Talk:Problem with fan noise
- 1 Problem with fan noise on R51 1829 L7G (ATI M9)
- 2 Thinkpad T42 Radeon Mobility M7
- 3 Thinkpad R32 with Radeon Mobility M6
- 4 Fan Control script: more safe version
- 5 X41
- 6 Fan speed control?
- 7 Further discussion
- 8 Works fine with APM instead of ACPI?
- 9 Rewiring the fan?
- 10 Secret sensor and the cause of fan always on
- 11 X21 overacceleration problem
- 12 Startup noise
- 13 Leftover bits
- 14 Embedded controller firmware disassembly
- 15 R50e with CELERON M: fan for life
Problem with fan noise on R51 1829 L7G (ATI M9)
On my R51 the fan is behaving like this:
- > 45C -> fan on;
- < 38C -> fan off.
By using cpufreq + laptop_mode + Xorg DynamicClocks + WiFi power management, I get the fan stopped time to time, but only for 3 minutes time (transition from 38 C -> 45 C). The cooling down cycle is taking 20 minutes in the best case.
I knew about the 'ibm_acpi experimental=1' trick, but in my opinion this is not very useful since nobody can guarantee that a temperature greater then 45 C will not damage the laptop and in the same time the transition time is very short (the laptop gets hot fast without fan).
Thinkpad T42 Radeon Mobility M7
When Xorg is running, the fan is always on and pretty loud ! Setting DynamicClocks does not help
it's clear that the GPU is the problem on the thinkpad :
after 10minutes with the fan off temperatures: 44 47 33 52 32 -128 24 -128
1: CPU 2: Mini PCI Module 3: HDD 4: GPU 5: Battery 6: N/A 7: Battery 8: N/A
Controlling the fan speed would be really cool !
What is the maximum temperature not to cross ?
Word on the 'net is that 85 degrees is the max operating temp for most of the Intel chips. I've seen some high 70's all the time (just put it on carpet for awhile and play some quake3 :). I wouldn't let your processor get much higher than 85...
"Word on the net" is usually synonym to bullshit. The datasheets are available for download at Intel's site, go read them. For the P4, 80C is the absolute maximum limit, and in the newer Pentiums, the chip itself will MCE and enable clock modulation if you try to make it exceed 80C. I didn't read the new Pentium-M's datasheet yet.
Letting the machine run hot is not a good idea, anyway. It shortens its lifespan considerably, especially the harddrive's. Don't play games with this.
Older versions of xorg (i.e. 6.7.0) don't seem to be able to use the DynamicClocks option although it's set in the xorg.conf. Search the log to find out if it's really used.
Thinkpad R32 with Radeon Mobility M6
Updating xorg-x11 from 6.7.0 to 6.8.2 and using Speedstep (with the ondemand module in this case) helped cooling the system down significantly:
- before updating the CPU was ~62 C in idle state, and got very near the critical temperature (72 C) during heavy load - I even got some freezes because of the heat ;)
- after the update the CPU is ~54 C in idle state, and still gets to about 68 C while under heavy load
The second sensor (which may be the GPU) is somehow fixed to 50 C (maybe a bug?)
The fan on the R32 is behaving like this:
- > 61 -> fan in state 2 (quite noisy)
- < 55 -> fan in state 1 (less noisy :) )
But I remember using my old SuSE distribution with kernel 2.4.16, apm and some old x11 version the fan actually stopped completely from time to time.
