Strange fan behavior [longish]

[George Macdonald]

A few weeks ago I put together 4 Athlon64 systems all based on the Asus
A8N-E. The builds were all smooth with no glitches... well, apart fron a
broken driver set from nVidia<ptui again>, I got everything working quite
quickly. All systems same Rev of mbrd, same BIOS. same drivers, etc., etc.
all bought in the same month, 3 of them in the same order.

I used Cool'n'Quiet with Asus' Q-Fan which controls the CPU fan speed
according to temperature. Then I figured I might as well load Asus'
PCProbe at least temporarily to check that things were working right.
Turns out that with 2 of the systems, at idle, the fan was turning at
~2400RPM vs. a max speed of ~3300RPM, while with the other 2 systems it was
800-900RPM and the lowest "alarm" setting is 800RPM... IOW a nuisance in
case the alarm goes off for a user.

For those who don't know, in normal business tasks, Athlon64 systems spend
most of their time at close to min temp. of ~33C or so with normal ambient.
What was happening with all 4 systems was at startup into Windows, the fan
speed was ~1500RPM and from there 2 of them would slowly ramp up to 2400RPM
and the other 2 would slowly ramp down to a touch above 800RPM

So I thought I'd try Speedfan (http://www.almico.com/speedfan.php) to see
what was going on. SpeedFan allows you to control fan speed manually by
setting a %age of max speed - a great program BTW. So with Q-Fan turned
off, and PCProbe uninstalled, the first 2 systems would lower the speed to
1500RPM at 45% drive, after which the speed would go up:
~1800RPM at 40% drive
~2100RPM at 35% drive
~2400RPM at 30% drive
stopped at 25% drive.

The other 2 systems went the other way of course:
~1500RPM at 45% drive
~1300RPM at 40% drive
~1100RPM at 35% drive
~800RPM at 30% drive
stopped at 25% drive

Apart from this possibly being useful info for anyone else, I'm puzzled as
to how this behavior comes about, I haven't tried switching fans around to
see if the anomaly follows the fan or the mbrd but since the mbrds are all
"the same", I'm assuming that AMD is fitting two different kinds of fans to
their heatsinks. In that case, since, AIUI, those fans are speed
controlled by a PWM drive, how in the hell can a fan run faster with a
lower duty cycle PWM? I'm pretty sure that this is not a mis-measurement
of fan speed due to PWM/tach signal effects - the fans which read a faster
speed at 30% drive *are* turning much faster.

With SpeedFan I can, of course, set a min drive of 45% and have all the
systems run the fan at 1500RPM at idle but I'd be interested to know if any
"fan experts" have an explanation for the oddball effect above??

 

[Ed]

A few weeks ago I put together 4 Athlon64 systems all based on the Asus
A8N-E. The builds were all smooth with no glitches... well, apart fron a
broken driver set from nVidia<ptui again>, I got everything working quite
quickly. All systems same Rev of mbrd, same BIOS. same drivers, etc., etc.
all bought in the same month, 3 of them in the same order.

I used Cool'n'Quiet with Asus' Q-Fan which controls the CPU fan speed
according to temperature.

I'll STOP READING HERE, sorry......

AMD64 stock fans have a built-in thermal fan speed controller, no need
for Q-fan.

Air temp / RPM speed (Approx.)
@ <=32C / 3000 (Min Speed)
@ >=42C / 6000 (Max Speed)

btw, I've had two A8N-E's running with BIOS 1008 for 11 months now with
X2-4200+ CPUs in them....
0 crashes.
0 lockups.
0 BSOD.

If you download the chipset drivers from Nvidia you may still have a "!"
in device manager in WinXP, but if you install them from the mobo CD it
installs this missing driver, or if you install PC Probe II it will
install the driver also. At least that's how they acted for me 11 months
ago.


Ed

 

[Franc Zabkar]

So I thought I'd try Speedfan (http://www.almico.com/speedfan.php) to see
what was going on. SpeedFan allows you to control fan speed manually by
setting a %age of max speed - a great program BTW. So with Q-Fan turned
off, and PCProbe uninstalled, the first 2 systems would lower the speed to

1500RPM at 45% drive, after which the speed would go up:
~1800RPM at 40% drive
~2100RPM at 35% drive
~2400RPM at 30% drive
stopped at 25% drive.

