Noctua violates PWM Spec, Causes Problems in Multi-Fan Configs

[Zarathustra[H]]

EDIT:

I wanted to make it excruciatingly clear that Noctua fixed my issue. They explained the problem, and offered to swap out all 18 of my fans. See more details here. Good job Noctua in addressing the issue head on, solving it, and winning a customer for life.


Original Post:

Bear with me on this one. I don't have a news link to post, because no one seems to be reporting on this yet, but it appears as if Noctua (and possibly EK based on what I have seen in other forums) have started breaking PWM compatibility with their latest fans.

I found this out personally the hard way in my ongoing build.




How PWM Fans Usually work:

There is a lower speed cutoff below which the fan does not spin. Above this cutoff the fan scales linearly with the PWM signal. 50% PWM duty cycle gets you 50% of the rpm. On my old 140mm Noctua iPPC-2000 pwm fans this worked something like this:

PWM Duty Cycle      Actual Fan Speed
0%                  0%
10%                 0%
20%                 20%
30%                 30%
40%                 40%
50%                 50%
60%                 60%
70%                 70%
80%                 80%
90%                 90%
100%                100%

The way things with Noctua's new fans:

Single Fan Configutation:

PWM Duty Cycle      Actual Fan Speed
0% - 30%            0%
35%                 20%
40%                 23%
45%                 26%
50%                 29%
55%                 33%
60%                 37%
65%                 42%
70%                 48%
75%                 54%
80%                 61%
85%                 69%
90%                 78%
95%                 88%
100%                100%

So, it's not too bad in a single fan per PWM channel configuration. Yes, it comes on more slowly and ramps up towards the end, rather than being linear, but most people probably wouldn't even notice.

The problem is, if you have a many fan water cooling solution, you are trying to control using PWM signal splitters, which used to work just fine, with all fans scaling linearly, the problem is much much worse.

Here are the new Noctua fan speeds using an 8-way PWM splitter:

PWM Duty Cycle      Actual Fan Speed
0% - ~75%            0%
80%                 20%
85%                 30%
90%                 45%
95%                 65%
100%                100%

So, you get no fan speed at all below 80% PWM duty cycle, then at 80% you get a very low fan speed, which ramps up exponentially until it hits 100% at 100. This is going to throw off just about every fan controller out there, resulting in a revvy peaky and overly loud fan control.

The really sneaky part about all this is that the change was rolled in without much fanfare, and unless you know the exact cutoff batch, there is no way of knowing if the Noctua fan you are buying has the old proper PWM implementation or the new problematic one. Noctua appears to be aware of the issue, with unconfirmed reports of them assisting other customers by replacing their fans with older models that function properly.

I can confirm this happens with Noctua fans, but according to the Aquacomputer community forums, the latest EK fans are doing the same thing. It is unclear to me why Noctua and EK are doing this (cost savings maybe?) According to the postings over there it has something to do with how/if the PWM signal is pulled up to 5.25V inside the fan per the Intel spec. I am not an electrical engineer - however - and cannot comment on this.

What is clear is, Aquacomputer (the manufacturer of the Aquaero fan controller) says the problem is entirely on the side of the fan. Noctua and EK have violated the PWM specification, and there is nothing they can do on their end to overcome this without risking damage to other fans.

A user in the Aquacomputer forums suggests it might be possible to make these fans behave like they are supposed to by modifying the PWM splitters to "add a +12V to 5.5V switching regulator and then a 5.5V to 5V LDO" or a simpler solution might be to "add a 4,7kOhm resistor or maybe also a 1k resistor between VCC (12V) and the PWM Signal pin"
but these fixes might actually kill other fans that properly implement the PWM spec, so they are not perfect.

I wanted everyone to be aware of this issue before buying fans, and winding up with the mess I am in, where unless Noctua Support comes through for me, I am out almost $500 on fans, and fan accessories and am beyond my Amazon return window, because of my large complicated build taking more than 30 days to get to the point where I had everything hooked up for testing.

I think the safest bet right now is just to avoid all Noctua and EK fans until further notice. An electronic/adapter fix maybe possible, but it is unclear at this point.

Consider this a PSA.