Concerning the maximum temperature of the CPU, I found that the critical temperature on the R32 for the CPU sensor is 72 C
# cat /proc/acpi/thermal_zone/THM0/trip_points )
Fan Control script: more safe version
ibm_acpi works well on my R50 and R51. But to rely on it completely, I modified the script in two ways:
1. It catches verious signals and turns the fan on before it quits
2. It turns off the fan under very strict conditions, leaving it on when unexpected errors occur.
Here is my script:
#!/bin/sh # july 2005 Erik Groeneveld, firstname.lastname@example.org # More conservatiev and saver version # It make sure the fan is on in case of errors # and only turns it off when all temps are ok. IBM_ACPI=/proc/acpi/ibm THERMOMETER=$IBM_ACPI/thermal FAN=$IBM_ACPI/fan MAXTRIPPOINT=65 MINTRIPPOINT=60 TRIPPOINT=$MINTRIPPOINT echo fancontrol: Thermometer: $THERMOMETER, Fan: $FAN echo fancontrol: Current `cat $THERMOMETER` echo fancontrol: Controlling temperatures between $MINTRIPPOINT and $MAXTRIPPOINT degrees. # Make sure the fan is turned on when the script crashes or is killed trap "echo enable > $FAN; exit 0" HUP KILL INT ABRT STOP QUIT SEGV TERM while [ 1 ]; do command=enable temperatures=`sed s/temperatures:// < $THERMOMETER` result= for temp in $temperatures do test $temp -le $TRIPPOINT && result=$result.Ok done if [ "$result" = ".Ok.Ok.Ok.Ok.Ok.Ok.Ok.Ok" ]; then command=disable TRIPPOINT=$MAXTRIPPOINT else command=enable TRIPPOINT=$MINTRIPPOINT fi echo $command > $FAN # Temperature ramps up quickly, so pick this not too large: sleep 5 done
I added this script to the other ones. Don't wander about my talk edits, i didn't realize i was on the talk page. Wyrfel 01:48, 13 Aug 2005 (CEST)
Same fan problem here on the X41. Once it starts it won't stop (unless it is _very_ cold outside). Undervolting the CPU doesn't help - still the same problem.
Fan speed control?
Only the X31 and X40 have an ACPI method for controlling the FAN speed (this is why ibm_acpi provides this functionality just for these models).
What will happen if we take the "FANS" method from the X40 DSDT, paste it into a iasl-disassembled DSDT of (say) a T43, recompile it and tell the kernel to use the patched DSDT? ibm_acpi will present the functionality, but it may or may not work.
--Thinker 16:16, 28 Sep 2005 (CEST)
Any risk of damaging the hardware when doing this? E.g. what does occur if the system overheats - will the CPU be destroyed are does it automatically switch of? As I've just bought a new X41 I don't want to take any stupid risks - but otherwise I'd say let's try it out.
--gst Thu Sep 29 18:14:13 CEST 2005
I think Intel CPUs have some built-in thermal protection, but I'd hate to test it. And of course, any fiddling with the hardware at this level might damage it. That said, when the CPU is mostly idle it keeps a reasonable temperature even when the fan is disabled, so as long as you keep an eye on both the CPU usage meter and /proc/acpi/ibm/thermal, things should be pretty safe temperature-wise. For extra safety you can force the CPU to its lowest speed via /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq. --Thinker 18:33, 29 Sep 2005 (CEST)
Anybody tried to slow down the fan and have it running at ~1000-1500rpm? That would make it almost silent and you could have it always running (like Apple does on Macs).
--Micampe 12:56, 7 Dec 2005 (CET)
Goot question. The embedded controller interface we have discovered allows only for setting a fan level between 0 and 7, not the actual RPM. And IBM has been ramping up the RPM assignment of each fan level (look at the model-specific data there), so that on the T43, even the lowest accessible level is at a pretty fast 3300RPM. The following might provide a way around this, if the hardware is capable of generating lower speeds:
- Patch the embedded controller firmware. You will need the information in this post and the following one, a lot of patience, and an active warranty.
- Find out how the embedded controller is setting the speed (maybe by disassembling it as above, but at least you don't need to load patched firmware). Maybe the EC is changing the speed by instructing another component over an interface (SMBus?) that can also be accessed directly by the CPU. In this case, maybe you can put the embedded controller in disengaged mode and control that component directly.
