Fan calibration and RPM fan curves #2333
Replies: 14 comments 83 replies
-
|
It's awesome! What happens if you hold the fan off with your hand and rpm feedback is lost? It will slowly go to 100%? |
Beta Was this translation helpful? Give feedback.
-
|
^ Well, that's where RPM and calibration comes in. When you start thinking in RPM, the "true" range in percentage becomes irrelevant. |
Beta Was this translation helpful? Give feedback.
-
|
Also @ionGL ended up with a 32 rpm resolution. Multiples of 2s allows for really quick math. |
Beta Was this translation helpful? Give feedback.
-
|
How do I download the old version? This is not working correctly for me and is way more complicated. The old version worked great for me with 0 issues. |
Beta Was this translation helpful? Give feedback.
-
|
Older versions: https://github.com/Rem0o/FanControl.Releases/releases |
Beta Was this translation helpful? Give feedback.
-
|
Thanks. |
Beta Was this translation helpful? Give feedback.
-
|
Can someone make a ELI5 of the calibration feature? I may be quite dumb and I'm not understanding the utility of the feature 😅 |
Beta Was this translation helpful? Give feedback.
-
|
Traditional way: Hey fan, please run at 77% power, whatever that is. Scenario with 2 fans, one is a 3000 rpm fan, other is a 1200 rpm fan. Calibrated way: Hey both fans, run at 1000 rpm. |
Beta Was this translation helpful? Give feedback.
-
|
Holy shit, that was a fast response lol. And now seeing it that way and finally understanding it, I feel even dumber for not getting it on my own 😂 What an amazing job you are doing with FanControl! Thank you so much @Rem0o 😊 |
Beta Was this translation helpful? Give feedback.
-
|
@Rem0o Oh that's an excellent ELI5. Thanks so much! |
Beta Was this translation helpful? Give feedback.
-
|
Quick feedback after playing with the new feature for a few minutes.
|
Beta Was this translation helpful? Give feedback.
-
|
^ I will add a new logic to wait for the fan to stabilize, but even then I think it won't cover your VRM fan as it is sooooo slow. You fall out of the "works for 95% of people" and this is why manual calibration input is possible ;) |
Beta Was this translation helpful? Give feedback.
-
|
Just to add to this feedback, similar experience here. Fans on my CPU cooler (Thermalright Peerless Assassin SE) take about 5 seconds to react, so the auto calibration for that one wasn't horribly off. Same with my case fans (Fractal Design Silent Series R3 140mm). However, fans on my deshrouded GPU (two Arctic P12 PWM PST CO fans) take about 20 seconds to settle their RPM, so I had to redo the calibration for these fans from scratch, RPM was off by about 200-300. Someone else here with the same CPU cooler said their fans took 10 seconds to stabilize - perhaps there's something else also impacting RPM reaction time? When I was calibrating my GPU fans, I noticed that whenever I set the % to 30 or below during manual fan calibration, the percentage would be forced to 0% for some reason. |
Beta Was this translation helpful? Give feedback.
-
|
On nvidia cards, the gpu fan control works like this: any fan % between 1 and 30% is still 0%. You can't play in that range. If you deshrouded and put normal fans on the gpu, you should connect them to a motherboard fan header and gain access to the full 0-100% pwm range. |
Beta Was this translation helpful? Give feedback.
-
|
That's good to know, thank you @ionGL. I'll keep them connected to the GPU though, I don't have any free headers on my motherboard and I don't mind not having that bottom range, the fans are already basically silent at 30% and I have them running in a semi-passive configuration anyway. |
Beta Was this translation helpful? Give feedback.
-
|
If you don't need that 1-30% range, the gpu header is fine. Those arctic fans should spin at a super silent rpm at 30%. Knowing about that 30% limit, you can have a graph around it, with a first point at 0% and the next one at 30%. Make sure your card can go to 0% at any temp. To test this, have the card loaded a bit, wait for 60C or so and manually set the fans to 0%. If they go to zero and stay there, you are good. Some card don't let you do that. Use NvApiWrapper and disable nvidia 0% hardware curve override, like this. |
Beta Was this translation helpful? Give feedback.
-
|
I know you could manually do it, but it would be nice if you could set a dwell time while taking automatic readings. I did manual and automatic and found that automatic switches to the next test too quickly before the fans had stabalised. For some, the difference was 200-300rpm at 100% for example. |
Beta Was this translation helpful? Give feedback.
-
|
In the different cases I tried, 8 seconds per 10% increments was plenty enough. However I knew it wouldn't be perfect for all cases. My plan is to have an algorithm that look for value stability before moving to the next increment, with a timeout of course. |
Beta Was this translation helpful? Give feedback.