 

[jeremyshaw]

Intel ATX fan spec, version 1.3 (seems to be the latest) - I'm wrong, it's not technically ATX spec, rather "4-Wire Pulse Width Modulation (PWM) Controlled Fans Specification"
https://www.glkinst.com/cables/cable_pics/4_Wire_PWM_Spec.pdf

Noctua PWM spec
https://noctua.at/media/wysiwyg/Noctua_PWM_specifications_white_paper.pdf

I didn't run into this problem when designing my own PWM controller for a PWM controlled fan (various modern Noctua PWM and Nidec PWM fans were used). Did you scope your Aquacomputer controller under load (tap one of the 8 channels when loaded up)? Aquacomputer is blaming others, but it's still valid to evaluate their claims and their controller's performance.

EDIT: of amusement, unless if there is a later "4-Wire Pulse Width Modulation (PWM) Controlled Fans Specification" updating the fan control segment, it seems Noctua misinterpreted Intel's example, when they state "Operation below 20% PWM duty-cycle is not officially supported in the Intel specification (undefined behaviour)." It's only an example, and 20% was chosen arbitrarily as a point below 30% (the allowed maximum value for the minimum fan speed). This has no relevance to the OP's claims, however. It's only an amusing diversion.

Followup to the edit, Intel published many specs openly on their website. The Fan control spec ain't one of them. Very strange omission.

 

[BrotherMichigan]

This is also a problem with be quiet! fans as well, I believe. An alternative solution: just run them with voltage control instead.

 

[1_rick]

Just so I'm clear, if you plug all 8 fans in, you get the third table, and if you then unplug 7 of those 8 fans, you get table 2, but if you have half of them plugged in, you get something somewhere in the middle? I assume you're not running into some kind of current limit. (I've never split more then 3 fans onto a single mobo header, so have no personal experience).

 

[travm]

I expect this has to do with fan features added to minimize the PWM "buzz" sound. Possibly make the fans more efficient to boot. Entirely a fan controller issue of that's the case. Not able to supply adequate current, or to much output impedance for whatever the fans have in place to manage the sound.

Entirely speculation. I don't own any of these things or have any inside info. Wcftech likely has more reliable information.

 

[Zarathustra[H]]

I expect this has to do with fan features added to minimize the PWM "buzz" sound. Possibly make the fans more efficient to boot. Entirely a fan controller issue of that's the case. Not able to supply adequate current, or to much output impedance for whatever the fans have in place to manage the sound.

Entirely speculation. I don't own any of these things or have any inside info. Wcftech likely has more reliable information.

The the last intel PWM spec seems to be version 1.3 from 2005. It says:
"This signal (the PWM signal) must be pulled up to a maximum of 5.25V within the fan."

According to what I am reading, this is what Noctua is failing to do properly.

Some of the hacks I am reading about to fix this issue, seem to involve pulling up the PWM signal to 5V before it enters the fan, but my electrical schematic skills are way too weak to say for sure.

Now, granted, that looks like a pretty shitty spec to me. It fails to give us an acceptable range, and just gives a max. Not sure if that is what is the problem

Here is what user Jibby posted on the Aquacomputer forums. He says it works for him, but he can't make any guarantees it won't damage anything long term.

I love engineering redlines

 

[jeremyshaw]

The the last intel PWM spec seems to be version 1.3 from 2005. It says:
"This signal (the PWM signal) must be pulled up to a maximum of 5.25V within the fan."

According to what I am reading, this is what Noctua is failing to do properly.

Some of the hacks I am reading about to fix this issue, seem to involve pulling up the PWM signal to 5V before it enters the fan, but my electrical schematic skills are way too weak to say for sure.

Now, granted, that looks like a pretty shitty spec to me. It fails to give us an acceptable range, and just gives a max. Not sure if that is what is the problem

Here is what user Jibby posted on the Aquacomputer forums. He says it works for him, but he can't make any guarantees it won't damage anything long term.

View attachment 234128

I love engineering redlines

That is a maximum allowable voltage level. Right underneath that line, it goes and recommends 3.3V pullup (Noctua interprets it as 3.3V/5V, on page 6). IMO, this is because 5V (TTL's Logic Level) isn't that common for logic and internal I/O anymore, and 3.3V is usually within the logic=HIGH trigger range for modern 5V devices (again, IMO, since this isn't exhaustive).

Also, all you showed was a standard CMOS inverter setup, handling the Aquacompter's output and feeding a signal into the Noctua fan, at a nominal ~5V level. I have no clue why you have a switching regulator, when any decently specced LDO is enough for this (extremely commonplace in Arduino & Arduino Accessories).