Then, there's the soldering-iron approach. Assuming the fan uses the standard wiring and that its speed determined by the provided voltage, you can make all speeds somewhat slower simply by putting a resistor in series with the fan. You can do that by splicing one of the wires or, to avoid voiding your warranty, just build an extension cable like this:
fan cable extension #------- Vfan --------#> >#---~~R~~---#> >#SYSTEM FAN#------- SENSE -------#> >#-----------#> >#BOARD #------- GND ---------#> >#-----------#> >#CONNECTOR
Use an appropriate resistor R, and make sure it doesn't get too hot (if it does, attach it to the fan's cooling assembly). Worked nicely for me a few years ago on some desktop motherboard. If you're into that kind of stuff, you can also use fancier electronics instead of the resistor to avoid the (minor) power waste.
It would be useful to have some pictures of the fan cable connector so we can get a matching pair (it's above the rear right corner of the PCMCIA slot, under the palmrest but maybe visible with just the keyboard removed).
--Thinker 14:36, 7 Dec 2005 (CET)
I've just found a very interesting thread regarding the same issue on HP notebooks. IMO it provides many insight information about heat/fan problems in general, the URL is: http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=853249 Especially the posts by the HP engineer "Andy Fisher" are very interesting. IBM should be able to provide the same BIOS fix as HP did (maybe I should have bought an HP notebook instead of a Thinkpad?).
I've also contacted IBM/Lenovo support via the website about the fan issue. Maybe it helps when others do this as well (especially people who bought larger quantities) so that this issue is taken serious by Lenovo. Is there already any official response to this problem?
--gst Thu Sep 29 19:40:34 CEST 2005
Two of the changes mentioned by the HP engineer make perfect sense here: raise the low trip points and make speed transition gradual. Oh, and get rid of the annoying beat pattern (a brief speed pulse every few seconds) it sometimes gets into!
But from our perspective, what would probably be best is to do the whole thing in software, providing the flexibility for personal preferences and smart decisions. The hardware would only enforce emergency override or throttle/shutdown for extreme temperatures. Then we could do cute things like having a software daemon lower the thresholds in a noisy environment (as judged using the built-in microphone) or when the laptop is on the user's lap (as judged by the built-in accelometers).
--Thinker 18:47, 30 Sep 2005 (CEST)
I noticed that on my T43 the fan is usually in one of two modes, low speed (around 3300 RPM, triggered around CPU=47deg) and medium speed (around 4100 RPM, can't figure out the trip condition). The former is nearly inaudible, but the latter is quite noticable in the absense of strong background noises.
Now, the problem is that once it has tripped into medium speed, it usually never comes back to low speed until the next reboot. So once it happens, to quiet things down I can only run one of the fan-disabling scripts given here. But with a disabled fan the T43 is not thermally stable, so it will spend its time moving back and forth between the hysteresis thresholds, i.e., toggling between 4100 RPM and 0 RPM every few minutes. This is quite silly and annoying, when staying at low speed would be both more stable and more quiet.
I hope someone will find a way to control the fan speed, or at least to reset the embedded controller's hysteresis state.
--Thinker 10:29, 6 Oct 2005 (CEST)
When you do changes to e.g. the Energy Schema in Windows or you eject the Thinkpad of the Docking Station it seems that the controllers state is rest. At least on the X41 the fan does stop until it reaches the threshold to start some minutes later. So it should be doable. --18.104.22.168
That's good. But just like a bunch of other functions (e.g., controlling the battery charge threshold), it probably uses low-level undocumented proprietary interfaces which are very hard to figure out without the help of IBM/Lenovo, who are in denial about the whole thing. --Thinker 01:40, 16 Oct 2005 (CEST)
Works fine with APM instead of ACPI?
