not instant CPU temps (or motherboards for that matter; they always spin the fans up instantly). As has been pointed out by others in this thread, CPU temps rise and fall instantly based on load, and once they hit a high value under sustained load they will climb over time as the CPU and cooling apparatus gets heat soaked, so ideally our cooler needs to move heat away. But AIO's don't really function at all without fans. You get minimal cooling without any airflow, so unless the fans run, it won't cool the CPU much at all even with the pump running at maximum speed.

Average CPU temp option would be nice. Speaking of pump speed and fan speed, it seems that the effect of pump speed for cooling hit the ceiling easily than fan's air flow. But the pump speed immediately affects the temperature of CPU. When I increased the pump speed to 1000 to 2000, the temperature drops 8C instantly.

 

Also while iCUE recently got updated and can now read the Ryzen CPU temperature for the 3900X, it's not every useful to run fans against because the max temp on the custom fan profile is 60 degrees. One option is to plug the fans into a Commander Pro to get a wider temp curve (up to 100 degrees).

 

I actually ended up doing just that for my gaming rig as I changed out all my fans two days ago for Silent Wings 3 fans to see how much of an impact that has on performance (I want more airflow for the same loudness level :) ). Unfortunately the RGB Platinum cooler does not seem to be able to properly interface to the Silent Wings 3 fans and while they are PWM fans, 95% makes them run about 50% speed, 40% stalls them out and the last 50% of speed seems to happen from 95% to 100% which is super annoying when trying to program a curve. But since I have a Commander Pro in there anyway to run all my case fans, I plugged them into that and that controls them nicely and I can set specific RPM values (not percentage values like the cooler requires for some reason).

Corsair software can now read 3900X temp? It's AMD's officially supported AIO. It should be zen2 ready asap :D:

Link to comment
Share on other sites

Thanks for the reply. That's nice finding. I don't have experience with cool-it aio.

 

Yeah I was not expecting that either. Its possible it's the fans, but they are pretty similar. I have changed to Silent Wings 3 fans now and I plan to rebuild my work rig this weekend in the P600S case also since I like it so much, with Silent Wing 3 fans also. Then I can compare apples to apples with the Asetek and Cool-IT pumps. Will be interesting to see how they end up, as to date they have also been in different cases and of course the fans are different.

 

I tried to check temp with radiator being took out of the case. Holy, you actually compared with aio installed in the cases, that would be a hell of a work.

 

Mostly I was testing with the front and top covers and and off. With most cases you can notice a big difference. With the P600S you can have the covers on but leave off the extra panels and get excellent cooling and it still is nice and silent. Best airflow case I have tried so far.

 

Looks like I need to adjust the fan curve. I tested how pump speed affect under full load with cpuz.

 

Pump speed

  • 1000rpm
    CPU:88C, Coolant:32C
  • 2000rpm
    CPU:80C, Coolant:28C
  • 2700rpm
    CPU:79C, Coolant:27C

(FANs are all 1100rpm)

 

That's interesting. I don't think I spent much time on the quiet pump setting. I can barely hear mine (either the Asetek or the Cool-IT) so I generally leave it in either balanced or extreme and I usually can't see a difference between those two settings. I wish we could do custom profiles for the pump, but we can only choose presets at the moment.

 

Corsair software can now read 3900X temp? It's AMD's officially supported AIO. It should be zen2 ready asap :D:

 

Yes, an update landed earlier this week and now I can see it's temps and use it as a control variable.

 

Here is my fan profile I used with the Corsair fans, and then the profile I use via the commander pro with silent wings 3 fans

 

6EqtwG4.jpg?1

 

jdX7Woy.jpg?1

Link to comment
Share on other sites

 Share

×
×
  • Create New...