Again, as noted earlier, scope the output of the Aquacomputer's output under load. Make sure it's actually meeting spec before moving on. It's clear the Noctua fan can work correctly without that extra device in the way. Even the most rudimentary deductive logic would require an investigation of Aquacomputer's device, since it was the extra variable.

 

[serpretetsky]

Also, all you showed was a standard CMOS inverter setup,

Close, if you look carefully the PMOS is actually connected backwards. Which means it's effectively working like a simple diode, allowing +5V down to the output with a small voltage drop. Mistake?

 

[Zarathustra[H]]

That is a maximum allowable voltage level. Right underneath that line, it goes and recommends 3.3V pullup (Noctua interprets it as 3.3V/5V, on page 6). IMO, this is because 5V (TTL's Logic Level) isn't that common for logic and internal I/O anymore, and 3.3V is usually within the logic=HIGH trigger range for modern 5V devices (again, IMO, since this isn't exhaustive).
View attachment 234130

Also, all you showed was a standard CMOS inverter setup, handling the Aquacompter's output and feeding a signal into the Noctua fan, at a nominal ~5V level. I have no clue why you have a switching regulator, when any decently specced LDO is enough for this (extremely commonplace in Arduino & Arduino Accessories).

Again, as noted earlier, scope the output of the Aquacomputer's output under load. Make sure it's actually meeting spec before moving on. It's clear the Noctua fan can work correctly without that extra device in the way. Even the most rudimentary deductive logic would require an investigation of Aquacomputer's device, since it was the extra variable.

Well, again, I am not skilled with the electrical stuff. It is outside of my domain, so I lack enough knowledge to speak intelligently on this issue. I'm collecting information from others and trying to make the most sense of it I can.

 

[odditory]

PWM is quite the rabbit hole. I acquired like 40 grey delta case fans from Mac Pro towers for like $20, thinking I could just resell them as Delta fans since they're beautiful and powerful. Oh hell no. Apple had their own proprietary thing going on with them. Modding and tweaking went nowhere to taming them.

I wish you good fortune to solving this. YOULL NEED IT

 

[schmide]

7v for life

 

[xx0xx]

From some reading I was doing on fan controllers when wanting to buy one myself, I ran into multiple instances of users saying that fan controllers often failed to supply the voltage/wattage that they claimed they could for fans and pumps hooked up via PWM (example review). Is it possible something like that could be at play here?

 

[Twisted Kidney]

Sometimes I think a more USB-like interface for fans and pumps would be a huge benefit.

That's about when I consider how much that would cost. I mean, Corsair has already decided 50 bucks is perfectly reasonable.

 

[rgMekanic]

voltage controlled fans ftw.

Sometimes I think a more USB-like interface for fans and pumps would be a huge benefit.

That's about when I consider how much that would cost. I mean, Corsair has already decided 50 bucks is perfectly reasonable.

NZXT Grid, it's what runs my 11 fans.

 

[Twisted Kidney]

voltage controlled fans ftw.



NZXT Grid, it's what runs my 11 fans.

I mean for each fan. That way you wouldn't have to worry about proprietary nonsense for lighting. A system that can identify the fan and its operational parameters to the BIOS and tuning can be done from there along with the aforementioned lighting.

Too many manufacturers that have spent too much money trying to wall off their gardens would have to sign on.

 

[Zarathustra[H]]

Too many manufacturers that have spent too much money trying to wall off their gardens would have to sign on.

This is everything that is wrong with the modern PC building landscape.

I don't want walked off ecosystems from any vendor. I don't want hardware that is dependent on software running on my machine (apart from a small, efficient driver)

Any fan controller I would ever want to use would need to be more configurable than what most motherboards are capable of, yet still be OS agnostic. If my fan control depends on having a functioning software running in the OS I'm out.

I'm OK with setup being done at one time in OS, but once set up, whatever controls my fans must be completely independent of and not running on the OS.

 

[zerogg]

This is a real bummer. I have used noctua alot in the past, and didn't expect this. I was planning on getting a bunch of the black nf a12s whenever those are finally released.
I did a quick search and didn't find anywhere else talking about it.

When you talked to support did they indicate if they were going to fix it in upcoming runs? Did they come through for you?

