Folding Overclock

So, I'm curious, how many of you are overclocking while folding? I know that an OC while folding is... questionable, given the chance of a WU failure with even a small OC.

So I'm currently running 1 EVGA 1080 FTW Hybrid at a +110 MHz OC on the GPU and 3 1070 SC (2 Black Ed.) with a +74 Mhz OC on the GPU. From what I've seen, OC can be done while folding as long as you don't touch the memory clock speed.

My 3 1070 are putting out 20k PPD more per unit due to the overclock @ 550-750k total. The 1080 seems to be gimped (even before the OC) and only doing ~800k with the OC.

Anyone else have numbers?

For systems that are for dedicated folding, I do not overclock, or if I do it is very minor (+50 mhz or less).

While the cost of folding is free in the cold winter, the cost of cards is not cheap. With a multiple GPU system, there is already enough heat as is, and folding is already shortening the lifespan of the cards. I don't use watercooling for a dedicated folding rig since financially and timewise it just doesn't make sense. Why spend all that money on waterblocks, radiators, etc for a bit of overclocking headroom when you could just get another GPU? How long will it take to set all that up and make sure there's no leaks? One bad connection and it's a lot of money potentially wasted.

Keep in mind we don't get paid to fold, so every $ matters. I'd imagine if you have access to a corporate cluster or datacenter with free electricity you could probably actually make a tidy profit folding on one of teams that pay.

From my experience cards that are run 24/7 are put under lot of stress.

There's a few other users like HayesK who are running lots more GPUs than I am. They could probably give you a better answer than me.

I kinda understand this so I actually set my target tempuratures (currently 79°) lower than stock. I'm new(ish) to the game of overclocking and don't have much knowledge of GPU lifespans. Is running a card at these temps 24/7 going to kill the card quickly?

If you set your temps lower than stock there's a chance you will see some throttling if your airflow is poor or you're overclocking a lot. Check the actual clock speed in GPU-Z or afterburner to figure out.

As for failure time, it's not like there's a guaranteed expiration date or something. When you're maxing out the card 24/7, the mean time before failure is probably going down by a significant amount. But it's impossible to know when. Your card could fail tomorrow, but it could be OK for the next 5 years (by which time it would be obsolete for folding anyway).

Also all cards can vary a lot in terms of PPD based on the WU. The average across several, however, for my GTX 1080 is around 750-800k.

Yeah, I've noticed this with my 1070. Out of the 3 I have, the card that isn't Black Edition consistently runs about 700k, while the other two do ~620k.

I'm actually noticing an issue with my 1080. It appears it won't go above (for more than 10 seconds) the base clock of 1721. GPU-Z is showing both Util and VRel for PerfCap. Power level stable around 37%. CPU use hovers 30-40%.



Are you able to get anything above your base?

I am currently folding with 3 GPUs:

- 2 Titan X Pascal (with an EVGA 1080 hybrid cooler modded onto each)...temps stay below 45 C at 28 C ambient
- 1 EVGA 1080 FTW (air cooled)...temps stay below 65 C at 28 C ambient

Both cards are heavily GPU overclocked...with sustained boosts of 2075 MHz to 2125 MHz while folding.

I am only doing +100 on memory (I would rather the power went to the GPU...but this is just me.)

I have never had a single failed work unit.

Also - electronics are mean to run. As long as you manage the heat and keep their temperature in check, they will run for a very long time.

Also - I agree that the folding performance of the 1080 is not what I expected it to be. I posted a thread in this forum about the folding performance of the Titan X Pascal versus my 1080 FTW when I first got my first one - hehe. It performed WAY better than I thought it would.



Edit: I also responded to your other post in the GPU section. You are using the wrong EVGA tool to overclock a Pascal card...you have to use Precision X OC (or MSI Afterburner). I use Precision X OC myself.

I sometimes buy factory OC card's but I don't push the OC any further than that or what the cards do themselves in terms of boost clocks. Its not worth the additional points, especially if the card or WU crashes at some point and power draw goes up as well.

JrClocker said:

I am currently folding with 3 GPUs:

- 2 Titan X Pascal (with an EVGA 1080 hybrid cooler modded onto each)...temps stay below 45 C at 28 C ambient
- 1 EVGA 1080 FTW (air cooled)...temps stay below 65 C at 28 C ambient

Both cards are heavily GPU overclocked...with sustained boosts of 2075 MHz to 2125 MHz while folding.