The other 2 systems went the other way of course:
~1500RPM at 45% drive
~1300RPM at 40% drive
~1100RPM at 35% drive
~800RPM at 30% drive
stopped at 25% drive

Apart from this possibly being useful info for anyone else, I'm puzzled as
to how this behavior comes about, I haven't tried switching fans around to
see if the anomaly follows the fan or the mbrd but since the mbrds are all
"the same", I'm assuming that AMD is fitting two different kinds of fans to
their heatsinks. In that case, since, AIUI, those fans are speed
controlled by a PWM drive, how in the hell can a fan run faster with a
lower duty cycle PWM? I'm pretty sure that this is not a mis-measurement
of fan speed due to PWM/tach signal effects - the fans which read a faster
speed at 30% drive *are* turning much faster.

The *first* thing to do is to *measure* the *voltage* across the fan.

If the fan has built-in thermal speed control, then it could be that
varying the voltage to the fan also confuses its temperature sensor.
To test whether the fan incorporates its own speed control, try
heating it with a hair drier or cooling it with spray freeze.

- Franc Zabkar

 

[George Macdonald]

I'll STOP READING HERE, sorry......

AMD64 stock fans have a built-in thermal fan speed controller, no need
for Q-fan.

Air temp / RPM speed (Approx.)
@ <=32C / 3000 (Min Speed)
@ >=42C / 6000 (Max Speed)

Where did you get those numbers from?

Maybe some have thermal control but not all... and 6000RPM - that's nuts.
I have 9 Athlon64 systems and none of them have a fan wjhich execeeds
3500RPM... and they've all been above 50C without Q-Fan. My SpeedFan tests
were done withut Q-fan too.

btw, I've had two A8N-E's running with BIOS 1008 for 11 months now with
X2-4200+ CPUs in them....
0 crashes.
0 lockups.
0 BSOD.

No "Event ID:51"s?? Do take a close look at Event Log/System.;-) The NI
is also broken - offloading doesn't work reliably... and if you take a look
at the "fix" in http://www.speaz.net/download/nvidia.rar it gives an idea
of the incomptence at work here: read the readme.txt, then look at the file
names. It will not do what it says it does.

If you download the chipset drivers from Nvidia you may still have a "!"
in device manager in WinXP, but if you install them from the mobo CD it
installs this missing driver, or if you install PC Probe II it will
install the driver also. At least that's how they acted for me 11 months
ago.

Oh that! The nForce4 6.86 driver package is broken and should not be used
by end users on a fresh WinXP install: 1) it will not install a PATA
driver; 2) yes it results in that ACPI related device with the "!"; and 3)
it is incomplete and it does not include MediaShield - just look at the
oem.ini file in the root folder. This package is for OEMs to do a custom
mix. The 6.70 package installs everything correctly.

 

[Ed]

Where did you get those numbers from?

XBIT LABS, sort of dated.....
http://www.xbitlabs.com/articles/cpu/display/athlon64-3200_3.html

My only guess is the air temp passing over the thermistor in the fan is
not the same in each PC case, or maybe some don't have the built in
controller? On the 3 amd64's I have (no opterons here) you can see a
green thermistor sticking out of the fan hub area.

Maybe some have thermal control but not all... and 6000RPM - that's nuts.

I've seen a few posts of AMD64 users complaining the fan was loud as
hell and spinning at around 6000 all the time, I just tell em contact
AMD and get a RMA, that's not normal.

I have 9 Athlon64 systems and none of them have a fan wjhich execeeds
3500RPM... and they've all been above 50C without Q-Fan. My SpeedFan tests
were done withut Q-fan too.

3500 RPM, that's about the fastest I've seen my 3200+ Newcastle go and
that's only in the summer while running under full load for extended
periods where the heat starts to build up some inside a full tower case.

The CPU temp doesn't really matter as long as it's heat isn't
recirculating right back into the CPU fan, it's all about the air temp
passing through the fan going over that thermistor, that's what really
controls the fan speed.

No "Event ID:51"s?? Do take a close look at Event Log/System.;-) The NI
is also broken - offloading doesn't work reliably... and if you take a look
at the "fix" in http://www.speaz.net/download/nvidia.rar it gives an idea
of the incomptence at work here: read the readme.txt, then look at the file
names. It will not do what it says it does.

Sounds a bit over my head, offload?, is that for the Nvidia firewall? I
don't use that, at the time I built these all I read was bad things
about it.

I did look in my system event log, the only error I see is the same one,
ID:7023...The Computer Browser service terminated with the following
error: This operation returned because the timeout period expired.

One nice thing about having Q-fan even with an AMD64 stock fan is using
it on a regular case fan, on my brothers A8N-E I connected his top case
fan (non thermal type) to the cpu fan header and plugged the CPU fan
into a case fan header (that has monitoring), this way the top case fan
only pushes more air only when the CPU warms up, so it's sort of like,
Cool n Quite Duo <g> even though Cool n Quite has nothing to do with CPU
fan control, which a lot of people on the net seem to believe and keep
repeating! <rolls eyes>.

Cheers,
Ed
--
SpeedFan Daskboard 2
4200+ @ 2.5GHz....34.0
AMD K8 Core.......35.0
Asus A8N-E........28.0
6800GS Core.......45.88
6800GS............35.0
160GB.............31.0
74GB..............29.0
Rear Fan..........2083
Front Fan.........1985
Chipset...........6750
Vcore.............1.38
+3.3V.............3.3
+5V...............4.97
+12V..............11.84
Vbat..............3.1

 

[George Macdonald]

The *first* thing to do is to *measure* the *voltage* across the fan.

If the fan has built-in thermal speed control, then it could be that
varying the voltage to the fan also confuses its temperature sensor.
To test whether the fan incorporates its own speed control, try
heating it with a hair drier or cooling it with spray freeze.

Seems logical but if the fan had effective temp control, I'd have thought
that having the CPU on CnQ, sitting idle at ~30C would have been sufficient
to see a speed drop but I don't see it... with a 100% drive manual setting
in SpeedFan, it runs at the same 3500RPM as when it's dual-load pegged with
Orthos at 60C. IOW heating it is not going to show any response and spray
freeze is not something I have laying around the house.

If it won't slow down at 30C, any possible thermistor control seems like a
waste of time. Then again, having a thermistor on a fan which is upstream
of the cooling load would not have the desired effect... kinda like a
thermistor controlled case intake fan, which is a daft idea.

The best solution so far is to just turn of Asus Q-Fan, forget PCProbe and
use SpeedFan with a max drive of 100% and a min drive of 45%, which gives a
fan speed on idle of ~1500RPM.

 

[George Macdonald]

XBIT LABS, sort of dated.....
http://www.xbitlabs.com/articles/cpu/display/athlon64-3200_3.html

My only guess is the air temp passing over the thermistor in the fan is
not the same in each PC case, or maybe some don't have the built in
controller? On the 3 amd64's I have (no opterons here) you can see a
green thermistor sticking out of the fan hub area.

I've seen a few posts of AMD64 users complaining the fan was loud as
hell and spinning at around 6000 all the time, I just tell em contact
AMD and get a RMA, that's not normal.


3500 RPM, that's about the fastest I've seen my 3200+ Newcastle go and
that's only in the summer while running under full load for extended
periods where the heat starts to build up some inside a full tower case.

So where do you get the 6000RPM number? It does *not* exist on any AMD CPU
heatsink I have. The chipset fans turn at 6000RPM or higher - maybe you're
confused here. If you want to run your temp & fan speed up to max on an X2
try Orthos (front-end to Prime95 which runs two threads) for 15 mins. It
took one of mine up to 60C - fan maxed at 3500RPM

The CPU temp doesn't really matter as long as it's heat isn't
recirculating right back into the CPU fan, it's all about the air temp
passing through the fan going over that thermistor, that's what really
controls the fan speed.

I see no thermistor on any of mine and no action to indicate it's there.
Then again, as remarked elsewhere, having a thermistor on a fan which is
upstream of the cooled load is a stupid design.

Sounds a bit over my head, offload?, is that for the Nvidia firewall? I
don't use that, at the time I built these all I read was bad things
about it.