 

[Zarathustra[H]]

This is a real bummer. I have used noctua alot in the past, and didn't expect this. I was planning on getting a bunch of the black nf a12s whenever those are finally released.
I did a quick search and didn't find anywhere else talking about it.

When you talked to support did they indicate if they were going to fix it in upcoming runs? Did they come through for you?

My interaction with Noctua Support is still underway. Things are progressing slowly, but that is to be expected with our little pandemic we have going on, so I am not holding it against them.

I initially contacted them on March 29th.

On April 3rd they responded as follows:

We have recently introduced an additional protection mechanism for our industrialPPC fans. Unfortunately, this protective function in combination some fan controllers can lead to problems with PWM control. In normal setups such as in a PC or if you only operate one fan per channel, the fan works as intended. Would you be so kind as to send us the lot codes of your fans?

By "some fan controllers" I'm fairly certain this means any fan control which uses one PWM signal over multiple fans. I could be wrong though.

I responded with my batch information on that same day, but have not heard back yet. Again, the world has bigger shit going on right now, so I'm expecting things to be slower than usual, so while I am a little frustrated that my project is taking forever, I am not upset with Noctua over the time this is taking.

It looks to me like this is a feature of their fans going forward. I have no idea if it impacts ALL Noctua fans, but by what I am reading on the Aquacomputer forums it sounds like pretty much everything new from Noctua and EK is impacted.

There is a suggestion in there that Noctua Support has been helping affected customers by shipping them fans from previous batches made before the change to address their issues, but presumably that can't go on forever.

It is also unclear to me what kind of protection a fan really needs.

 

[Zarathustra[H]]

For what it is worth, this appears to be the official response from Aquacomputer:

Noctua had the great idea to add some kind of protection circuit to newer revisions of fans which is causing this problem. We do not know what they did there in detail but we know it is causing the problem because it is also Noctua that confirmed that to one of our customers. Their solution was that they offered the customer fans from an older batch that do not suffer from this problem.

There is no way for us to solve this - or at least not without raising new problems with regular working fans. We already sent out modified Splitty9 active boards for testing purposes to affected customers. Feedback so far is that it only partially helped but a normal operation with affected Noctua fans is still not possible. Noctua will have to fix this with a new revision and we know for sure that they are well aware of this problem.

 

[zerogg]

Thanks for the reply, I will be watching this. I imagine it was frustrating figuring out what was wrong.

I wish they would walk back this change.
I have been using a aquaero 6 with pwm splitters for rad fans as well. I don't understand the need for "protection" on fans either?

 

[Krazy925]

This post made me bust out my old gentle typhoons until you get an answer about if this is a permanent issue.
I was gonna replace some corsairs with noctuas but imma wait now. I use a fan controller.

Perhaps they need two models.

 

[Deleted member 243478]

PWM is quite the rabbit hole. I acquired like 40 grey delta case fans from Mac Pro towers for like $20, thinking I could just resell them as Delta fans since they're beautiful and powerful. Oh hell no. Apple had their own proprietary thing going on with them. Modding and tweaking went nowhere to taming them.

I wish you good fortune to solving this. YOULL NEED IT

It isn’t hard making a PWM fan controller with a 555 timer. I made one for my finger chopping Nidec VA450DC fans.

 

[clockdogg]

...It is also unclear to me what kind of protection a fan really needs.

Unless your system operates in an ISO 5 clean room, fans need protection from the horrors of dust bunny invaders. Sure, they look light and fluffy on the outside, but the freeloaders will collect on the trailing edge of your pristine fan blades. And no amount of RGBling will remove the disgusting grime.

Oh... maybe you meant electrical protection... never mind.

 

[Arcygenical]

It's not hard to make a circuit that takes PWM signals, converts it to analog, with voltages ranging up to PWM VCC... With a beefy enough board and a step up, might be able to get this to 12v without a huge pain in the ass.

I'm just using a 6$ 50w PWM controller with a knob on it from aliexpress and it works for my needs. Much easier to hack one of those up that your motherboard, or its PWM signaling. I have 6 120mm radiators in the loop, I adjust that knob seasonally.

 

[Zarathustra[H]]

It's not hard to make a circuit that takes PWM signals, converts it to analog, with voltages ranging up to PWM VCC... With a beefy enough board and a step up, might be able to get this to 12v without a huge pain in the ass.

