Overclocking a GTX1070

[royaleagle]

Hey guys,

It has been a while since I didnt use FAH. I was wondering if you could help me on the subject of overclocking.

Is memory or core clock more important ?
Should I lower my power limit, what is the best ratio between power limit(consumption) and PPD ?

Thanks a lot

Happy folding !

[JimboPalmer]

I know nothing about this topic, but core clock will be more important than memory clock. I wish your GPU a long life.

[Joe_H]

Unless you have a Founders Ed. 1070, they mostly come factory overclocked already. You might get a slight increase on the core clock, but it might be at the expense of not being stable for folding calculations.

[SteveWillis]

I went through an over clocking phase (Linux). Once I stopped doing it my folding problems were reduced by 95%.

[bruce]

You might get a slight increase on the core clock, but it might be at the expense of not being stable for folding calculations.

I second that thought. If you boost the core clock beyond what it already is, you'll find that some projects will push the GPU into unstable territory. Even if it's only an occasional WU, the loss of one WU every N (for some N) will counteract any benefit you might get from the overclock.

[HayesK]

Have a pair of EVGA 1070 FTW cards folding in linux, caseless, with settings +25 GPU, -200 memory, 80% fan and power limit 130.

[Nathan_P]

I would just let the gpu boost itself,as long as you get plenty of cooling to the card it will boost to somewhere around 1900-2000mhz. Manually overclocking to gain the extra few % over that may well shorten the life of the card and lead to lots of stability issues. My 1070 is factory overclocked and i let the boost sort itself out, its currently clocked at 1987mhz core and 7604 memory on linux

[rwh202]

I would just let the gpu boost itself,as long as you get plenty of cooling to the card it will boost to somewhere around 1900-2000mhz. Manually overclocking to gain the extra few % over that may well shorten the life of the card and lead to lots of stability issues. My 1070 is factory overclocked and i let the boost sort itself out, its currently clocked at 1987mhz core and 7604 memory on linux

This!
The recent generation of cards ship with pretty optimal voltage / frequency curves and boost as high as power and thermals allow.
If you want increased efficiency, then just bringing the power target down pulls you into the more efficient part of the curve. I find ~130 W optimal for my 1070 Ti and 1080s, 170 W for 1080 Tis. Maybe try 120 for a 1070. It might limit boost by 70 MHz, but saves 25% power.

[royaleagle]

Have a pair of EVGA 1070 FTW cards folding in linux, caseless, with settings +25 GPU, -200 memory, 80% fan and power limit 130.

This!
The recent generation of cards ship with pretty optimal voltage / frequency curves and boost as high as power and thermals allow.
If you want increased efficiency, then just bringing the power target down pulls you into the more efficient part of the curve. I find ~130 W optimal for my 1070 Ti and 1080s, 170 W for 1080 Tis. Maybe try 120 for a 1070. It might limit boost by 70 MHz, but saves 25% power.

That's the answer I was looking for, optimizing and underclocking thanks, I'll try your settings HayesK

[toTOW]

The only interesting change is to set power limit to the maximum allowed (and maybe temperature limit too if you have efficient cooling) and let the card do the job ... the boost system is a real pain for overclockers

[ProDigit]

1030 and 1050 user here (and soon an additional 1060).
I've read online a few negative comments about underclocking from factory, and how it affects folding.

Not only do you get lower performance linear with the underclocking (eg: underclock 10%, get 10% less work done), which in itself is not that bad,
But you also lose bonus points for delivering work later.
Users underclocking their cards need to take this in consideration. The PPD drop seems rather sharp.
If you just are concerned about finishing WUs, at the most cost effective solution (after all FAH is not paying your electric bill),
You'd have to keep in mind that slower WU runs, will mean your PC will have to run longer per unit.
If you're running your PC 24/7 you might end up with a slightly lower electric bill (usually underclocking graphics card will result in up to $5 per month per graphics card, depending on the Graphics card); but you might as well turn off CPU folding, and end up with higher power savings; or plug your graphics card in a PCIE 1x slot for faster cards.

We really need some scientific benching done, set as an example for other users here.
I think it would be interesting to see some research, backed by hard numbers, to see the effects of underclocking on those pascal cards (GT1030 - GTX 1080 and RTX cards), and how underclocking relates to power consumption and PPDs.


Same can be said about overclocking,
Usually overclocking will cause more power draw, at a comparative small performance gain.
From the few articles I read online, these pascal cards at under 10%, would net a <10% performance gain.
Especially the lower end models, as they are easy to cool.
Unlike previous models at 28nm, where a 6% overclock would net a 8-9% increase in power consumption, or even older graphics cards, although 28nm cards have been in circulation for quite some time now.

These numbers I looked into, are for gaming and 3D benchmarking, not folding though; but it's the closest I can compare it to.

[Paragon]

Hi,

I recently wrote an article on my experiences getting the most PPD out of an air-cooled EVGA 1070. In summary, it came down to using some older drivers that work better for folding (372.90), plus manually setting the core clock to a reasonable number that the card would boost to during gaming (about 2000 MHz). The reason I needed to do this was because, for some reason, my card was not boosting by itself, but was instead in some sort of low power mode (P2 state). If anyone in these forums can shed some light on that, I'd be interested to hear it.

Details here:
https://greenfoldingathome.com/2019/01/ ... -gtx-1070/

I also have done some quick power limit tuning on a GTX 1060. Basically yes, you can gain efficiency by lowering the power limit, but you lose points. I didn't have time to exhaustively test this, but the two settings I used (50% and 100% power limit) are backed up by a few days of stats. Full article here:

https://greenfoldingathome.com/2019/02/ ... wer-limit/

[bruce]

I also have done some quick power limit tuning on a GTX 1060. Basically yes, you can gain efficiency by lowering the power limit, but you lose points. I didn't have time to exhaustively test this, but the two settings I used (50% and 100% power limit) are backed up by a few days of stats. Full article here:

https://greenfoldingathome.com/2019/02/ ... wer-limit/

FAH's bonus points system boosts the total points based on faster completion. Power utilization goes up based on effective clock rate and so does PPD -- but not limearly. Your clockrate is an independent variable and you're trying to compare those two curves which don't match precisely. Thus there is a PPD vs. power setting with an optimum which is somewhat less that the highest stable power setting but not necessarily the stock clock rate either. Finding thhe optimum is probably a lot of work with only slight net differences.