Offloading of packet checksum and segmentation to the hardware - bad things
happen with multi-threaded apps like Eudora: error 10053 & 10054 - Google
it. There's also supposed to be offloading of some Firewall functions
but AFAIK it's never worked at all and I've never even tried to load NAM.

 

[Franc Zabkar]

Where did you get those numbers from?

Maybe some have thermal control but not all... and 6000RPM - that's nuts.
I have 9 Athlon64 systems and none of them have a fan wjhich execeeds
3500RPM... and they've all been above 50C without Q-Fan. My SpeedFan tests
were done withut Q-fan too.

See page 32 of the "AMD Athlon 64 and AMD Opteron Processors Thermal
Design Guide":

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26633_5649.pdf

"The fans used in those measurements were 70x15mm and nominally varied
speed from 3000 RPM to 6000 RPM over the temperature range of 32degC
to 42degC."

Page 30 of the same document:

"The significant temperatures for the AMD Athlon 64 and AMD Opteron
processor systems are 32°C and 42°C. These temperatures correspond to
room temperatures of 25°C and 35°C, respectively, with an additional
7°C allowed for the temperature rise from the system external
temperature to the system internal temperature at the processor
location."

- Franc Zabkar

 

[Franc Zabkar]

Seems logical but if the fan had effective temp control, ...

See page 30 of the "AMD Athlon 64 and AMD Opteron Processors Thermal
Design Guide":

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26633_5649.pdf

"Systems with high levels of acoustic noise can generate customer
complaints and can influence the acceptance of a product. Therefore,
AMD recommends using fans that have speed control implemented on the
fan.

Usually, a thermistor is used to sense the airflow temperature at the
fan to determine the fan speed."

... I'd have thought
that having the CPU on CnQ, sitting idle at ~30C would have been sufficient
to see a speed drop but I don't see it... with a 100% drive manual setting
in SpeedFan, it runs at the same 3500RPM as when it's dual-load pegged with
Orthos at 60C. IOW heating it is not going to show any response and spray
freeze is not something I have laying around the house.

If it won't slow down at 30C, any possible thermistor control seems like a
waste of time. Then again, having a thermistor on a fan which is upstream
of the cooling load would not have the desired effect... kinda like a
thermistor controlled case intake fan, which is a daft idea.

That occurred to me, too.

The best solution so far is to just turn of Asus Q-Fan, forget PCProbe and
use SpeedFan with a max drive of 100% and a min drive of 45%, which gives a
fan speed on idle of ~1500RPM.

You still haven't explained why reducing the PWM duty cycle increases
the fan speed. If there is no automatic speed control, then the fan
voltage *must* be increasing when it shouldn't be. A quick measurement
with a DMM should verify this. You *do* have one of these lying about
the house, don't you?

- Franc Zabkar

 

[George Macdonald]

See page 30 of the "AMD Athlon 64 and AMD Opteron Processors Thermal
Design Guide":

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26633_5649.pdf

"Systems with high levels of acoustic noise can generate customer
complaints and can influence the acceptance of a product. Therefore,
AMD recommends using fans that have speed control implemented on the
fan.

Usually, a thermistor is used to sense the airflow temperature at the
fan to determine the fan speed."

Hmm, 3-year old docs. I wonder.

It would seem so. <shrug>

Maybe I'll try the hairdryer with 100% drive this week-end, to see if I can
get the fan to turn faster than 3500RPM.

That occurred to me, too.


You still haven't explained why reducing the PWM duty cycle increases
the fan speed. If there is no automatic speed control, then the fan
voltage *must* be increasing when it shouldn't be. A quick measurement
with a DMM should verify this. You *do* have one of these lying about
the house, don't you?

Well I have two which behave that way and two which behave the other
way.Ô_õ I was hoping somone else would "explain" it and your thermistor
confusion seems reasonable for want of something better. I have no idea
how a thermistor responds to a pulsed voltage??

As for a DMM, nope I still have my 25-year old analog VOM but I'm not even
sure I want to risk damaging anything with no spare fan on hand - I'd have
to be able to measure on both sides of the supposedly present thermistor
which I haven't seen yet. It can't be the sensor/drive chip putting out
more voltage on the mbrd since they are all the same.