I'm just using a 6$ 50w PWM controller with a knob on it from aliexpress and it works for my needs. Much easier to hack one of those up that your motherboard, or its PWM signaling. I have 6 120mm radiators in the loop, I adjust that knob seasonally.

I'm not sure you'd want to bring the PWM signal up to 12v. That might start harming things.

Intel spec is a max of 5.25v, but they strongly recommend 3.3v

 

[Zarathustra[H]]

I just realized I am WAY overdue in updating this thread with Noctua's response.

I have been moving (which is hell on earth) so that's my excuse.

Here is what they told me:

Greetings from Noctua!

Thank you very much for your patience!

The issues have occurred because we have increased the protection of the PWM control line in order to further improve electronic robustness in industrial environments. This extra protection has increased the capacitance of the PWM line slightly. In a situation where multiple fans are connected in parallel, the capacitance of the PWM line will end up in parallel, increasing the capacitance. This means that the PWM control line driver will have an increased capacitance load to drive, which can cause the signal to deteriorate. If a large number of fans is used, it is therefore important that controller or splitter board have independent outputs for each fan. If they don’t and if the output driver is sensitive to increased capacitance this can cause a signal degradation, which, in return, can lead to reduced fan speeds when controlling multiple fans.

We have already reverted our production back to the original version in 2019 in order to make sure that there are no more compatibility issues with controllers and we are working on a new, updated version that will feature the extra protection while still being fully compatible.

They proceeded to offer exchanges for all 18 of my fans for newly manufactured fans on their reverted production line, which I accepted.

For what its worth, my old fans which exhibited the issue had a date code starting in 1908. (which I can only presume is August 2019) The new ones I received had a date code starting in 1911. I do not know when the switchover occurred.

Anyway, kudos to Noctua on responding well to an unfortunate situation. As anyone who has worked in engineering design or manufacturing knows, eventually mistakes happen. I try not to judge companies based on whether or not they have mistakes (eventually they hit everyone) but rather on how they respond to them when they happen, and in this case Noctua did the right thing, explained the situation and offered to fix it.

They even sent me fans priority express shipping from Austria which I can't imagine was cheap, and was totally unnecessary.

So, after briefly flirting with removing them from my preferred brand list, I can say they are back with a vengeance. It's nice to know that a company takes care of it's customers.

 

[vegeta535]

I just realized I am WAY overdue in updating this thread with Noctua's response.

I have been moving (which is hell on earth) so that's my excuse.

Here is what they told me:



They proceeded to offer exchanges for all 18 of my fans for newly manufactured fans on their reverted production line, which I accepted.

For what its worth, my old fans which exhibited the issue had a date code starting in 1908. (which I can only presume is August 2019) The new ones I received had a date code starting in 1911. I do not know when the switchover occurred.

Anyway, kudos to Noctua on responding well to an unfortunate situation. As anyone who has worked in engineering design or manufacturing knows, eventually mistakes happen. I try not to judge companies based on whether or not they have mistakes (eventually they hit everyone) but rather on how they respond to them when they happen, and in this case Noctua did the right thing, explained the situation and offered to fix it.

They even sent me fans priority express shipping from Austria which I can't imagine was cheap, and was totally unnecessary.

So, after briefly flirting with removing them from my preferred brand list, I can say they are back with a vengeance. It's nice to know that a company takes care of it's customers.

Did it fix your issues?

 

[Zarathustra[H]]

The thing many companies don't realize about PR issues, is that while they are unfortunate, if handled correctly, they can show how committed a company is to their customers, and improve reputation.

The strategy should be:

1.) Acknowledge the issue. (customers hate denials)
2.) Explain what happened. (Customers also hate opaque interractions)
3.) Offer to fix it, and do so in a reasonable time frame.

Noctua did this, and it has made me want to continue to buy their products, as I know that they will solve issues when they occur in the future.

There is a reason people still speak ill of EK 10 years after the Nickel plating issue.

 

[Zarathustra[H]]

Did it fix your issues?

That, I haven't been able to test yet.

I am going out on a limb a little bit here and taking them at their word, but they told me they reverted their production line to the previous specification, and the previous specification is known good to me, so I am taking them at their word. It would be profoundly stupid of them to lie about this considering how verifiable it is.

I have had another issue with my build which means it is down right now, so I cannot test. The fact that I am moving has further exacerbated this. Everything I own is in boxes, and I just had not had the time to troubleshoit my build or test fans yet.