-
|
Yes, a stability check could be the solution. Or, on the "Only the controls with a paired fan speed sensor will be tested" card, an option "Test each % for x seconds" could be added. And i think that once a fan is calibrated, opening its "manual fan calibration" card should default to the % the fan is at that moment. No need to jump at 50% each time this is opened just to check the fan map. |
Beta Was this translation helpful? Give feedback.
-
Great call. Will do. |
Beta Was this translation helpful? Give feedback.
-
|
Hi Rem0o, Would be great to integrate your new feature to GRAPH mode with little switch between % and RPM. What do you think?
And add RPM to the card
By the way:
|
Beta Was this translation helpful? Give feedback.
-
|
The graph can be turned into rpm mode from the 3dots menu, like all other fan curve. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you Rem0o, what about window size? Should i open new topping? |
Beta Was this translation helpful? Give feedback.
-
|
Yes ^ |
Beta Was this translation helpful? Give feedback.
-
|
Possibly the biggest step forward in fan control - I can't believe nobody has done this before. Fan percentage is, in all honesty, just the middle-man we have to go through to get to rpm, and direct-rpm control lets us get rid of all this "find start/stop percentage" nonsense that's never consistent across fans. I would suggest a margin +/- of tighter than 50 rpm merely because, at the low end, 50 rpm can be a noticeable noise increase/reduction. Also, regarding the "holding time" of the calibration steps, the Phanteks T30 (and probably others) has a LOT of inertia, so it takes a long time to spin up to maximum and to respond to changes. Indeed, the calibration gave a maximum speed of 1869 rpm, when the true maximum is just over 2000. Instead of a fixed holding time, is there a way to poll the rpm and use a "be happy with the result once it no longer varies by 5%"? As always, thanks for a wonderful program👌🏻. |
Beta Was this translation helpful? Give feedback.
-
Will be working on a value stability tool, which should handle all cases with no hard coded time delays. |
Beta Was this translation helpful? Give feedback.
-
|
Give https://github.com/Rem0o/FanControl.Releases/releases/tag/V184 a go and see how it goes for auto-calibration. |
Beta Was this translation helpful? Give feedback.
-
|
I love this idea, but my issue is that calibration doesn't work with some of my fans. I use Corsairlink for several fans and with my system the time it takes for the controller to issue speed changes to the fans takes too long. They change at about 15-20 RMP per second, so the when I do the auto configuration and it ramps up to 100%, then down to 90%, 80%, etc. the fans on that controller haven't even made it up to speed before the test gets down to 60% and even then it can't keep up to the change as it drops. Would it be possible to add an option we can toggle that does an extended test for this and all the assisted setup? Personally I would have no issue letting the configuration run for 5-10 minutes instead of having to set all my start, stop and RPM settings manually, which takes much much longer. Maybe have it user assisted, a prompt asking the user to verify the fan has stabilized at the current speed before changing. |
Beta Was this translation helpful? Give feedback.
-
|
Is that a parameter in ICUE or something in the corsair firmware, how slow/fast the fan RPM can change? I would be very surprised this is the default, as it is incredibly slow. |
Beta Was this translation helpful? Give feedback.
-
|
Not that I ever saw, ICUE can change the fan speed 1 of two ways, by commanding the controller's fan channel to a certain percentage or RPM. When ICUE commands percentage the rate of change is very slow, which is what EvanMulawski's CorsairLink does for the commander core (see link https://github.com/EvanMulawski/FanControl.CorsairLink#:~:text=The%20speed%20ramping,be%20set%20manually. ), when commanding by RPM the change is nearly instant. I have used the CLI program https://github.com/liquidctl/liquidctl and it can command the controller by RPM but I assume because FanControl works by percent EvanMulawski saw no need to implement that in CorsairLink. |
Beta Was this translation helpful? Give feedback.
-
|
I see, weird limitation still from Corsair. |
Beta Was this translation helpful? Give feedback.
-
|
I agree, my older corsair controller didn't have that limitation. In actual use it's not really an issue, so long as my fan curve isn't too steep you never really notice how slow the fans ramp up or down. |
Beta Was this translation helpful? Give feedback.
-
|
Hi, @Rem0o Please forgive me if my question has been asked before, I'm going crazy because of the problem I'm having. I have a "ZOTAC GAMING GeForce RTX 3070 Ti Trinity OC" video card (3 fans), after calibration I find two gpu fans ("the first one" on the far right and "the second one" on the left and center) and that's not problem. My problem is that the fans stop frequently, I don't want it to go up from 0rpm because the noise is very annoying, I want to make a curve that runs constant at low temperature, increasing according to the temperature but never stopping completely. I want to do something like the "fan stop setting" that Zotac had in the old software but removed in the new versions, is there any way to override it?
|
Beta Was this translation helpful? Give feedback.