 

[Franc Zabkar]

Maybe I'll try the hairdryer with 100% drive this week-end, to see if I can
get the fan to turn faster than 3500RPM.

You could try both types of fan in the one box by connecting the
second as a chassis fan. Then you could warm up the chassis fan
without hurting your CPU.

Well I have two which behave that way and two which behave the other
way.Ô_õ I was hoping somone else would "explain" it and your thermistor
confusion seems reasonable for want of something better. I have no idea
how a thermistor responds to a pulsed voltage??

It doesn't need to. AIUI, the fan voltage is not pulsed. The pulses
are averaged by an RC combination and this smoothed DC value is then
buffered by an op-amp whose output drives a transistor. I have tried
to find an online photo of your mainboard but none clearly shows the
electronics in the vicinity of the fan connectors. I can't be sure but
I *think* I see a MOSFET at both the CPU and chassis fan headers.

Another reason why you would avoid direct PWM is that a typical
brushless DC fan employs 3 Hall effect sensors for electronic
commutation. A pulsed voltage would probably wreak havoc with such an
arrangement. And then there's the tacho output ...

As for a DMM, nope I still have my 25-year old analog VOM ...

Well, I'm gobsmacked ...

... but I'm not even
sure I want to risk damaging anything with no spare fan on hand - I'd have
to be able to measure on both sides of the supposedly present thermistor
which I haven't seen yet.

I haven't yet seen a temperature controlled fan, but I would think
that the thermistor would be internal to the fan and not externally
accessible. In any case you don't need to measure the thermistor, only
the fan voltage and its tacho output.

If the fan's internal temperature sensing utilises a potential divider
such as the following, then a fall in the supply voltage could
conceivably be sensed as an increase in temperature:

|---> transistor ---> fan motor
|
+12V o--- R --|-- NTC ---o ground

As the NTC warms up, its resistance decreases and the voltage from the
divider also decreases. A reduction in the supply voltage produces the
same result.

It can't be the sensor/drive chip putting out
more voltage on the mbrd since they are all the same.

The initial premise was that *everything* was identical, including the
fans, yet clearly something is not. The obvious point of demarcation
is the fan connector. If the voltage increases when Speedfan commands
it to increase, then you can be sure that the "problem" is in the fan.
Otherwise you are merely speculating.

- Franc Zabkar

 

[George Macdonald]

On Wed, 18 Oct 2006 22:35:07 -0400, George Macdonald
<fammacd=!SPAM^[email protected]> put finger to keyboard and
composed:

It doesn't need to. AIUI, the fan voltage is not pulsed. The pulses
are averaged by an RC combination and this smoothed DC value is then
buffered by an op-amp whose output drives a transistor. I have tried
to find an online photo of your mainboard but none clearly shows the
electronics in the vicinity of the fan connectors. I can't be sure but
I *think* I see a MOSFET at both the CPU and chassis fan headers.

Another reason why you would avoid direct PWM is that a typical
brushless DC fan employs 3 Hall effect sensors for electronic
commutation. A pulsed voltage would probably wreak havoc with such an
arrangement. And then there's the tacho output ...

This article http://www.almico.com/sfarticle.php?id=1 led me to believe
that at least some systems actually send the pulsed voltage right to the
fan... complete with a description of the possible problem of the strobing
effect on the tach output.

I haven't yet seen a temperature controlled fan, but I would think
that the thermistor would be internal to the fan and not externally
accessible. In any case you don't need to measure the thermistor, only
the fan voltage and its tacho output.

I have an old case fan with temp control and there's a device sticking out
in the airlfow near the hub, which is a blue "blob" with a red head. The
two legs *could* be probed... with a *very* steady hand.

 

[Franc Zabkar]

See page 32 of the "AMD Athlon 64 and AMD Opteron Processors Thermal
Design Guide":

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26633_5649.pdf

"The fans used in those measurements were 70x15mm and nominally varied
speed from 3000 RPM to 6000 RPM over the temperature range of 32degC
to 42degC."