I am only doing +100 on memory (I would rather the power went to the GPU...but this is just me.)

I have never had a single failed work unit.

Also - electronics are mean to run. As long as you manage the heat and keep their temperature in check, they will run for a very long time.

Also - I agree that the folding performance of the 1080 is not what I expected it to be. I posted a thread in this forum about the folding performance of the Titan X Pascal versus my 1080 FTW when I first got my first one - hehe. It performed WAY better than I thought it would.



Edit: I also responded to your other post in the GPU section. You are using the wrong EVGA tool to overclock a Pascal card...you have to use Precision X OC (or MSI Afterburner). I use Precision X OC myself.

Click to expand...

Thanks for the reply. Not sure what I did, but my 1080 is now stable at 2088 and taking the OC. I really want a Titan card, but I'll likely wait until Volta and pick up 2 or 3 with a new build. From what I've seen, they're supposed to have exponentially higher performance over Pascal.

I'll trade in my cards an get a new Titan when they come out...these were my first Titans...and they won't be my last!

You should really check out Precision X OC. It works with Maxwell cards, but it's meant for Pascal. Notice your GPU voltage is reading 0 V?

I downloaded and started using X OC as soon as you recommended it. Also, 1080 dropped back down to base clock.

Have you adjusted the power setting for the card?

The voltage regulator? Haven't touched that. Recommendations? 100%?

In addition to GPU and memory overclock sliders, there are also power and voltage sliders.

The power slider allows for the card to draw more than it's rated power. Your card will downclock if it can't get enough power. My recommendation here is to set it to the maximum setting. If you have a FTW card, there is a secondary BIOS on the card that increases this power limit further. To activate this, you have to power down the PC, move the switch on the card (it's labeled master/slave), unplug AC power from the PC (wait about 5 mins), and then turn on the PC. The Precision X OC will then show additional power setting available.

For the voltage slider - leave this alone at the beginning. Additional voltage for the Pascal cards works differently than other NVIDIA GPUs. Don't increase unless your GPU overclock is unstable (the default voltage boost algorithms are pretty good in these cards.). If you get an overclock failure, increase in 25% increments to get it stable.

Maxing the voltage slider without needing it will cause the card to hit the power limit sooner than it needs to.

For my 3 Pascal cards (2xTitan and 1x1080 FTW) I only had to increase the voltage slider to 50%...and maxed out the power slider.

Overclocking my cards brought the GPU boost clock from 1700-ish MHz to 2100-ish MHz...or 25%-ish increase...corresponding to a 25%-ish PPD increase from all 3 cards.

Don't use the folding@home app to test your card stability. I mainly use the Heaven benchmark...letting it run for 20 to 30 mins. This is a pretty hard load on the card, and getting this to jot crash it a good sign. Next, I load up The Witcher 3...the game hits the GPU hard.

If I'm stable for these 2, then I seem to be stable everywhere.

Good luck!

With a set 100% voltage the card is still limited to 1721. GPU-Z is showing the voltage @ .81.

Shouldn't the voltage setting force 100% use?

The PerfCap Reason is Util...which means that the card is not being 100% utilized by anything that's driving it.

Right click on the desktop and select the NVIDIA tool. Go into the card setting and select "Prefer Maximum Performance" or something like that for the Power setting. This will force the card to not go into power savings mode, and always be at its maximum clock.

I have mine set this way, and the cards only downclock to 1418 MHz when idle. It's not a big deal for me as the cards are not idle very often...

I'd start by setting the voltage to 0%. The GPU boost algorithm in the Pascal Cards is different...it boosts the GPU voltage by bins.

Set the power target to 120% (max) and voltage to 0%.

I've always had the NVIDIA option set to prioritize performance.

I've switched the 1080 over to the secondary power setting and F@H started at 2113 MHz clock. It's been stable now for 5+ minutes. I'll keep you updated.

Thanks for all your help so far, JrClocker.

Points for the team man!

My 1080 FTW (non-hybrid) seemed to have a much more stable GPU clock on the secondary bios...even though the card shows it doesn't come anywhere near the maximum power limit. The only thing I can figure is that there are short "burst demands" for power that don't make it through to the "average power draw" graphs that the utilities show.

Once I started overclocking my GPUs, I stopped CPU folding...the ±points daily variation on the GPUs was larger than what I got from CPU folding - hehe.

Yeah, I haven't attempted to fold on the CPU since day 1. I'm still able to use my PC when I'm folding as long as it's only on the GPU.

Today's numbers: