Folding Forum

Hi,

I have a 12x 1080 Ti rig mining various alcoins and thought I'd give folding a shot.

I understand the CPU processing power and and PCIe bandwidth matter (which isn't the case when mining) but I don't know how much.

What would be the PPD of a 1080 Ti on riser (because it's in a rig)?

What would be a good starting point for me, information-wise? I mean a good thorough article or blog post or...

Mod edit: merged two posts

Welcome to the folding support forum.

There is a long topic on this subforum - viewtopic.php?f=38&t=28847 - that discussed this issue as recently as last Fall. Some newer topics in FAH Hardware have also touched on this issue. Basically each GPU will require use of one CPU core of a processor that is not at the very low end of power. If you are running Linux the PCIe bandwidth is less of an issue, but 1X slots will reduce throughput. 4X or better PCIe v.3 slots general are sufficient for almost full throughput from reports by those who have tested this.

Well, that changes the plans...

Maybe I can fold with a 1080 Ti in a 16X slot and mine with the 11 others.

We're doing real computations here, not basic maths ...

Just give it a try and post the results here. I assume everyone's curios

I've seen reports, over on Discord I think it was, that 1x 3.0 pcie slots do ok

I heard you need have Linux and one CPU thread for each GPU and still pcie 3.0 1x looses 20% PPD - may be worth it.

I guess I lose about 20% on a 1080 on a 1x 2.0 on Linux

My airflow is fucked up if I install one 1080 Ti in a full size slot.
So risers are mandatory and my CPU is a piece of crap (2 threads), it hardly seem worth it to go through the hassle of learning the specifics of folding with such limitations from the start.


But I thank you for the information!

I always assumed people running big rigs like this would go with liquid cooling solutions for the GPU's, to remove points of failure from the design and ensure better temperatures on the cards.

I also wouldn't ever run a 1080 next to another. The air gap is just too small. I have had cards overheat in the past, when ran side by side. If I was putting down the type of money required to build that kind of setup, I would consider liquid cooling a must.

Anyways I would also be interested in the OP's numbers running 12 cards simutaneously for 24 hours. Even if only as an experiment. As for running from an x16 slot, not massively bothered. We know what a card can generate in a x16 slot. What would be more useful is a hardware breakdown and numbers for 12x cards folding in whatever configuration he uses. While the CPU will bottleneck it. I am still interested in it from an experimentation standpoint.

v00d00 wrote:I always assumed people running big rigs like this would go with liquid cooling solutions for the GPU's, to remove points of failure from the design and ensure better temperatures on the cards.

The price of a watercooling for 12 cards is crazy. It's a much more efficient solution to run them on an open frame with one or two box fans moving air around.
Very noisy, but a huge watercooling would be as well.

Anyways I would also be interested in the OP's numbers running 12 cards simutaneously for 24 hours. Even if only as an experiment. As for running from an x16 slot, not massively bothered. We know what a card can generate in a x16 slot. What would be more useful is a hardware breakdown and numbers for 12x cards folding in whatever configuration he uses. While the CPU will bottleneck it. I am still interested in it from an experimentation standpoint.

I had really shitty results, under 1000000 ppd with the 12 cards: I assume the CPU couldn't handle the workload at all.

If it were a 6 core/12 thread cpu which are now cheaper available from AMD or Intel then I guess it would make 12 x 1mill PPD max on Linux = 12mill PPD total.

The celeron is supposed to have two cores.

Anyway, not trying that again

Amstellodamois wrote:The celeron is supposed to have two cores.

Anyway, not trying that again

That's most definitely under-powered. One thread per GPU is recommended when running OpenCL drivers.