If a fan can really turn "~2400RPM at 30% drive", would it not be
reasonable to expect that it could turn more than 6000RPM at 100%
drive? Could it be that Speedfan is maxing out at 3500RPM because the
HW monitor IC's count register is itself maxing out? Does changing the
divisor affect the range of the readings? Does Speedfan have a debug
mode where you can see the contents of the various registers? Assuming
the fan is a temperature controlled type, could it be that there is
poor thermal tracking between its thermistor and the heatsink?

- Franc Zabkar

 

[Franc Zabkar]

This article http://www.almico.com/sfarticle.php?id=1 led me to believe
that at least some systems actually send the pulsed voltage right to the
fan... complete with a description of the possible problem of the strobing
effect on the tach output.

Hmmm, I notice that the author doesn't address the issue of
commutation which AFAICS would create a much bigger problem. The
following article appears to support my reservations in this regard.

"Application of PWM Fan Control":

http://www.coolingzone.com/Guest/News/NL_May_2003/JMC/JMC_May_2003.html

This drawing illustrates the principle of electronic commutation:
http://www.ece.msstate.edu/~hagler/May2001/02/FIG2.JPG

I think you'll agree that the Hall effect sensors will be all over the
place if the supply is chopped. The abovementioned fan control article
suggests that PWM control is only feasible if the power to the fan's
electronics is uninterrupted. According to the article, this
necessitates a 4-wire design.

For a PWM-to-DC conversion solution, see AMD's "Open Platform
Management Architecture Specification":
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/32200.pdf

The document states that "the motherboard must provide the necessary
circuit that converts PWM pulses from the mCard output to an analog DC
level that is then amplified to actually control the fans." See Fig
15, section 8.4, page 69 for a suggested circuit.

Here is a photo of your motherboard:
http://www.thetrailingedge.com/images/products/asus-a8n-e-big.jpg

I notice that it appears to use ITE's IT8712F multi-IO chip. Here is
its datasheet:
http://www.iteusa.com/product_info/file/pc/IT8712F_V0.9.1.pdf

Section 9.6.2.2.17 (page 107) lists a selection of 8 PWM frequencies
ranging from 5.86kHz to 375kHz. The example used by the author of
Speedfan assumed that the PWM frequency was of the same order of
magnitude as the rotational frequency, namely some 10s of Hz. I would
think that if the tach signal were chopped at 5.86kHz, say, then each
tach pulse would appear as several hundred pulses. Clearly this is not
happening in your case, so I'm inclined to believe that Asus has opted
for DC averaging. Having said that, I still can't see anything other
than a MOSFET and a capacitor near the fan connectors, and then only
at the CPU fan and front chassis fan. This appears to be confirmed by
the manual which states that "the Asus Q-Fan function is supported
using the CPU Fan and Chassis Fan 1 connectors only". In the absence
of an oscilloscope I'd use the AC voltage range on a DMM to confirm
whether there is an AC component. This voltage would be measurable at
rotational frequencies but probably not at 5kHz and above.

BTW, the Asus motherboard manual shows a BIOS screen with a CPU fan
speed of 4962 RPM and a chipset fan speed of 5443 RPM. The CPU
temperature is 48 degC and Q-Fan is disabled.

Here is an interesting post:
http://www.silentpcreview.com/forums/viewtopic.php?t=18921&sid=4054b97fb8d073b37a98e63a712774c9

The author alludes to an Asus P5GD2 motherboard manual which states
....

"Q-Fan 2 ... allows you to select the type of cable connected to the
CPU fan connector. Set to [PWM] when using a 4-pin CPU fan cable. Set
to [DC] when using a 3-pin CPU fan cable." Options [PWM][DC]

I notice also that your manual states that a fan can draw up to 2A.
This would suggest that, if the MOSFETs were operating in linear mode,
then they could be required to dissipate up to 6W (1A at 6V). In PWM
mode the dissipation would be close to zero. So maybe you're right and
I'm wrong. Is it time to break out a scope?

- Franc Zabkar

 

[Franc Zabkar]

This article http://www.almico.com/sfarticle.php?id=1 led me to believe
that at least some systems actually send the pulsed voltage right to the
fan... complete with a description of the possible problem of the strobing
effect on the tach output.