-
|
RPM mode will not help you if you don't fully understand how the fan works in % mode. RPM only translates percentages from the calibrated fan map. Try the ! Force apply option. Your gpu fans might not work at 48%. It's too low and the fan turns off, so the bios compensates and sends a higher % to try and spin it back up. Open the manual fan calibration and try to add more points to the map. This way, when/if you decide to use RPM mode, it will have a better understanding of what % to set when you request a RPM.
Add 49%, 55% etc. If you can make it to take 48, 47 somehow and it's stable, you will have access to a lower % as your idle speed, as you say you don't want to turn off the fan. |
Beta Was this translation helpful? Give feedback.
-
|
Also put the start % at 0. Just set it like in this pic bellow and then play around with manual control to find where your fan starts and stays on.
|
Beta Was this translation helpful? Give feedback.
-
|
I tried what you said but I can't get below 892 rpm mate, it's very frustrating... but thank you very much for your support, I will do some more research, I will let you know if I find something |
Beta Was this translation helpful? Give feedback.
-
|
@ondrhn the fans on your GPU might just be limited at that minimum speed and that's it. You'll most likely have to deal with that minimum speed. |
Beta Was this translation helpful? Give feedback.
-
|
I saw a post on a forum suggesting that changing the power mode in the control panel works, in my settings it was "Prefer maximum Performance", when I selected "normal" and tested it after the restart my fans didn't work at all (during the furmark test the temperature went up to 83 degrees and they still didn't work .d) it looks like a hard coded setting for some video cards and we can't manipulate it with curve as far as I understand (we used to be able to turn off this control in zotac's firestrom software but they removed that option) |
Beta Was this translation helpful? Give feedback.
-
|
Hi, |
Beta Was this translation helpful? Give feedback.
-
|
For your "real name" request, you basically mean a new text field to the control cards next to the existing name? I assume you know the existing field is editable right? You may just give it a really long name if you want to add additional information about that fan. Why the next for a second field? As for the RPM range, I know that's useful information, but the clutter is starting to get real on the control cards. The speed cards are meant to be hidden, tbh I might remove them completely in the feature, only leaving them for the pairing process and that's it. I may add information to the tooltip when you hover your mouse over the RPM? The info is available when you click on the "calibrated [x] " label, but that might be a bit more conveniant. |
Beta Was this translation helpful? Give feedback.
-
|
Something like: |
Beta Was this translation helpful? Give feedback.
-
|
Tried running automatic calibration with v184 (.NET 8) a couple times. All 10 fans are paired (three intake, two cooler, one exhaust, three GPU) and automatic calibration eventually reports six of them as done before exiting. None of the fans' marking changes from uncalibrated to calibrated, nothing in log.txt. I can manually calibrate both fans which report as done and which do not but RPMs are not displayed when editing their graph curves, possibly because there's a mix control which combines several graphs and the fans are assigned to the mix. In some cases the graphs route to multiple fans. Known issues? |
Beta Was this translation helpful? Give feedback.
-
Er, no, often GPU fans get uncomfortably loud. Noise can be an issue with case, cooler, or rad fan mounts at higher RPMs as well, depending what's installed and how it behaves. It's also not really worth spending time characterizing operating points which won't be used. While I tend to go up to 100% for completeness and out of curiosity, sometimes with earplugs, it's odd the cal range isn't boundable, the stepping configurable, and adaptive defaults employed.
Can't say I see being able to manually perform tedious labor as much of a feature, but sure. Don't agree ~30 s per point times 20 points times all the fans in a build is very little time—for a typical ATX build it's a couple hours for reasonable settling to set point and subsequent RPM and noise averaging. Easily more if you're not getting a reliable noise floor. If multiple fan configurations are being evaluated those stack on additional hours. It's not a prohibitive effort but I wouldn't trivialize getting up at four five in the morning to go to the office on a weekend to run measurements when it's quiet or scheduling work around changes in building HVAC activity. If noise isn't particularly a concern that widens the measurement window, shortens the time per point somewhat, and enables concurrent settlement of multiple fans and RPM averaging in parallel when fan to fan interactions aren't much of a concern. Still a decent amount of work to do manually, though.
I'm aware. Unsure why this point is being made but that's ok.
I'm aware. This is, in my experience, routine practice for checking part to part variations and interactions between individual fans and fan headers. A bit on the higher end—when ordering a custom prebuilt the assembler's unlikely to ship it with this degree of customization, for example—but it's something which comes along naturally with the measurement intensity being discussed here. |
Beta Was this translation helpful? Give feedback.
-
This should now be much better. A linear regression test is done on the RPM values until the average "ramp" is flat at each step, to make sure the values are stable. Also, the value for a given step will be the average value of the last few input points, which again should increase accuracy.
Not exactly, it's the other way around. The graph doesn't "push" a value to the fan, the fan "pull" a RPM value from the graph, then does the conversion itself using its calibration table/map. As for @twest820, the primary goal of the calibration is to normalize control whatever the fan is. For example, my own case has 3 types of case fans, each with different max RPM values. To make them run at the same speed, I need to create different curves for each type of fans, and/or apply offsets to the fans to try and synchronize their speed. With RPM mode, I can create a single fan curve with RPM values, from let say 700 rpm to 1200, apply it to all these different fans, and they will run at the same speed, always. That's it, there is nothing more to it. |
Beta Was this translation helpful? Give feedback.
-
Yes, obviously a graph doesn't push anything. The control pushes the corresponding % saved in the calibration table, to match the RPM in the selected RPM curve. I don't see an issue with testing at 100% any fan. Even if loud. How uncomfortable can they be if you can't tolerate them for 20 seconds? There is no way around this if you want to know its full map. This is what i do with any fan. Set it to 100%, write in a .txt the rpm. Then go to 95% and so on. It takes 10 seconds at most to stabilize and get a good reading when you have small % jumps. Do this until it stops, note the exact %. Then from go 1 by 1 the other way around until it starts spinning. You only have to do this once/fan and you have everything about it, saved for future reference. A few minutes to get a complete fan map is nothing. Anyone who tinkers with stuff like this finds it trivial. |
Beta Was this translation helpful? Give feedback.
-
This is, in my experience, entirely typical. Haven't seen RPM matching mentioned as much of a priority for anyone else, though. Even if all fans are of the same type with negligible part to part and header to header variations it's likely flow resistances vary among mounts. So individual fans end up with distinct m³/h-RPM ratios. Since case, cooler, and rad manufacturers don't provide flow resistance coefficients you're left to guess(timate) the various operating points and resulting airflow balance. The math's the same with multiple fan types, just with a more diverse set of m³/h-RPM ratios. As much as practical I try to use GPUs' default fan curves. For a default front to back air cooled ATX build that leaves three intakes, two cooler fans, and a rear exhaust. Typically I have all six of those on mix controls taking in per fan CPU Tctl and dGPU graph curves and graphs for all relevant drives. Depending on the case and cooler plus variations in what was available and replacement of worn fans there's 2–5 different fan designs in use (maybe six but I don't think I've done that yet). Generally every fan design has a different spin up point and maximum RPM and, within multiple fans of the same type, response curves often differ noticeably due to unit to unit variation, varying behavior among motherboard headers, and changes in flow resistance. Complexity scales up with inclusion of top and bottom/shroud fans like you'd expect. Within this class of approaches there's a convenience to being able to nudge a fan up by 50 RPM directly rather than by a few percent of drive. But you'll still end up looking at the fan card to see what drive the control point adjustment's actually giving, how it's feeding through the mobo, what the RPM ends up being as a function of the control temperature mix, and maybe checking audio spectrometry. So the practical difference between bumping RPM and bumping drive is pretty small.
FWIW, in air cooled builds I've found varying RPM among fans consistently offers some beneficial combination of lower noise-normalized temperatures, lower noise, lower dust accumulation, and lower wear. Details depend on the build but I'd expect generally similar results from most current cases (Lian Li Lancools, Fractal North/Meshify/Torrent, etc.), tower coolers (Thermalright, ID-Cooling, 3RSYS, DeepCool, Scythe, etc. etc.), and fans of interest (P12, P28, PH-F120T30, Silent Wings, NF-A12x25, ...). It's a bit different with rads where the convention's to use the same fans at the same drive and bottom to top or back to front flows are more common. |
Beta Was this translation helpful? Give feedback.
-
Picked up v185 yesterday. Automatic calibration is now successful for the mix of fans I've checked across a couple machines. Behavior's not great when viewing the cals, though, as even just opening the manual cal dialog to review the results and closing out of it is fairly slow. Absent another option for viewing (short of saving the config, extracting the cal points from the .json, and plotting elsewhere) I'm unsure the dialog should suspend automatic control just to view a curve. Opening the manual cal dialog with the MSI B550 mobo I have available frequently causes fan control channels to get stuck showing ? 10%, ? 40%, or similar for their drive percentages after closing out of the dialog. Doesn't matter if there's a cal or not on the pairing. I can get channels unstuck by switching into manual control and changing drive levels but sometimes it takes several attempts. Since looking at one fan typically causes three or four other fans' drive levels to get stuck this can get old fast. Haven't seen drive levels get stuck with the ASRock B550 I've also tested on. So far, anyways. |
Beta Was this translation helpful? Give feedback.
