Jump to content
Corsair Community

What value is the H100i actually reporting to CPU_FAN?


Synergist
 Share

Recommended Posts

I'm using an H100i RGB Platinum connected as advised to the CPU_FAN header on my ROG MAXIMUS XI HERO. Just bought this H100i to replace an H100x - at least that, being controlled by AI Suite, span its fans reliably! I thought AI Suite was garbage until I experienced iCue crashes for the first time yesterday...

 

I've spent a while this evening trying to figure out potential causes for the repeated iCUE service crashes, which seem to cause the fans to briefly ramp to 100%, then fall to zero before reinitialising with a quiet/balanced profile - which results in an overheated CPU.

 

I was also curious to see that the device was apparently reporting an additional RPM which didn't match any of the AIO fans to CPU_FAN. I thought at first it was just a combined value of bother AIO's radiator fans, but setting them both to zero RPM mode in ICUE resulted in CPUFAN/CPU0 still showing around 4500-5000 rpm.

 

I then wondered if it was some multiple of pump speed, so lowered that to Quiet - no change to CPU0.

 

Cranked both AIO fans to maximum 3000 rpm - CPU0 now reported around 3600 rpm, so I started to wonder if it was a CPU pump with some odd multiplier or offset... Nope. Nothing makes any sense. It seems to increase or decrease in line with changing the pump speed profile, but the value it in/decreases by isn't transposable to anything like the actual pump values.

 

The CPU0/CPU_FAN value also doesn't appear to match with anything recognisable on the board. What might it be reporting?

 

 

 

I'm also getting fed up with the Corsair iCUE service crashing and holding the last known value, while the AIO fans spin down to zero RPM - resulting in a CPU idling at over 70 celcius with no alarms. I'm now also aware of the years of discussion about iCUE crashes on this forum!

 

Other apps like OCCT or HWiNFO have no problem reading Winbond chipset status and temps, why is iCUE having such a hard time even just existing? And is anyone using a known good build of iCUE or CLINK which doesn't experience these driver/service faults? I've tried two previous versions of iCUE - 3.27.68 was working until this week, so I rolled back to 3.26.95, same problems; currently trying 3.25.60. However this feels more like a Corsair bus driver / USB issue, which is nuts on a top of the line motherboard and an otherwise-stable machine with a normal amount of peripherals (and some connected via the monitor's separately powered USB3 hub).

 

Danger now is that I lose all trust in this cooler - besides reporting nonsense RPMs to the motherboard, if it fails and stops the fans while rendering or doing something like Folding@Home, it could toast mine and other's CPUs. Bad news.

Edited by Synergist
Link to comment
Share on other sites

First, if you are running other monitoring tools (as you indicated, you run OCCT and HWInfo) that read from the Corsair devices, you will have issues. I don't know about OCCT but HWInfo definitely will read from the Corsair devices and definitely will cause well-known and documented issues.

 

The hardware device can only handle communications from one application at a time. It has absolutely nothing at all to do with the drivers, regardless of what it "feels" like. It's a simple device limitation. When there is another app in there trying to access the device at the same time, you will have issues; the device simply doesn't support it. Period. The only thing that you can do - with HWInfo at least - is to disable that in the "Safety" tab AND you need to disable/unload the persistent driver.

 

This issue is well known and the workaround is also well known. I know that I've posted it up here HUNDREDS of times in many, many, many, many discussions. 99% of the time this issue is remedied by using the working around that I mentioned. If you don't want to use iCUE at all, then you can use that software to monitor your iCUE devices but not control them. You'll have a fixed fan curve profile based on the coolant temperature, not on the CPU temperature. Regardless, having your fan curve tied to the coolant is the optimum method - it shouldn't be tied to CPU temperature. This, too, has been discussed at length on here.

 

TBH, IMHO, this is the fault of HWInfo. They know that this is an issue but insist on enabling this by default. While it's a simple fix, it's not one that many users are aware of off the bat. It'd be simple for them to fix it too ... just don't access the device. Or check if the Corsair Service is in the list of active processes and not access the device. Or disable it by default. Because the Corsair coolers aren't the only ones that have issues with this.

 

As for the value that the Platinum provides to the fan header - I don't think it really corresponds to anything except to tell the BIOS that something is providing cooling to the CPU.

Link to comment
Share on other sites

First, if you are running other monitoring tools (as you indicated, you run OCCT and HWInfo) that read from the Corsair devices, you will have issues. I don't know about OCCT but HWInfo definitely will read from the Corsair devices and definitely will cause well-known and documented issues.

 

The hardware device can only handle communications from one application at a time. It has absolutely nothing at all to do with the drivers, regardless of what it "feels" like. It's a simple device limitation. When there is another app in there trying to access the device at the same time, you will have issues; the device simply doesn't support it. Period. The only thing that you can do - with HWInfo at least - is to disable that in the "Safety" tab AND you need to disable/unload the persistent driver.

