On Demand GPU Fan Control for GPU Parking

[UaVaj]

is there an application out there that will engage the gpu fan on demand?



window's ULPS is a great feature for multi gpu setup when not gaming. it literally turns off the other 3 idling slave gpu (aka gpu parking).

sadly it does not have a cool down cycle after gaming.

right after gaming. ulps simply stops the fan on all slave gpu. and that causes those gpu to becomes heat soaked.

like to run the fan for ~5 minute afterward to allow those gpu to cool down before entering ultra low power state.

 

[mikeymikec]

It depends on what you mean by 'on demand', surely you don't want complete manual control of the fan 100% of the time.

MSI AfterBurner allows you to specify your own fan profile, to specify particular fan speeds for particular GPU temps and establish a fan increase curve accordingly. I used to use it with my 5770 because the default 'idle' fan speed seemed about 1000RPM higher than it needed to be (ie. GPU temp didn't increase during idle after reducing the fan speed).

You can create other fan profiles, I suppose you could have one during gaming and activate another one after gaming.

 

[VirtualLarry]

Is ULPS a Windows thing, or a video-driver thing, or both?

 

[Flapdrol1337]

right after gaming. ulps simply stops the fan on all slave gpu. that causes those gpu to becomes heat soaked.

Not really, and not really a problem if it did.

 

[jkauff]

If you have a newer Asus motherboard and an Asus GPU, you can use AI Suite 3 to control the GPU fans.

 

[KingFatty]

Are you sure it takes 5-10 minutes for the GPUs to cool down? My experience saw the typical GPU cooldown being more like less than one minute to return to ambient temps.

 

[amenx]

Are you sure it takes 5-10 minutes for the GPUs to cool down? My experience saw the typical GPU cooldown being more like less than one minute to return to ambient temps.

It will vary from card to card. Not all cards share the same fan speeds or curves or thermal dissipation characteristics. Having said that, 5-10 min seems above the norm.

 

[utahraptor]

I believe it is common practice to have a racing pedal set up for GPU control. It is usually configured so that depressing it increases the clock speed as well as the fan speed within a configured range. When FPS drops you put the hammer down.

 

[Flapdrol1337]

I thought the fan was controlled with the throttle, if you slow down your gpu melts.

 

[utahraptor]

I thought the fan was controlled with the throttle, if you slow down your gpu melts.

Generally the throttle only controls overclocking beyond stock limits. With regard to fan speed, depressing the pedal simply adds a boost in RPM beyond the normal cooling profile. By preemptively cooling before the OC ups the temp stability can be preserved.

Of course if you want to be more advanced you can join the many users who have also incorporated a shifting stick to directly control the turbo increments.

 

[mikeymikec]

5-10 minutes does seem a bit much to me too; I wonder if there's dust in the HSF for the graphics card.

 

[UaVaj]

some of you are getting confused.



let me try again.

this has NOTHING to do with fan profiles while gaming.

this has ALL do with what the FAN is DOING when "back on the desktop" AFTER a HOT gaming session.

with ulps is enabled (ulps is a windows feature).
as soon as the game exit and back on the desktop. the fan on ALL slave gpu "STOPS".
those gpu were cooking hot just 1 second ago. and are now being heat soaked due to zero circulation.

still lost? just like a turbo on a car. after a hard run session. you let the car idle for a bit. to allow the turbo to cool down before turning off the car.



need a lite gpu application to keep the fan running at low speed or whatever speed to allow the gpu to cool down before the fan calls it quit per ulps.

btw. the setup is 290x x4.

 

[mikeymikec]

Sounds like it might be a driver bug (and/or a serious oversight). Normally fan profiles (default or otherwise) AFAIK run on a principle of "if temp = x make fan do y".

While 'parking' secondary GPUs sounds like a great idea for power saving, I would have thought that there would be a safeguard in, such as checking the GPU temp before parking the GPU, or even manage the fan speed independently of the GPU parking so that fan management speeds continue to respond to GPU temps regardless of circumstances.

Having said that, AMD might have checked the effects of parking a secondary GPU and its fan immediately after gaming and found that heat dissipation from the card in general was within acceptable boundaries because it had been powered down. Perhaps they also based it on the assumption that there would be chassis fans to help the secondary parked graphics cards to dissipate heat effectively.