On my X41 the fan starts after about 10 minutes of use and doesn't stop (until it is rather cold in my room - and even then it runs most of the time ;) A friend of mine who has a X41 too (though another model) and who does use NetBSD and APM doesn't experience this problem. He claims that the fan only comes up if the system is not idle. So either it is colder in his room, the X41 model which he has doesn't have this flaw or APM does use different tresholds than ACPI.
- Then why not just try the
acpi=offkernel parameter and see what happens? --Thinker 18:14, 30 Sep 2005 (CEST)
I currently don't have physical access to the X41. Will try in a few days.
Rewiring the fan?
Since IBM/Lenovo shows no intention of fixing their embedded controller firmware or releasing its specs, how about getting the embedded controller out of the loop? I'd be happy as a clam if my fan was hard-wired to work at a constant 3000RPM, with temperatures kept at bay in software through CPU frequenty control.
Assuming the fan has the standard 3-wire connector, we can probaby keep the sensor and ground wires untouched, and rewire the positive wire to some nearby current source of appropriate voltage (through a resistor, for fine-tuning). The trick would be to find an easily tappable source that can handle an extra 2W and has the appropriate voltage (i.e., just slightly higher than what the fan needs to rotate at that RPM, so we don't waste too much energy in the resistor). Any idea what are the typical fan voltages and what would be an appropriate hookup point?
--Thinker 01:59, 16 Oct 2005 (CEST)
Secret sensor and the cause of fan always on
This discussion was moved to Talk:Thermal sensors.
X21 overacceleration problem
On my X21 2662-66U, the fan runs all the time. That's not an issue, because of how hot my laptop runs.
However, at load, it runs really hot. I can hear the fan attempt to accelerate past the maximum speed, click three times, and drop back down to the max. It's not the same as the pulsing problem, yet similar.
New fan, ordered direct from IBM, due to my old fan running WAY too hot, even underclocked. As it is, I'm running underclocked because 95 C at full load is just NOT my cup of tea. 40 C idle temps at 500MHz, around 60 at 700MHz. 70 C load at 500, 95 C load at 700.
My T43 accelerates the fan to very high speed twice during startup: immediately after power-on, and during the bootloader prompt (i.e., after the BIOS but before Linux loads). This is independent of the unit's temperature, and is loud enough to cause some ugly stares when sitting in a lecture. Is this a universal phenomenon, worth mentioning in the article page?
Potential solutions: the first fan acceleration can't be fixed by anything but an EC firmware update. For the second acceleration (after BIOS), a workaround can be implemented by adding basic fan control code (i.e., a few accesses to I/O ports 0x62 and 0x66 as explained in the fan control specs and Linux's drivers/acpi/ec.c) to the bootloader. GRUB 2 has a module infrastructure that should make this pretty easy, and other boot loaders can also be patched.
--Thinker 20:36, 27 Dec 2005 (CET)
The following data didn't belong where it was posted, and was deleted when the containing section swre moved to How to control fan speed:
- On my T41 (gentoo-sources-22.214.171.124 ) I noticed that after unloading the fan module the fan noise stopped. With the module loaded the fan was working even at very low cpu temperatures, without the module it's ok so far.
- On a 770X the fan can be fully controlled through ACPI. Thermal Zone THM5 (possibly the battery/charging circuit, it's definitely warmer when using 5v PCMCIA cards and AC) triggers it to turn on and not ever off by design. Can be solved by a custom DSDT, which also makes use of the variable-speed features of the fan (will release this once I've finished tweaking and testing it).
--Thinker 20:48, 20 January 2006 (CET)
Embedded controller firmware disassembly
This thread points to, and discusses, a commented (partial) disassembly of the embedded controller. Following this lead, it may be possible to patch the firmware (perhaps transiently) to change the fan level selection algorithm and/or the RPMs.
--Thinker 07:10, 24 February 2006 (CET)
R50e with CELERON M: fan for life
I am such a loser. I bought a R50e with CELERON M CPU, and thus will never get longer than 90 seconds of peace before the next 62 to 54 degree fan cycle. --jidanni