 

This issue is well known and the workaround is also well known. I know that I've posted it up here HUNDREDS of times in many, many, many, many discussions. 99% of the time this issue is remedied by using the working around that I mentioned. If you don't want to use iCUE at all, then you can use that software to monitor your iCUE devices but not control them. You'll have a fixed fan curve profile based on the coolant temperature, not on the CPU temperature. Regardless, having your fan curve tied to the coolant is the optimum method - it shouldn't be tied to CPU temperature. This, too, has been discussed at length on here.

 

TBH, IMHO, this is the fault of HWInfo. They know that this is an issue but insist on enabling this by default. While it's a simple fix, it's not one that many users are aware of off the bat. It'd be simple for them to fix it too ... just don't access the device. Or check if the Corsair Service is in the list of active processes and not access the device. Or disable it by default. Because the Corsair coolers aren't the only ones that have issues with this.

 

As for the value that the Platinum provides to the fan header - I don't think it really corresponds to anything except to tell the BIOS that something is providing cooling to the CPU.

 

Really helpful response, I appreciate the detail! Interesting that the cooler provides a weird value to CPU0 but doesn't accurately represent something like the cumulative fan RPM, a missed opportunity and it would have been useful to have.

 

 

I was also surprised that after the most recent iCUE crash, the AIO fans stopped and although after a reboot they span up at POST, they quickly stopped again until I loaded iCUE once in Windows. I would have expected the device to either always fail safe at a 75% constant or revert to a coolant-based curve as you say.

 

I'm currently running iCUE with a moderate fan curve based on package temp. I keep an eye on coolant temperatures and was getting concerned when they rose more than five celcius above nominal, what would you consider the threshold for more aggressive fan speeds?

 

Also is there a neat way to get the device to fallback to a coolant based curve if the iCUE service fails, or should it do that already?

 

I went straight in and disabled the CLink/Asetek option in HWiNFO's safety tab; I'll also disable the persistent driver tomorrow and see if I can still provoke it ;-)

 

Now to try and figure out how to reduce the spin down and up times. The hysteresis on the RPM change for a curve or fixed RPM change seems much too slow.

 

 

Sidenote - a few reasons why I initially presumed a USB related problem and not controller conflicts... First supposition was USB bus power demands, then when that didn't pan out perhaps an interrupt issue or even an XHCI issue (my previous X99 board was a bad one for that). It was interesting that I could provoke the issue by changing fan curve profiles in fairly quick succession, but on a clean boot with minimal adjustments in iCUE it seemed stable to a point.

 

I'm learning every day, thank you again for explaining.

 

I've had HWiNFO open recently to monitor temps with the previous and current cooler. Ironic that this has been contributing to cooler controller problems. Never had a single issue with my old H100i v2 and CLink4. :)

Link to comment
Share on other sites

There really isn't a hysteresis - or only a very mild one - in the cooler. If you create a custom fan curve that's based on the coolant temperature, it'll save that profile to the cooler. But if you base a curve on the CPU temp - which requires the software to be running to be active - it'll default to ... uhh ... maybe the last curve it had that was based on coolant? Would that be the "Silent Mode" in your case?

 

For the temp source for the fan curve - take a look at question A3 in the linked AIO cooler FAQ. In short, pump temp is the proper one. Once you think about the thermodynamics of how liquid cooling works, CPU temp as the control variable just doesn't make a lot of sense.

 

Regarding creating a fan curve - take a look at question A17 in the same FAQ.

 

Those settings in HWInfo should do the trick. Personally, I'm a big fan of HWInfo and I use it on a number of machines frequently. However, I run the portable version so it never installs that persistent driver. And I have the settings saved in the .ini file on the network share that I run it from so no matter what system I'm running it on, I have those settings already saved. It caused issues in Link too but it wasn't as noticeable - iCUE communicates with the device (for RGB control) much more frequently that Link did so it's far more likely to overlap with an access attempt from other software. And ... setting the fan curve to use CPU temp as the control source also increases how often iCUE is communicating with the controller.

Link to comment
Share on other sites

I also choose to run the portable version to avoid the persistent driver, so I'm glad to hear that. Worthwhile noting that CPU-based curves rely on iCUE running (of course, how could the cooler read the die temps?) but internal coolant based curves get saved to the unit, that's nice. I can only presume that the last coolant-based curve I set was the silent curve, hence the spindown. More testing on that to come I think. I did note that a few of the default profiles were apparently coolant temp only with no option to change the sensor.

 

I presumed the fan RPM ramping was entirely being done in software and to have no option to adjust the rise/fall speed was nuts. Even the rubbish AI Suite has that option, albeit in fixed increments. Appears this behaviour is hardcoded, with only fan speed and temp references in the text configs? Given up peeking DLLs for the moment :)

 

 

I've now created an adjusted fan curve using the H100i Platinum Temp as the reference; the curve editing is awkward as the edit box isn't designed for a curve with a range of 5 degrees :) And a shame you can't specify decimal places for the threshold temperatures. However it does seem to be working as expected and the pump is on Extreme.

