Folding Forum

foldinghomealone wrote:I updated the database.

Any comments on it?

Looks great! Thanks for putting this together. I'd love to see more people submit data .
Maybe throw a standard deviation score in there eventually.

Looks about right.

It's always hard for these sorts of things. I think there's a good case for ignoring the top 1% and the bottom 1% at the very least because sometimes machines do get rebooted during units and there are also very weird units sometimes that generate massive points.

My TitanX is a case in point. That high/low is such a wide variation.

I added stdev in the PPD summary.
However I think that spikes are not as important as long the sample size is high enough.

Prophet_of_83 wrote:I'm curious to know what drives top performance? Is it # of CUDA / Stream cores, or clock speed?

Looking at the google doc sheets, it looks like the nVidia xx80 series (eg 2080, 1080, 980) generally outperform the xx70 (eg 2070, 1070, 970) and likewise the xx60 series (2060, 1060) in that order. What is it about the xx80 series that gives it the edge? More RAM, more CUDA cores, higher clock speed?

edit: when I say the xx80 serious outperforms the xx70 series, i mean a 980 can outperform a 1070, and a 970 can outperform a 1060, so the generation isn't a good indication of rank, but the series is a better indicator of rank.

Like an automobile manufacturer, they have a top tier product, and a bottom tier, with some number of steps between.

x80 is top tier, with most cores, highest clock, most shaders, best cooling, glitzy lighting, and whatever else might appeal.
x70 is almost as good
x60 is kinda "entry level" gaming, it has a lot less of everything
x50 is the no-frills version, often available in low-profile for servers, and in smaller form factors for micro- and mini- computers, AND, the x50 version needs to additional power plugs. It runs off the 75 watts available in your PCIc slot.

Given the tiers above, each generation just gets more powerful. I have a 1650 parked side-by-side with a 780, and the 1650 earns more PPD, because it's newer and more powerful. I am often told that the 780 should outperform the 1650, but that's not what I'm seeing.

Hope that helps some.

I reuploaded my HFM export with frame bonus. Thanks for doing this foldinghomealone, it's looking really good!

I updated the database.
Thanks for the data.

@all:
don't be shy to upload current HFM export

PS.
I will make a new topic on the Hardware page
viewtopic.php?f=38&t=34510

Kebast wrote:

NoMoreQuarantine wrote:In an effort to make a predictive ranking of GPU performance, I took the specs of all of the desktop GPUs since 2016 for both AMD and NVIDIA ...

I think you should at least go back one more generation. Several of the Nvidia 900 series cards would fall toward the middle of your list, and can be found cheap on ebay.

I would say, modern AMD GPUs are about as efficient as GTX9xx to GTX10-series, but much faster.
Most of your PPD is coming from QRB, which means you'll really want to invest in a faster GPU, rather than a previous gen GPU.
For instance,
A GXT1080Ti nets a lower PPD count as an RTX 2060, but folds at ~225-250W vs 170W for a 2060.
Once you cap power on the GPUs, the RTX 2060 runs at even better efficiencies.
So you'll be running at least $50 a year on electricity extra on the 1080Ti. I you want to keep the card for 2 years, that's $100 added to your purchase price.
For that reason, I think lower than 1660 isn't really worth researching, or buying at a near to new price.
GTX 10th gen, are great GPUs for gaming, but no continuous load.

Kebast wrote:...
I uploaded a file.

Kebast, I would have a question regarding your CPU exports.
Do you fold with all cores/threads or do you limit it a certain amount?

Juggy wrote:Thanks foldy

Gigabyte RTX 2060 Super Gaming OC Windforce
PCIe 3.0 @ 16x
Windows 10 1909

Device - Geforce RTX 2060 Super
Compute - OpenCL
Precision - Mixed
WU - Real
Accuracy check enabled
NaN Check - Disabled
Run length - 1 minute

Score - 86.0815
Scaled Score - 259.764
Atoms - 64614

My 2080 Super arrived this morning

Gigabyte RTX 2080 Super Windforce OC
PCIe 3.0 @ 16x
Windows 10 1909

Device - Geforce RTX 2080 Super
Compute - OpenCL
Precision - Mixed
WU - Real
Accuracy check enabled
NaN Check - Disabled
Run length - 1 minute

Score - 116.178
Scaled Score - 350.584
Atoms - 64614

@Juggy, but how many ppd does it get?

HaloJones wrote:@Juggy, but how many ppd does it get?

Struggling to get work at the moment so it has only done 1 WU and that estimated around 2 million PPD.

HaloJones wrote:My guess would be WuHistory.db3 that is the file that all the results get written to and which you query when you go to "Tools, Work Unit History Viewer"

Maybe I am being a bit dense but where do I find the Tools menu?

Never mind, I see I need HFM setup to gather data.

foldinghomealone wrote:

Kebast wrote:...
I uploaded a file.

Kebast, I would have a question regarding your CPU exports.
Do you fold with all cores/threads or do you limit it a certain amount?

No, I only give cpu slot 12 threads. This is my main desktop, so even then those 12 are sometimes doing other things.

I created a post to help people get HFM.NET setup viewtopic.php?f=16&t=34516

NoMoreQuarantine wrote:I created a post to help people get HFM.NET setup viewtopic.php?f=16&t=34516

Thank you for this, very handy. I found an article from 2009 with a very old client that I couldn't figure out.