I guess if I wanted to test whether AMD had properly tested heat dissipation on a parked card, I'd stick a temp sensor on a heatsink nearer a memory module and/or a power distribution point and monitor temp reduction after a game over say a period of twenty minutes (or more if the temp still hasn't dropped to something reasonably healthy).

I remember in the days of the P4/AMD wars I was afraid of a scenario whereby active cooling would be required for a period after the computer is shut down, but perhaps it never came to that or perhaps <insert physics/thermodynamics explanation here>.

 

[Flapdrol1337]

some of you are getting confused.



let me try again.

this has NOTHING to do with fan profiles while gaming.

this has ALL do with what the FAN is DOING when "back on the desktop" AFTER a HOT gaming session.

with ulps is enabled (ulps is a windows feature).
as soon as the game exit and back on the desktop. the fan on ALL slave gpu "STOPS".
those gpu were cooking hot just 1 second ago. and are now being heat soaked due to zero circulation.

still lost? just like a turbo on a car. after a hard run session. you let the car idle for a bit. to allow the turbo to cool down before turning off the car.



need a lite gpu application to keep the fan running at low speed or whatever speed to allow the gpu to cool down before the fan calls it quit per ulps.

btw. the setup is 290x x4.

A gpu is nothing like a turbo. A gpu is directly connected to the pcb, if it is hot it is always transferring that heat to the pcb. Same goes for all other components on the pcb.

If you shut everything down at the same time everything will slowly cool down.

You're trying to solve a problem that doesn't exist.

 

[PhIlLy ChEeSe]

is there an application out there that will engage the gpu fan on demand?



windows ULPS is a great feature for multi gpu setup when not gaming. it literally turns off the other 3 idling slave gpu.

sadly it does not have a cool down cycle after gaming.

right after gaming. ulps simply stops the fan on all slave gpu. that causes those gpu to becomes heat soaked.

like to run the fan for 5-10 minute afterward to allow those gpu to cool down before entering ultra low power state.

What multi GPU'S do you have? AMD has CCC, in it you can crank up the fan speed manually. And Nvidia cards you can use EVGA precision 17, also you can use MSI after burner which is the same application as precision. In them you can set the fan profile or as I do just crank up the fan

 

[utahraptor]

Sounds like it might be a driver bug (and/or a serious oversight). Normally fan profiles (default or otherwise) AFAIK run on a principle of "if temp = x make fan do y".

While 'parking' secondary GPUs sounds like a great idea for power saving, I would have thought that there would be a safeguard in, such as checking the GPU temp before parking the GPU, or even manage the fan speed independently of the GPU parking so that fan management speeds continue to respond to GPU temps regardless of circumstances.

Having said that, AMD might have checked the effects of parking a secondary GPU and its fan immediately after gaming and found that heat dissipation from the card in general was within acceptable boundaries because it had been powered down. Perhaps they also based it on the assumption that there would be chassis fans to help the secondary parked graphics cards to dissipate heat effectively.

I guess if I wanted to test whether AMD had properly tested heat dissipation on a parked card, I'd stick a temp sensor on a heatsink nearer a memory module and/or a power distribution point and monitor temp reduction after a game over say a period of twenty minutes (or more if the temp still hasn't dropped to something reasonably healthy).

I remember in the days of the P4/AMD wars I was afraid of a scenario whereby active cooling would be required for a period after the computer is shut down, but perhaps it never came to that or perhaps <insert physics/thermodynamics explanation here>.

I am glad you mentioned the driver errors. If you are manually shifting through turbo bins with a stick you may want to use gloves if you find your hand slipping off during intense gaming. That or cut back on the greasy snacks.

As for parking a hot card, a common error is to park the gpu over combustible materials at the end of a gaming session prior to it cooling down. I know I lost a rig that way.

 

[UaVaj]

GPU Parking is the correct descriptor.



looks like there is no lite application to keep the fan spinning for a short duration prior to gpu parking.


1) perhaps run an non demanding game to maintain cpu fan prior to gpu parking.
or
2) just disable ulps. (just have to de-dust the slave gpu more often)