Folding Forum

Hey all, I'm brand new to folding - something I've been wanting to do for a while, but the current COVID-19 situation brought it to the forefront again so I finally jumped in. I do have some questions on the PPD I'm seeing with this card though.

Paragon put together an excellent GTX 1080ti folding review (thank you!) that really excited me for what this card should be capable of. However, I'm not getting similar results compared to the article, or similar results to anything else I've seen online either. My PPD is actually averaging higher and I'm trying to figure out why. Are others with 1080 ti's getting similar results and not reporting it? Could this be due to recent fahcore optimizations? Or am I just fortunate with my current configuration?

i7-5930k @ 4.3 GHz
32GB DDR4 @ 2800 MHz
Evga 1080 ti Hybrid +103 MHz
Windows 7 - 7601 (I know I know - please don't turn this into an OS thread. There's many reasons I'm not happy with Windows 10 - working up the courage to face the change . In any event, it's certainly not hurting PPD)

PPD varies a bit between projects, as drivers and cores are updated, and with the exact hardware version you have. You would need to go through all those details (and possibly others) between your system and Paragon's to find an explanation.

I run a GTX1080ti in win7 as well. The PPD you are seeing on this is at the high end of what expected on Windows but yes seems to be running well.

Worries me that if I do switch away from Win 7 I might see a negative impact. Guess I won't know till I try. Thanks guys

This is just a guess.

You are getting mostly Core_22 Work Units, as all the of COVID-19 work is new.

Once life calms down and researchers get back to their main research, you will get some Core_21 WUs, these tend to be about 15% less points.

(All new work is on Core_22, but existing project continue to use what ever core they were started on)

Liquid51 wrote:Paragon put together an excellent GTX 1080ti folding review (thank you!) that really excited me for what this card should be capable of. However, I'm not getting similar results compared to the article, or similar results to anything else I've seen online either. My PPD is actually averaging higher and I'm trying to figure out why. Are others with 1080 ti's getting similar results and not reporting it? Could this be due to recent fahcore optimizations? Or am I just fortunate with my current configuration?[/img]

I'm running a 1080 ti (Aorus Waterforce) on win10 and it also generates a higher PPD than Paragon's: around 1.5M point/day.

EDIT:


EDIT2: There are different flavors of 1080 ti. I see that the reference card has a Base Core Clock of 1480 MHz (Boost: 1582), whereas yours is at 2062.5 and mine at 2037.5 according to GPU-Z. That's probably where the gain in PPD comes from.

I find the CPU clock speed has a significant effect on time to complete work units. Every so often the cpu does some sanity checking,as I understand it, on the cpu.

Appreciate everyone's input.

@Jimbo - you're right. Seems that everything is currently core_22 WUs

@ajm - I did add a bit of an OC to the gpu (+103). It was hitting high 1900's before that (whatever the default Nvidia boost 3.0 configuration figured it could do I suppose), but still pulling decent PPD (1.5M+). Probably a hard thing to determine with the varying WU's and other hardware components, but I'm sure the +500MHz you and I are both seeing over stock boost speeds is the main contributor there (I'll add that while the highest WU I've seen was 1.8M PPD - it was only once - and roughly half run at 1.6-1.7M. I have plenty in 1.4-1.5M range as well)

@ipkh - My processor is getting up there in age, but she's happily plugging away at 4.3GHz. I wonder how impactful it would be if it was running at base boost (3.7GHz). Oh well, I definitely am not bothering to find out

ipkh wrote:I find the CPU clock speed has a significant effect on time to complete work units. Every so often the cpu does some sanity checking,as I understand it, on the cpu.

No, it doesn't.
When you check your GPU load (in Afterburner or in taskmanager with compute) then you will see drops in GPU load every few minutes. The drops last only a few seconds and appear every few minutes. The drops are what you are referring to as "sanity checking".
A faster CPU will not have an effect when you consider the whole process (request a WU, getting it assigned, download it, compute it, upload it).
Saving a second or so doesn't make a difference.

My experience is that CPU roughly equals the WU completion time. And that I do get better results with higher cpu clocks. This is sustained 100% cpu usage, not minor short term bursts of activity.

You have 100% cpu usage while GPU folding? Something's not right.
CPU should be idle, basically.

GPU folding (at least on nVidia) requires one core per GPU to supervise the data transfers. That core will be fully loaded, because it uses a "spin wait" between transfers. This is not ideal but is required by the current GPU drivers.

I'm aware of that, however I thought one NV GPU needs one thread.
But that's not what ipkh is saying.

Ipkh talks about 100% CPU usage. This would mean ALL cores/threads.

And spin wait is not influenced/improved by increased CPU clocks.

Some projects will use all the threads available on the CPU to perform sanity checks on checkpoints.

Ok, thank you for clarifying.

But still my understanding is that sanity checks don't require much time (compared to GPU computing) and therefore overclocking of the CPU doesn't improve GPU folding performance.
Please correct me if I'm wrong.