GPU memory OC and longevity- gaming/part time mining

Napoleon

[H]ard|Gawd
Joined
Jan 27, 2003
Messages
1,073
[H]ello -

I haven't OC'd a GPU in the last 10 years and am considering overclocking the memory to maximize hash rate in my gaming rig. It's a single GPU, 1080FW, currently stock but I'll be installing a heatkiller full coverage waterblock in the next month (it's sitting next to the computer).

I didn't used to care about how long the card lasted but I'd like to stretch this one for several years; do the same considerations about CPU OC apply here? Namely, keep temps and collages good and it should be fine. It should be noted I have an X58 hex core OC' D over 4ghz so clearly I'm not afraid of all risk (although it isn't expected to to last as many years)

I'm planning on lowering the power target, declocking the core and increasing the memory for hash rate. For gaming I'd do stock clocks and power target (I play old games)
 
temps and voltage are what will really determine how long your cards will last, not clock speeds.

That said mining can push your card to the limit for extended periods of time so if you're worried about longevity I'd recommend mining with reduced TDP.

Whattomine recommends +150mhz core clock, +500mhz mem clock, and 65% tdp on a 1080, if you've got a good chip you might be able to push it further. I wouldn't be afraid of running it 24/7 at even 100% tdp if you've got it under water with the VRM's cooled. If your water block does not do a good job at keeping the VRM's cool I'd avoid high power draw algo's.
 
temps and voltage are what will really determine how long your cards will last, not clock speeds.

That said mining can push your card to the limit for extended periods of time so if you're worried about longevity I'd recommend mining with reduced TDP.

Whattomine recommends +150mhz core clock, +500mhz mem clock, and 65% tdp on a 1080, if you've got a good chip you might be able to push it further. I wouldn't be afraid of running it 24/7 at even 100% tdp if you've got it under water with the VRM's cooled. If your water block does not do a good job at keeping the VRM's cool I'd avoid high power draw algo's.

Apparently the Heatkiller IV block is pretty good at cooling the VRM's so I should be hopefully be good.

I imagine the TDP is set as low as possible as long as stable, correct? Interesting, if I can run it OC'd stable at a lower TDP, why wouldn't I always do that even for gaming? Perhaps the increase to the power target is only for achieving max overclocks...

If I can get above stock speed with less power, seems like a win to me :). I imagine for mining I'd reduce the CPU OC to keep overall system power consumption down, right?

Thanks for the info, you've proven you know a bit about this so if it's ok with you, I might PM you at some point with a few questions sometime in the future
 
Feel free to PM me with any questions.

Nvidia cards kboost based off their power targets (auto OC) so if you set your TDP lower with an overclock you may or may not see an improvement in performance. An easy way to see this in action is to fire up an intense GPU load like mining or furmark then lower your power target in MSI Afterburner (or your preferred GPU overclocking software). You will see the core clock reduce as you lower your power target because it can't increase its voltage to keep the card in its auto-oc stable range. As you manually overclock though you will be increasing the default offsets so even at a lower power target you will have higher clocks. This is where the good old fashioned silicon lottery will come into play. You may also find that you can reduce your power target by a significant amount without the clocks auto reducing. On my 1060 mining rig I can reduce power target to 85% before the core clock slows down.

I can't speak to if this is load specific though. I set specific over/under clocks for specific mining algo's and that is it. I'd imagine in games and under different workloads I may find that at 85% the cards aren't boosting as aggressively as they would be at 100%.
 
Whattomine recommends +150mhz core clock, +500mhz mem clock, and 65% tdp on a 1080, if you've got a good chip you might be able to push it further. I wouldn't be afraid of running it 24/7 at even 100% tdp if you've got it under water with the VRM's cooled. If your water block does not do a good job at keeping the VRM's cool I'd avoid high power draw algo's.

What don't I get about GPU overclocking?

Upping the GPU and memory clock, but lowering the TDP? How does that not violate the laws of physics? :) Serious question, I don't get it. Could you educate me?
 
What don't I get about GPU overclocking?

Upping the GPU and memory clock, but lowering the TDP? How does that not violate the laws of physics? :) Serious question, I don't get it. Could you educate me?
It must be a relative thing, like it’s base is 150 higher buy the boost fluctuates with TDP. Like 100 percent Power boosts 1850 or 2000 with +150 OC, buy at 85 percent Power boosts 1700 but 1850 with +150 OC. I’ll have to play with it
 
What don't I get about GPU overclocking?

Upping the GPU and memory clock, but lowering the TDP? How does that not violate the laws of physics? :) Serious question, I don't get it. Could you educate me?

It must be a relative thing, like it’s base is 150 higher buy the boost fluctuates with TDP. Like 100 percent Power boosts 1850 or 2000 with +150 OC, buy at 85 percent Power boosts 1700 but 1850 with +150 OC. I’ll have to play with it

Napoleon is right, there is a relative boost factor to this, there is also some silicon lottery in play also.

Say you bought a new intel 8700k and overclocked it 500MHZ without increasing voltages you aren't going to increase power usage, if you managed to lower the voltage at stock clocks and overclock it a little it wouldb't be breaking physics, but you are a slave to how good your silicon is just like if you were trying to reach 5ghz with over-volting.

kboost increases your voltage and clocks automatically and dynamically on your gpu. When you bring down your tdp you aren't going to be able to boost as high, but most nvidia cards have pretty decent headroom for overclocking so you can still push it a bit to make up for this. Its not unlike lowering voltages at stock clocks on a CPU but with more fuckery and some actual overclocking on memory.
 
The speed of the card is not related to how much power you put into it. Its how much speed you can get ouf of it, at lower power settings.
 
Lowering the power limit in afterburner is just a way to force lower voltage since afterburner doesn't let you lower voltage.

I'll use my card as an example.
1328mhz boost @ 1.2v. TDP is usually around 84% at load (the TDP varies by card, you can monitor it in afterburner or gpuz) . In this case if you set the power limit to anything higher than 85% it doesn't change anything.

Set the power limit to 82% and it drops the boost table down to 1316mhz @ 1.187v. Overclock the core to +12 and it runs 1328mhz boost @1.187v, stock clocks but lower voltage and less power usage. Drop the power limit more and it will lower the clock and voltage again.
 
Thanks guys. Great explanation.

So I assume the risks of this (overclocking but not, using undervolting)... Is BSOD's ?

I mean otherwise the vendors would under-volt at normal clock speeds, right?
 
Thanks guys. Great explanation.

So I assume the risks of this (overclocking but not, using undervolting)... Is BSOD's ?

I mean otherwise the vendors would under-volt at normal clock speeds, right?

You need to test for stability just like if you were overclocking.
 
Back
Top