Underclocking GPU - HD 7970

St1ckM4n

Posts: 2,887   +628
Heyo,

I have a factory overclocked HD7970, running at 950MHz core (or 1GHz in the 2nd BIOS). I want to lower my temps, so I will be experimenting with underclocking. I'll be posting my results here.

Feel free to chime in if you have any tips or knowledge.
 
Heyo,

I have a factory overclocked HD7970, running at 950MHz core (or 1GHz in the 2nd BIOS). I want to lower my temps, so I will be experimenting with underclocking. I'll be posting my results here.

Feel free to chime in if you have any tips or knowledge.
If you don't mind me asking what are you current temps and are they so excessive you feel you need to do that or is it just too loud for your taste? Is it a reference blower (Seeing how its got 950 base clock) and is it just to loud for you? Just curiosity speaking here...

Maybe you should just try an Gelid or Arctic cooler for around 50 bucks. Might help you out if noise and temps are an issue for you.
 
It's this card, and in particular it's in my HTPC. It's one of the quietest and best coolers. I've currently got all fans (except GPU) set to slowest speed, which makes the GPU run around 80C at 80% speed, and the CPU got up to 90C apparently...

But, that was with terrible airflow around the case, terrible fan speeds, and no A/C turned on at the time. Also a problem is that GPU dumps heat inside the case.

This weekend I'll have time to do some more testing. I'll ramp up the base fan speeds to get more airflow going, check it with A/C turned on, and turn on fan increase for CPU as temp rises. I'll try it with stock 7970 clocks, and if it's still too much I'll have to dial it down... After all, it's running 1920x1080 so it's plenty of GPU power.

Edit: Yes, I realise how utterly stupid it is to use a 7970GHz that dumps heat inside this case when it's shoved into a TV stand. :p
 
Last edited:
It's this card, and in particular it's in my HTPC. It's one of the quietest and best coolers. I've currently got all fans (except GPU) set to slowest speed, which makes the GPU run around 80C at 80% speed, and the CPU got up to 90C apparently...

But, that was with terrible airflow around the case, terrible fan speeds, and no A/C turned on at the time. Also a problem is that GPU dumps heat inside the case.

This weekend I'll have time to do some more testing. I'll ramp up the base fan speeds to get more airflow going, check it with A/C turned on, and turn on fan increase for CPU as temp rises. I'll try it with stock 7970 clocks, and if it's still too much I'll have to dial it down... After all, it's running 1920x1080 so it's plenty of GPU power.

Edit: Yes, I realise how utterly stupid it is to use a 7970GHz that dumps heat inside this case when it's shoved into a TV stand. :p
Oh ok that makes since, I don't know which case you have but maybe if you can force airflow to the back with a small fan addition you could get that extra hot air out of the computer.
 
Under-clocking a bit certainly won't be noticeable with your setup, and might help temps somewhat. Also, if your case fans are getting old, some of the newer designs may have a better airflow-to-noise ratio, helping a little more.
 
Undervolt? Depending on your factory settings it could be much higher than it needs for the clockspeed you are running at.
 
I've read some forums about what exactly has the most impact on high-end GPU temps, and I didn't find anything conclusive. Obviously core clock impacts some, but people also said that memory clock heavily impacts temps.

Using MSI Afterburner I reduced the memory to 1300MHz and the core to 860MHz I believe. Temps/noise is fine at the moment, and I will probably bring it back to stock to check again. It was recommended not to touch voltage or power% for the 7970GHz, although I may try this @hellokitty[hk]

Not sure how my case fans are with airflow/noise ratio, but they are pretty much brand new Noctua's throughout, and they are pushing a decent amount of air IMO.
 
I don't know what card you have but there's a good chance your voltages are already beyond stock, and far beyond what you need for and underclocked card.

Of course if you limit the power the temperatues will follow.
 
I have this card.

I thought reducing power% can lead to instability since the card may be starved of power?
If you reduce the clock speed, it won't need 100% power. Just figure the percentage of clock speed reduced and reduce the power almost the same percentage. This should further reduce the temps since you're not feeding it power it doesn't need.
 
I might be wrong but the way I understand it is that AMD cards now have a dynamic clock speed, which increases whenever possible, but is bounded by a couple factors to remain reasonable.

The first is the absolute turbo clockspeed, it won't go higher than that no matter how much headroom it has.
The second is temperature/fanspeed, it will continue to ramp up clocks and fan speed until it hits the maximum temperature and maximum fanspeed.
The last is power consumption.

So in other words if you bring down the clockspeeds and lower the power, it will ramp up clockspeeds and take more power until it hits that power limit, I'm pretty sure it doesn't push itself to BSOD.

AFAIK that card has default voltages of 1.112v. A lot of factory overclocked cards have something like 1.256v, and I think a stock card by default has 1.05v unless the manufacture changed something. For example I think I remember that an HIS version has 1.05v stock.

If you want the best possible temperatures, pick a clockspeed that gives acceptable performance, cap it at that clockspeed, and lower the voltage until you get artifacting.
Generally it's more efficient to run at lower clockspeeds so you may find that you need exponentially less voltage/power to run underclocked.

I should note that this will be time consuming and possibly frustrating. I spent about a day undervolting all the power states on my phone and went through fixing a cold brick, some boot loops, a few restores, many reinstalls, and lockups/reboots.
Luckily it only takes a few minutes to check for stability otherwise it would have been much worse.
 
Last edited:
I have a Gigabyte 7950 and the card would start throttling itself at about 72C on the core.
When running furmark the VRM's would reach 115 C and that was at stock voltage and speed of 1GHZ(stock).
The card would also slow itself to 900mhz in some games,in Furmark it slowed right down to 500
In Heaven the VRM's would still go over 100C.

I found a program called VBE 7.0.0.7b which allows you to edit the bios's of 79 series cards.
I noticed that the core voltage was 1.25v,which was over kill for this card.
At the moment I'm running at 1,1 GHZ and at a voltage of 1,125 and the VRM's just on 100C in Furmark and reach about 85 on a hot day in Heaven.
The core get's to about 60C now on a hot day,can't wait to see what it reaches during the winter months.

You can also use VBE to get rid of the turbo feature,set a manual fan speed and set the power limit.
 
Back