Resolved How-To: FIX MAC PRO FAN SPEEDS when using PC (non-EFI) Graphics Cards

[h9826790]

@friolz: 5 minutes is pointless. You can do it for 1 second. The point is that the PC GPU starts in 3D acceleration mode when the computer powers on, and the Mac doesn't tell it to go into 2D mode. But starting and closing a 3D application makes the Mac tell the GPU to calm down and enter 2D mode, which makes the GPU go into power-saving mode which in turn generates less power draw on the PSU which in turn makes the Mac's miscalibrated fans run quieter.

There are two ways to fix the fan noise:

1. The way I proposed, which reconfigures the Mac fan speeds to be based on heat instead of how many watts they draw from the PSU.

2. The way you and many others proposed, which starts and closes a 3D application to get the GPU into low-power 2D mode which means less wattage which means slower Mac fan speeds. However, that method doesn't solve the noise you'll get if you play games or use any other 3D applications, since the power draw will then rise again and so will the Mac fan noise.

There is a third way, actually (which would benefit even people who use method 1):

3. nVidia could recode their Web Driver to put the GPU into 2D mode at startup, since macOS itself isn't smart enough to do it. This would save more electricity even for people who use method 1. And it would completely eliminate the need for method 2. Does someone want to take care of the task of contacting nVidia and trying to get them to do this?

I doubt if the problem is the GPU stay in 3D mode.

When the fan spin fast after boot, the power draw is actually low.

I personally use method 1. If your theory is right, the card will stay at 3D mode regardless I use a custom fan profile. So, it will draw more power and stay hot, and eventually spin up the fan. However, this is clearly not the case on my Mac. It looks more like a simple SMC bug that make the fan spin fast, and a little bit stress on the GPU can make the fan back to normal.

In fact, there is a 4th solution. Install the card at slot 2. If someone don't need all 4 slots, this is actually a simple solution.

 

[friolz]

@friolz: 5 minutes is pointless. You can do it for 1 second. The point is that the PC GPU starts in 3D acceleration mode when the computer powers on, and the Mac doesn't tell it to go into 2D mode. But starting and closing a 3D application makes the Mac tell the GPU to calm down and enter 2D mode, which makes the GPU go into power-saving mode which in turn generates less power draw on the PSU which in turn makes the Mac's miscalibrated fans run quieter.

There are two ways to fix the fan noise:

1. The way I proposed, which reconfigures the Mac fan speeds to be based on heat instead of how many watts they draw from the PSU.

2. The way you and many others proposed, which starts and closes a 3D application to get the GPU into low-power 2D mode which means less wattage which means slower Mac fan speeds. However, that method doesn't solve the noise you'll get if you play games or use any other 3D applications, since the power draw will then rise again and so will the Mac fan noise.

There is a third way, actually (which would benefit even people who use method 1):

3. nVidia could recode their Web Driver to put the GPU into 2D mode at startup, since macOS itself isn't smart enough to do it. This would save more electricity even for people who use method 1. And it would completely eliminate the need for method 2. Does someone want to take care of the task of contacting nVidia and trying to get them to do this?

That was a typo. I meant 5 seconds. The AppleScript reflects this (5000 ms).
[doublepost=1482518448][/doublepost]But: I have an R9 280X, which I flashed successfully. Shouldn't it be in 2D mode after system startup?

 

[friolz]

I'm sticking with method 2. Correct me if I'm wrong, but method 2 will keep the power consumption low.

 

[cdf]

In my opinion, method 2 is the best solution because it restores the intended fan control. Here are my comments on this matter:

For fan control, method 1 writes to the SMC chip in a way not intended by Apple. In the long run, this might be problematic because the chip can only be written to a finite number of times. There is also the question of properly setting the temperature parameters. This question, however, has been very well answered in this thread. On the other hand, method 2 brings back the intended fan control, and once the script runs, it is out of the way.

It is also possible that method 2, barring the initial stress, diminishes the power consumption of the card. This remains to be proved, though. For my video card, running a graphic-intensive app for one second is not enough; at least 30 seconds are required (whether it is with LuxMark or a CUDA computation). It is strange that this would be necessary to enter the lower power mode.

Putting the card in slot 2 indeed lowers the fan speed, but the speed is still higher than normal.

I believe that this fan issue is an SMC bug. And luckily, when the video card is in slot 1, we can recover from this bug if the SMC, on its own, sufficiently spins up the fans. The SMC spins up the fans as the PCI power consumption increases, which happens when stressing the card.