-
|
Looks like automatic calibration introduces a regression. If a header doesn't have a fan plugged in then that control channel gets set to 0% start and some high value for stop—I got 96% on channel and 97% on another. If a fan's then plugged in it won't do anything, meaning running a cal breaks existing fan curves and mostly blocks manual control. Manually performing a calibration does not reset the stop and start values. So you end up with a fully calibrated and configured fan that doesn't work. Workaround is to manually reset the control channels' stop and start percentages though, interestingly, I initially got a total disable with stop set back to 0% and start still at the 0% default. Had to bump start to 1% to restore normal operation and then use the - button to drop it back to default. Seems to be working fine after that. Also, the middle mouse wheel doesn't jog the drive level by 1% in manual cal the way it does in manual control. |
Beta Was this translation helpful? Give feedback.
-
|
All hardware I've run recent builds on has had all headers paired for months. Haven't tested to see if auto cal can work out new pairings. If FanControl was open source the first place I'd look for this would be to see why the stop percentage is set when there's no response. Might also be the code that decides there's no fan cal is missing to revert the stop percentage. |
Beta Was this translation helpful? Give feedback.
-
|
So you had your fan paired in the software, then unplugged a fan then ran calibration and the output of that auto calibration makes no sense. Is that what you are describing? |
Beta Was this translation helpful? Give feedback.
-
|
That's in the right direction but, as no cal is assigned to the header, it's probably simpler to say auto has side effects when it doesn't set anything. I'm getting the impression FanControl assumes pairing is an indication a fan's plugged in, rather than treating pairing as just a binding of names. This expanded definition of pairing seems susceptible to problems in use cases such leaving a header open for fan testing or leaving a paired header open to test a different a fan config. |
Beta Was this translation helpful? Give feedback.
-
|
Hello. can someone maybe upload a good fan curve for Tuf 4080 super. reason i use fancontrol is that i layed my hand on my gpu's back plate they other day, idle mode it was on 56-57 degrees celcius. 0%fan. |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
done that still wont accept curve
Sendt fra Outlook til iOS<https://aka.ms/o0ukef>
…________________________________
Fra: Rem0o ***@***.***>
Sendt: Sunday, April 7, 2024 3:24:29 AM
Til: Rem0o/FanControl.Releases ***@***.***>
Cc: PappaJam ***@***.***>; Comment ***@***.***>
Emne: Re: [Rem0o/FanControl.Releases] Fan calibration and RPM fan curves (Discussion #2333)
image.png (view on web)<https://github.com/Rem0o/FanControl.Releases/assets/14118956/4930a8f9-a1cb-45bd-9ed6-0191523fc45d>
image.png (view on web)<https://github.com/Rem0o/FanControl.Releases/assets/14118956/da7cf106-de95-47ea-9ef1-566522671d90>
image.png (view on web)<https://github.com/Rem0o/FanControl.Releases/assets/14118956/6e6e81a3-76b5-4402-837f-cec7d2d8cb70>
—
Reply to this email directly, view it on GitHub<#2333 (reply in thread)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BHUPL5XPKSRR455QWETWIW3Y4CN43AVCNFSM6AAAAABD7GMDNKVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TAMZSGYYTS>.
You are receiving this because you commented.Message ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
You need to "uncheck" the manual control option. |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone. Is there some FAQ about FanControl? FanControl does not see any sensors in my laptop. Is it OK, because my laptop doesn't have any supported sensors or I am doing something wrong? Panasonic SF-SZ6 based on Intel Core I5-6300U. |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
Not supported |
Beta Was this translation helpful? Give feedback.
-
|
Hi everyone |
Beta Was this translation helpful? Give feedback.
-
|
yes |
Beta Was this translation helpful? Give feedback.
-
|
@Rem0o Thanks for your help ! |
Beta Was this translation helpful? Give feedback.
-
Hello all,
Working on a big new feature: calibrated fans + rpm fan curves.
Features:
Screenshots should tell you the idea:
Manual Calibration page:
UI:
This is still a work in progress/experimentation.
Posted to gather comments, questions and suggestion.
Like, how accurate would you want the calibration to be? +-100 rpm? 50? 25?
Fire away.
Beta Was this translation helpful? Give feedback.
All reactions