 

6ZzMfke.png

 

Interestingly, my unit's pump had an intermittent buzzy rattle at certain RPMs which mostly went away when set to the Extreme profile. After running the pump at max for a few days, when I did lower the pump profile to test, the rattle seemed to have gone. It amused me that bedding in apparently fixed it.

 

Only other thing left to do is figure out how to rebalance my USB port usage, apparently the H100i draws enough current to overwhelm the root hub. Since hooking it up to one of the internal USB headers, both of the case-top ports have been inoperable :D Computers never fail to frustrate.

Edited by Synergist
Link to comment
Share on other sites

All coolers report 1/2 the pump speed to the motherboard not fan speed so a quiet watercooler with a pump running at 1800 rpm will show 900rpm at the fan header. I guess its because a constant is needed to be fed back to most motherboards but newer ones do seem to have specific headers for coolers and pumps now.
Link to comment
Share on other sites

All coolers report 1/2 the pump speed to the motherboard not fan speed so a quiet watercooler with a pump running at 1800 rpm will show 900rpm at the fan header. I guess its because a constant is needed to be fed back to most motherboards but newer ones do seem to have specific headers for coolers and pumps now.

 

Not the case. That's one particular series of cooler (e.g. H100i V2). The Pro coolers report about the exact pump speed (IIRC) and the Platinum cooler report a mystery number.

Link to comment
Share on other sites

Not the case. That's one particular series of cooler (e.g. H100i V2). The Pro coolers report about the exact pump speed (IIRC) and the Platinum cooler report a mystery number.

 

And if I ever find out WHAT the value actually represents, I'll let you know. I currently suspect pump RPM plus some arbitrary value.

 

What is your ambient temperature?

Having the curve band that narrow usually isn't necessary. Pushing the fans to 80% at 24C is a bit extreme ...

 

Indeed, I'm loathe to let the coolant go over 30 :roll: I started off with a fairly aggressive curve. I ran some CPU soaks and checked temps - frustrating that coolant temp is reported to a tenth of a degree, but you can only set curve points by whole degrees :[pouts:

 

After a couple of hours of Folding@Home pegged at about 75% CPU, once the work units finish, CPU package and cores return to idle at ~40 celcius with browser & a few apps open. The coolant settles to between 23.2 - 23.5 celcius, just under the room ambient (as measured on desk, a short distance above top exhaust) with fans sitting around 1600-1800 rpm. They spool up quickly and max out around 2600 rpm in line with coolant temp from CPU load.

 

At 100% utilisation, F@H is on par with C4D for CPU stress - looks like AVX levels of workload, all cores easily at or above 80 C :eek:.

 

I'm aiming to keep coolant at or under 29 celcius (current ambient+5) during load, my goal's to avoid excessive heat soak as this 9900K gets very hot very quickly despite its modest OC.

 

Now I'm more familiar with the rad's dissipation I've tweaked the curve; AIO fans spin at ~1600 rpm with coolant temp at about 23 celcius. Curve looks steep but seems OK. I'm sure it'll get tweaked again...

 

EBQWzdi.png

Edited by Synergist
Link to comment
Share on other sites

If you want to run the fans that hard, it's up to you. However, keep in mind the relationship of coolant temp increase to CPU temp increase is 1:1. So when you coolant temp goes up 0.7C, you have only added 0.7C to the CPU temp. I would expect a routine coolant delta of around +8C for the H100i + 9900K and a more moderate fan slope. 5C is likely as low as you can go.

 

Running that steep a curve is going is probably going to be more dynamic than you would like and someone opening a window somewhere in the house might cause a fan speed change. Your baseline will slide around with room temperature and its not easy to keep that level outside of a lab environment. You might be better off toggling between two profiles - one with a high fixed fan speed rate and the other with something quieter for light desktop work. If you have a Corsair mouse or keyboard, that can be bound to one key/switch for easy toggling back and forth.

Link to comment
Share on other sites

That integration sounds nice, shame I'm CM and Razer :)

 

I agree, I've already set up a couple of profiles and I've jiggled the more aggressive curve a bit more. Perhaps I was overly concerned with dissipating heat but I'll keep an eye on temps. Shuffled them back more, 70% threshold is currently at 26C. Coolant's steady at 24.6 with ~1000 RPM fans.

 

As you say this room will edge up a few deg by June so will definitely have multiple profiles. I wonder if hotkey toggling is also possible via AHK or eventghost. How do Corsair do it, iCue integrated with keyboard hotkeys?

Edited by Synergist
Link to comment
Share on other sites

 Share

×
×
  • Create New...