I found this article:

Why and How to Control Fan Speed for Cooling Electronic Equipment
http://www.analog.com/library/analogdialogue/archives/38-02/fan_speed.html

Notice that the waveforms in Fig 1 are identical to those in the
Speedfan article (a byte comparison of the GIFs confirms this).
Therefore it's possible that Speedfan's author may not have performed
any measurements of his own.

Notice also that the Analog Devices article states that High-Frequency
PWM "must use 4-wire fans".

- Franc Zabkar

 

[GIRunit]

it could be a faulty thermistor. its a rare occurance with computers,
but it happens.

 

[George Macdonald]

If a fan can really turn "~2400RPM at 30% drive", would it not be
reasonable to expect that it could turn more than 6000RPM at 100%
drive?

A 70 mm fan turning at 6000RPM? Think about it. The 2400RPM at 30% is IMO
the abnormal case - the other two turn at ~900RPM at 30%.

Could it be that Speedfan is maxing out at 3500RPM because the
HW monitor IC's count register is itself maxing out? Does changing the
divisor affect the range of the readings? Does Speedfan have a debug
mode where you can see the contents of the various registers? Assuming
the fan is a temperature controlled type, could it be that there is
poor thermal tracking between its thermistor and the heatsink?

It's not just SpeedFan - PCProbe showed the same weirdness in fan relative
speeds. The chipset fan reads to >6000RPM so I wouldn't think its a
problem with the sensor chip's counting ability; anyway I believe it's
established that the fan's built-in thermal control is supposed to track
intake air temp rather than heatsink. IOW it's supposed to compensate for
high room ambient temps where the inside case temp is ~40C.

 

[Franc Zabkar]

A 70 mm fan turning at 6000RPM? Think about it.

Please re-read the passage I have quoted immediately above.

The 2400RPM at 30% is IMO
the abnormal case - the other two turn at ~900RPM at 30%.

Maybe, maybe not. I have no technical data on the inner workings of
temperature controlled fans, but I would think that such a fan would
have its own internal PWM circuitry to drive the motor. If the reduced
voltage causes the fan's electronics to believe that the temperature
is high, then the PWM duty cycle would be 100%, which at 3.6V (12V x
30%) may only produce 2400RPM for a 6000RPM fan. At full supply
voltage, when the temp sensor is working properly, the motor turns at
its correct speed.

It's not just SpeedFan - PCProbe showed the same weirdness in fan relative
speeds. The chipset fan reads to >6000RPM so I wouldn't think its a
problem with the sensor chip's counting ability; anyway I believe it's
established that the fan's built-in thermal control is supposed to track
intake air temp rather than heatsink. IOW it's supposed to compensate for
high room ambient temps where the inside case temp is ~40C.

From my reading of the literature, I believe that any attempt to
control the speed of a standard brushless DC fan, using high frequency
PWM, is flawed. Furthermore I believe that your tests demonstrate that
a thermal fan *must* be run at 100% duty cycle. If I were you I would
use a 4-wire fan and wire it so that its motor gets an uninterrupted
power feed. Otherwise I would disable Q-Fan and forget about all your
other fan control utilities.

- Franc Zabkar

 

[Franc Zabkar]

If I were you I would
use a 4-wire fan and wire it so that its motor gets an uninterrupted
power feed.

Sorry, I should have written that the power to the fan's electronics
should not be interrupted.

- Franc Zabkar

 

[Franc Zabkar]

anyway I believe it's
established that the fan's built-in thermal control is supposed to track
intake air temp rather than heatsink.

Well, the following document suggests otherwise, although I can't see
how you could sense the processor's temperature unless the thermistor
were screwed down onto the heatsink.

Cool ‘n’ Quiet™ Technology Installation Guide for AMD Athlon™ 64
Processor Based Systems:
http://www.amd.com/us-en/assets/content_type/DownloadableAssets/Cool_N_Quiet_Installation_Guide3.pdf

"All AMD Athlon 64 Processor-In-A-Box packages include thermally
controlled fans. A thermally controlled fan detects the current
temperature of the processor using a thermistor, and when a lower
temperature is detected the fan speed and noise is then reduced. Upon
detection of a higher temperature the fan speed is resumed at full
speed to appropriately cool the processor."

Here is one fan with an external thermistor:
http://www.nidec.com/designoptions/thermalfan.htm

- Franc Zabkar