MP 1,1-5,1 Control GPU fan - R9 Nano

[Scrotorr]

In a couple of my Mac Pros I have R9 Nanos as the GPU. I haven't been able to get them working with SMBIOS updates to get full acceleration - but that's not a huge deal. A bigger deal is the fan speed not adjusting. It's staying way too low, so under load I'm getting temps on the gpu above 72C. I can crank the PCI fan speed and exhaust fan speed to help compensate, but that's enough to drive me nuts.

If nothing else could the curve be updated in the GPU BIOS?

Thanks!

 

[h9826790]

In a couple of my Mac Pros I have R9 Nanos as the GPU. I haven't been able to get them working with SMBIOS updates to get full acceleration - but that's not a huge deal. A bigger deal is the fan speed not adjusting. It's staying way too low, so under load I'm getting temps on the gpu above 72C. I can crank the PCI fan speed and exhaust fan speed to help compensate, but that's enough to drive me nuts.

If nothing else could the curve be updated in the GPU BIOS?

Thanks!

72°C is quite cool for GPU that under load.

 

[KeesMacPro]

under load I'm getting temps on the gpu above 72C. I can crank the PCI fan speed and exhaust fan speed to help compensate, but that's enough to drive me nuts.

Based on personal experience (and ofcourse by reading here) a few things you could consider concerning the airflow in the cMP:
- It seems to be most common to install the GPU in slot1 . This way the backplate/ pcb of the GPU is completely isolated: there will be no airflow underneath at all. Installing the GPU in slot 2, it will be in the center of the PCI fan airflow, cooling down both sides of the GPU.
This will lower the GPU Temps .

- I assume increasing the Exhaust fan rpm will not lower the GPU T significantly, since it's in another compartment.
The PSU fan though, blows air through the PSU compartment (obviously) , but also through a couple of slots above the SATA bays, resulting in an increased airflow around the PCI slots (especially PCI slots 3/4).
Especially in case of an empty SATA bay (3 and) 4, increasing the PSU rpm might lower the Temps too.

 

[h9826790]

- It seems to be most common to install the GPU in slot1 . This way the backplate/ pcb of the GPU is completely isolated: there will be no airflow underneath at all. Installing the GPU in slot 2, it will be in the center of the PCI fan airflow, cooling down both sides of the GPU.
This will lower the GPU Temps .

This doesn't fit my own test result. The graphic card install in slot 1 run coolest.

The card I used to test is a 1080Ti which has back plate.

I suspect that's due to hot air rise. Therefore, the higher the slot you use, the warmer the air to cool down the GPU.

From memory, I never see any back plate is designed to cool down the GPU (it may help to cool down the PCB, if that's not for pure protection). The GPU is always cool down by the main heatsink. Increase airflow on the back side won't help.

 

[KeesMacPro]

This doesn't fit my own test result. The graphic card install in slot 1 run coolest.

The card I used to test is a 1080Ti which has back plate.

I suspect that's due to hot air rise. Therefore, the higher the slot you use, the warmer the air to cool down the GPU.

From memory, I never see any back plate is designed to cool down the GPU (it may help to cool down the PCB, if that's not for pure protection). The GPU is always cool down by the main heatsink. Increase airflow on the back side won't help.

That's interesting!

Did you test it without another heatsource in slot1 ?

The 1080TI is a GPU with a big heatsink.
Anyway, i tested it on a GTX 780TI Reference (and the good old GT120 a while ago).
In case of a GPU with a small heatsink e.g. R9 NANO or the GT120 the results could be different .

I agree with you that the heatsink is meant to cool down the GPU , but since the PCB is part of the surface of the compartment enclosed by the cover of the GPU , I assume it might have some influence on the T.
Besides the heatsink is mainly for the processor , so if we look at the Temps of only the die (in e.g. MFC), the T of the pcb will be (almost) irrelevant.

About the function of the backplate : I suppose so too :they're (at least in most cases) for optical reasons/protection.
I do remember though having read about backplates installed with heatpads for e.g. the regulators , RAM or so...