 

[h9826790]

In my opinion, method 2 is the best solution because it restores the intended fan control. Here are my comments on this matter:

For fan control, method 1 writes to the SMC chip in a way not intended by Apple. In the long run, this might be problematic because the chip can only be written to a finite number of times. There is also the question of properly setting the temperature parameters. This question, however, has been very well answered in this thread. On the other hand, method 2 brings back the intended fan control, and once the script runs, it is out of the way.

It is also possible that method 2, barring the initial stress, diminishes the power consumption of the card. This remains to be proved, though. For my video card, running a graphic-intensive app for one second is not enough; at least 30 seconds are required (whether it is with LuxMark or a CUDA computation). It is strange that this would be necessary to enter the lower power mode.

Putting the card in slot 2 indeed lowers the fan speed, but the speed is still higher than normal.

I believe that this fan issue is an SMC bug. And luckily, when the video card is in slot 1, we can recover from this bug if the SMC, on its own, sufficiently spins up the fans. The SMC spins up the fans as the PCI power consumption increases, which happens when stressing the card.

My understanding is that you won't write anything to the SMC chip. It's just a custom fan profile after loading to desktop. The parameter stay in the RAM, not the SMC chip.

And we create a custom fan profile because the native one is too low when the ambient is hot and the computer only under light load. In this case, the fan won't spin up at all because the power draw is low, however, both the PSU and the GPU can be quite warm.

On the other hand, if the ambient temperature is low, the fan is too noisy when the computer under stress. Because again, the fan speed only base on the power draw, but not the temperature.

That's why in this thread, we discuss how to use MacsFanControl to override the native fan setting. If you read the 1st page, you will realise the native fan setting can let the PSU run very hot. This is the main reason why we make our own fan profile.

 

[cdf]

My understanding is that you won't write anything to the SMC chip. It's just a custom fan profile after loading to desktop. The parameter stay in the RAM, not the SMC chip.

This is also my interpretation. The control parameters that we set in the fan control application are not written to the SMC. However, the question remains whether the application writes to the SMC to take the fan control out of automatic mode and to adjust the fans according to the profile, or just issues instructions and nothing gets written.

In any case, thanks to the information posted in this thread, we have two viable methods (each with their advantages) that we can choose from.

 

[h9826790]

This is also my interpretation. The control parameters that we set in the fan control application are not written to the SMC. However, the question remains whether the application writes to the SMC to take the fan control out of automatic mode and to adjust the fans according to the profile, or just issues instructions and nothing gets written.

In any case, thanks to the information posted in this thread, we have two viable methods (each with their advantages) that we can choose from.

I am 99% sure nothing get written to the chip (but don't know how to prove).

MacsFanCobtrol just set a target RPM by midifing SMC key, you can do exactly the same thing in terminal (with the help of SMCfancontrol).

If the software write something to the SMC chip, the setting should stay there after software uninstallation (or disable auto load at start up), but it's clearly not the case.
[doublepost=1482587860][/doublepost]In fact, I wonder if writing the SMC chip is that easy, because this may open the window to use a 5,1 dual CPU tray on the 4,1.

 

[SteveJobzniak]

I doubt if the problem is the GPU stay in 3D mode.

When the fan spin fast after boot, the power draw is actually low.

I personally use method 1. If your theory is right, the card will stay at 3D mode regardless I use a custom fan profile. So, it will draw more power and stay hot, and eventually spin up the fan. However, this is clearly not the case on my Mac. It looks more like a simple SMC bug that make the fan spin fast, and a little bit stress on the GPU can make the fan back to normal.

In fact, there is a 4th solution. Install the card at slot 2. If someone don't need all 4 slots, this is actually a simple solution.

I remember reading that all graphics cards have 2 circuits, one for 2D and one for 3D, separated like that to allow the cards to save electricity when 3D isn't needed. And the cards start out in "All on!" mode for compatibility. And that it's up to the OS to tell the card to disable the 3D to save power.

Someone somewhere sometime (see, I can't remember the source) said that macOS doesn't tell the PC card to disable the 3D circuit. So the power draw is high enough to trigger the fans.

The 3D app method doesn't stress anything, especially not when only ran for 5 seconds (not enough time to generate heat), so it doesn't make any sense that the "heat stress" is the reason why starting 3D mode and quitting 3D mode is silencing the fans. It makes more sense that leaving the 3D app tells the driver to tell the card to shutdown the 3D circuits to save electricity, which the SMC detects as less need to run the fans high.

This is especially likely considering that low-energy PC graphics cards don't have this fan speed problem. Those generate as much HEAT, but less WATTAGE. And they don't spin up the fans. So it does seem to be the wattage that SMC bases its fan speed on.

As for Slot 2 not triggering the fans, that may be because they don't have wattage sensors in slot 2? Perhaps Apple assumed the graphics card is in slot 1 and only put high-power sensors there, since historically that was the only 16x PCIe slot? I don't know.