BTW: the T of the pcb might not influence that much on the die itself, but will on other parts on the pcb.

 

[h9826790]

That's interesting!

Did you test it without another heatsource in slot1 ?

Anyway, i tested it on a GTX 780TI Reference (and the good old GT120 a while ago).

I agree with you that the heatsink is meant to cool down the GPU , but since the PCB is part of the surface of the compartment enclosed by the cover of the GPU , I assume it might have some influence on the T.

About the function of the backplate : I suppose so too :they're (at least in most cases) for optical reasons/protection.
I do remember though having read about backplates installed with heatpads for e.g. the regulators , RAM or so...

BTW: the T of the pcb might not influence that much on the die itself, but will on other parts on the pcb.

My test was carried out without another card in slot 1. But 4x HDD in all 4 bays. That may also makes the PCIe compartment warmer (especially in higher slot).

Some back plate may help to cool down the VRAM, but that should has little to no help to cool down the GPU. In case of R9 Nano, the HBM can only be cooled by the main heatsink.

 

[KeesMacPro]

My test was carried out without another card in slot 1. But 4x HDD in all 4 bays. That may also makes the PCIe compartment warmer (especially in higher slot).

Yes, I replaced all HDDs a long time ago for blades and SSDs.
It does heat up more with spinners!

 

[Scrotorr]

Thanks for all the input and discussion! I'll keep an eye on the card and maybe swap it for the Fury X I have lying around to see if that's any better.

@h9826790 - how are your temps on the Radeon VII. I just picked one up but I haven't decided if I'm keeping it in my Windows rig or putting it in my home Mac Pro. My Windows rig only has a 4790k, while my Mac Pro has dual x5690 - so there's not a huge improvement one way or the other - Better single thread on the 4790k but much better multithreaded on the dual x5690. I suppose I could toss Windows on the Mac Pro too...

 

[h9826790]

Thanks for all the input and discussion! I'll keep an eye on the card and maybe swap it for the Fury X I have lying around to see if that's any better.

@h9826790 - how are your temps on the Radeon VII. I just picked one up but I haven't decided if I'm keeping it in my Windows rig or putting it in my home Mac Pro. My Windows rig only has a 4790k, while my Mac Pro has dual x5690 - so there's not a huge improvement one way or the other - Better single thread on the 4790k but much better multithreaded on the dual x5690. I suppose I could toss Windows on the Mac Pro too...

I fine tuned my Radeon VII's voltage to reduce power draw, and makes it to run cooler.

How to - downvolt a Radeon VII

Just upgraded from RX580 to Radeon VII (it's just $350 from a local seller, like new condition, with 2 years+ warranty. I really can't skip it :D) Anyway, as usual, all AMD GPU are seriously overvolted from factory (Default setting in Wattman shows 1801MHz @1093mV). Properly downvolt the card...

forums.macrumors.com

I also replaced the stock graphite pad by liquid metal. TBH, this is more for fun, the factory graphite pad on the Radeon VII is really good. Liquid metal can only reduce the peak temperature by 3°C. For my PNY XLR 1080Ti, liquid metal can reduce the GPU temperature by ~15°C.

Anyway, the temperature scale on the Radeon VII is different from the Nano. Because the sensors are at different locations. On the Radeon VII, everything is base on the junction temperature (AKA hot spot temperature), not the traditional edge temperature. Which makes the Radeon VII can easily run at 100C, that’s completely normal (when under stress).

However, by downvolting it, I can reduce the GPU temperature by about 20°C, but let it perform better.

This is what It get when running Luxmark stress test.

The GPU hot spot temperature stabilised at 79°C, which is same as 66°C in the traditional temperature scale. Which allow the GPU clock speed to stay above 1700MHz, and HBM2 were overclocked to 1200MHz.

 

[Scrotorr]

That's good to know - I'll have to look at downvolting mine. I was a little worried about the temps seeming so high, but that makes sense if they're using junction temp.