Anyway I am about to sell my Mac Pro so I can't investigate. I'm a MacBook Pro user now.

 

[MriX]

Does someone know if that fan speed issue only happens with Nvidia cards?

Now I have a GTX680 Mac flashed installed on my MacPro4,1 and I have this problem. I would be interested if a flashed R9 280 also has this fan issue

 

[h9826790]

Does someone know if that fan speed issue only happens with Nvidia cards?

Now I have a GTX680 Mac flashed installed on my MacPro4,1 and I have this problem. I would be interested if a flashed R9 280 also has this fan issue

Both Nvidia and AMD cards are affected.

 

[bookemdano]

This is another "bug" that can and should be reported by Mojave users using the "recommended" Sapphire RX 580 and MSI RX 560 cards

 

[h9826790]

This is another "bug" that can and should be reported by Mojave users using the "recommended" Sapphire RX 580 and MSI RX 560 cards

The problem is this bug seems only happen on the 4,1 (including flashed to 5,1), but not the genuine 5,1. It may be not quite appropriate to tell Apple that "we hit a bug because we perform an unofficial firmware upgrade to run a non-supported OS"

 

[star-affinity]

The problem is this bug seems only happen on the 4,1 (including flashed to 5,1), but not the genuine 5,1. It may be not quite appropriate to tell Apple that "we hit a bug because we perform an unofficial firmware upgrade to run a non-supported OS"

Fans run fast for me too on my 5,1 after a cold start, but it goes away after some time of use or if using a graphically demanding app.

 

[cdf]

Since some real 5,1's are also affected by this bug, I would like to emphasize the request that Mojave testers with recommended cards report it to Apple. Maybe we can finally get this issue resolved!

 

[star-affinity]

Since some real 5,1's are also affected by this bug, I would like to emphasize the request that Mojave testers with recommended cards report it to Apple. Maybe we can finally get this issue resolved!

Is it happening also when using an eGPU with a Thunderbolt 3 Mac?

 

[MriX]

Both Nvidia and AMD cards are affected.

I also have a MSI GT 730 (Kepler Edition and passive cooled) and this one does not have the fan bug.
So not all cards seem to reproduce this bug.

 

[misanthrophy]

Just need your opinion.

My cMP is:

MP4.1 flashed to 5.1 running High Sierra.
2x 3,46GHz CPU's with Liquid Metal as "thermal paste" and Stock Cooling System.
All 4 bays loaded with HDD's.
Right now stock GTX 120 as GPU installed (Nvidia GeForce GTX Titan X comes in next week)
PCIe card with NVMe SSD for the system.

When I am running Handbreak on my MP both CPU's are working very hard and for my fan profile they are at 63-70C maximum. Thats fine, the still remain silent even though they are running at 3275RPM (CPU A -- 70C), and ~1325RPM (CPU B -- 63C). I am unable to get them both at the same temperature without spinning the BOOST A fan to the maximum which makes it loud. Even in idle there is a difference of up to 5C, but as seen in other screenshots, that seems okay, because I am not the only one.

After reading that thread here carefully, I changed the fan behaviour of the power supply fan to get down to under 40C.

Before that fan profile, the Power supply was at around 50C and more in auto mode with this work load (without even the Titan X installed) and the fan was spinning with 600RPM.

I tried to manually increase the RPM to see when I can hear it, so until 950 it is still fine, but the PSMI Supply AC/DC Supply 2 was still above 40-45C.

So I changed it to a sensor based method. Setting is now at 35C -- 45C to let the fan spin faster.

So after ~1h of Handbreak encoding with that setting my PSU is at ~39C with ~1250RPM. That means I can really hear the fan. I mean it is still acceptable, but would you say, I can change the settings to higher temperatures which will lower the RPM's? Or is my PSU just dusty and cleaning it (removing the whole PSU and clean it with compressed air) should do the job and lower it automatically.

PS: Ambient Temperature is ~30C.

 

[h9826790]

Just need your opinion.

My cMP is:

MP4.1 flashed to 5.1 running High Sierra.
2x 3,46GHz CPU's with Liquid Metal as "thermal paste" and Stock Cooling System.
All 4 bays loaded with HDD's.
Right now stock GTX 120 as GPU installed (Nvidia GeForce GTX Titan X comes in next week)
PCIe card with NVMe SSD for the system.

When I am running Handbreak on my MP both CPU's are working very hard and for my fan profile they are at 63-70C maximum. Thats fine, the still remain silent even though they are running at 3275RPM (CPU A -- 70C), and ~1325RPM (CPU B -- 63C). I am unable to get them both at the same temperature without spinning the BOOST A fan to the maximum which makes it loud. Even in idle there is a difference of up to 5C, but as seen in other screenshots, that seems okay, because I am not the only one.

After reading that thread here carefully, I changed the fan behaviour of the power supply fan to get down to under 40C.

Before that fan profile, the Power supply was at around 50C and more in auto mode with this work load (without even the Titan X installed) and the fan was spinning with 600RPM.

I tried to manually increase the RPM to see when I can hear it, so until 950 it is still fine, but the PSMI Supply AC/DC Supply 2 was still above 40-45C.

So I changed it to a sensor based method. Setting is now at 35C -- 45C to let the fan spin faster.

So after ~1h of Handbreak encoding with that setting my PSU is at ~39C with ~1250RPM. That means I can really hear the fan. I mean it is still acceptable, but would you say, I can change the settings to higher temperatures which will lower the RPM's? Or is my PSU just dusty and cleaning it (removing the whole PSU and clean it with compressed air) should do the job and lower it automatically.

PS: Ambient Temperature is ~30C.

If I were you. I won’t try to make both CPU run at the same temperature. That’s meaningless. The down stream CPU of course will run warmer even the NB is not there. All we need is just “within limit”, and there is no max temperature difference limit between the two CPU. Then why care about that?

Your PSU temperature is fine. Even though it may be a bit dusty inside.

 

[misanthrophy]

If I were you. I won’t try to make both CPU run at the same temperature. That’s meaningless. The down stream CPU of course will run warmer even the NB is not there. All we need is just “within limit”, and there is no max temperature difference limit between the two CPU. Then why care about that?

Your PSU temperature is fine. Even though it may be a bit dusty inside.

I just want to make sure, my system will run for as long as possible while doing heavy loads of work, so I try to optimize everything I can, haha.

But as many other people are also facing the difference of a few Celsius, I guess I am fine.

Will try again to clean it properly and see if that will do the job.

Thanks in advance. At least I know my system is within the range where it runs safely.

 

[MriX]

Yesterday my MSI RX560 arrived and I have no higher fan speeds.
So for me the “smc issue” only appears when I install a graphics card (Mac or PC) which requires additional power cables.

 

[friolz]

I also have the MSI RX560, but the fan speeds are higher.

 

[Drolma-la]

Does someone know if that fan speed issue only happens with Nvidia cards?

Now I have a GTX680 Mac flashed installed on my MacPro4,1 and I have this problem. I would be interested if a flashed R9 280 also has this fan issue

I recently replaced the original GPU in my MacPro5,1 with a (secondhand) Sapphire AMD HD7950 Mac Edition GPU. Everything seemed fine, but as I shut the computer down the first time after the installation the fan went bananas. Horrific racket. Tried everything (serious dust eradication, running 3D graphics-intensive test programs). What finally worked was moving the card from slot 1 to slot 2.

So the fan speed issue happened with my original MacPro5,1 and an AMD card theoretically cleared for Mojave.

 

[guido4]

Hi,

I wanted to share my experience on this issue. I'm using a Mac Pro 5.1, dual X5690, 96GB RAM, 512 GB SSD, three monitors, and a flashed GTX 780.

When I'm booting this setup to Mavericks, fans are running quietly on minimum speed. Even when rendering complex 3D images with Blender, fans stay on lowest speed level. The first screen shot shows fan speed under Mavericks after the computer has been running for 15 mins and then 5 mins of intensive 3D rendering.



So PSU temperatures of 42°C or PCIE ambient of 40°C seem to be no reason for Mavericks to turn on the fans on a cMP.

Now when booting the same system to El Capitan, PCI and PS fans speed up within a minute, although no work was done at all. PSU Temp is 27°C and PCIE Ambient 25°C, still PS fan is over 2.200 and PS fan over 1.400 U/min.



So I have the impression that fan speed regulation of a cMP is depending on the MacOS version you are running.

Kind regards
Guido

 

[mrtang42]

The problem is this bug seems only happen on the 4,1 (including flashed to 5,1), but not the genuine 5,1. It may be not quite appropriate to tell Apple that "we hit a bug because we perform an unofficial firmware upgrade to run a non-supported OS"

Try to add a point that reference Vega 56 does not have this issue at all on my flashed 5,1(4,1 originally). It only happens to RX 580 which I have tried both sapphire nitro+ and sapphire pulse.

 

[cdf]

Try to add a point that reference Vega 56 does not have this issue at all on my flashed 5,1(4,1 originally). It only happens to RX 580 which I have tried both sapphire nitro+ and sapphire pulse.

Is there a pixlas mod in this equation?