Lower PPD with several projects

[Ichbin3]

Hi there,
I run a RTX 2060 Super with undervolting at normaly 170W the whole system.
I get normly 1.4 Mill PPD in the good projects.
But there are also some with significant decresed PPD and also less usage of power without me changing anything.

Code: Select all

Projects	PPD	Watt
16905	844000	140W
13302	1051713	150W
11760	1095624	150W
11759	1174800	150W
13408	1183000	150W
11764	1194000	150W
13408	1090347	150W
16441	1335313	
16442	1366975	
14457	1368112	
14465	1374485	170W
14456	1402600	170W
14457	1450442	170W
14450	1470000	170W
14448	1481000	170W

I'm not at beta.
Does anyone know about that and may have some hints?
Thanks!

[foldy]

Project 16905 has low atom count which cannot use all shaders of RTX 2060. That is why it only got 844k PPD.
But other projects like 11759 have high atom count so that is not the problem.

RTX 2060 will get 1.2 Mill PPD average so the other projects ranging from 1.0 Mill PPD to 1.4 Mill PPD look good.

Be sure you have one CPU core exclusive to feed the GPU. Nothing else to do. Happy folding!

[Ichbin3]

Thanks for explanation.
Of course the CPU core which feeds the GPU is running separatly.

How comes that some projects use the full power (170W system = 100W GPU) and some with lower PPD only 80% of GPU-power and Afterburner still shows 96% of GPU-usage?
And even if I change Afterburner into the reset mode (full power, GPU runs with 180W,System with 250W) it does not change so much in PPD, only goes up to 1.2 Mill

I'm confused ...

[PantherX]

Welcome to the F@H Forum Ichbin3,

Project 16905 has 28,000 atoms
Project 11760 has 62,227 atoms
Project 14448 has 216,769 atoms

You can see that there's a pattern where the usage of Wattage is increasing as the atom count increases.

A project with low atom count can use less shaders than a project with high atom count. Thus, even though your GPU is being used 100% the amount of power used will be dictated by the amount of shaders being used.

[MeeLee]

Which is what makes successfully overclocking pretty difficult on these GPUs.
A setting that might work for a slower WU (lower power draw), will definitely slow down the larger atom count WUs.
A setting that's great for a higher atom count, will only use more power, and potentially running the GPU frequency too high on low atom count WUs.

[foldy]

I would recommend to stay with power optimized settings. Highest boost clocks take much more power usage for only small speed benefit. 1.2 Million PPD average seems good enough for me.

[markdotgooley]

I have an RTX 2060 and an RTX 2060 KO on the same motherboard and depending on projects I can get PPD displayed as anything from 1.4 million to nearly 2.2 million. Lots of sensitivity to those projects...

[Ichbin3]

I'm with a RTX 2080S now and there is not always this dependency - lower atom count means lower PPD.

Anyway and btw ...
here is my recommendation for this 2080S
I have it running with 125W, this seems to be the best balance between power and PPD and also the most economical setting.
nvidia-smi -i 0 -pl 125
I'm earning nearly 2Mill PPD.
All increased power higher than this brings only lower increasing of PPD.
Means, if I increase power for 3.3% it brings only 1.3% higher PPD.
Increasing power from 165W to 230W (est 39%) brings only 8% more PPD.

[MeeLee]

125W for a 2080 seems a bit low. Your GPU would be running at 1500Mhz or so.
Remember that on top of the '125W', you also have to take account for the TDP of the CPU, and PSU efficiencies.
So a 125W may translate to 200-225W at the wall.
Then 1500Mhz may not seem very interesting.
I found the optimal value for the 2080 around 145W, and the 2080 Super, should be around 150-155W.
It'll also depend on temperatures.

Another thing to remember is, that when you run those high costing RTX GPUs, your PPD will be lower, your WUs will be lower.
And in a year or two, it would have been much better to have folded at a slightly higher PPD count and wattage, because the 2080 won't be fast enough anymore in 2 to 4 years.
It'll be like a 1080 or 1080Ti. It's still ok, but nothing compared to a 2080Ti.

[Ichbin3]

'Should be' is a huge difference to 'it is'
Look by yourself.
It is running at 1680 MHz.
This are my measured data:

Code: Select all

Proj 14467											
GPU W		Syst W		TPF			PPD		power incr		PPD incr.
105			150		00:02:37		1342760		100,00%		100,00%
110			155		00:02:26		1624006		103,33%		120,95%
115			160		00:02:15		1678281		106,67%		124,99%
120			165		00:02:06		1844994		110,00%		137,40%
125			170		00:02:00		1981595		113,33%		147,58%
130			175		00:01:57		2042271		116,67%		152,10%
135			180		00:01:56		2064308		120,00%		153,74%
140			185		00:01:54		2098526		123,33%		156,28%
165			210		00:01:50		2177989		140,00%		162,20%
200			245		00:01:47		2243710		163,33%		167,10%
230			275		00:01:45		2289595		183,33%		170,51%

Code: Select all

Power Readings
        Power Management            : Supported
        Power Draw                  : 122.34 W
        Power Limit                 : 125.00 W
        Default Power Limit         : 250.00 W
        Enforced Power Limit        : 125.00 W
        Min Power Limit             : 105.00 W
        Max Power Limit             : 300.00 W

Code: Select all

Clocks
        Graphics                    : 1680 MHz
        SM                          : 1680 MHz
        Memory                      : 7500 MHz
        Video                       : 1560 MHz

Code: Select all

Temperature
        GPU Current Temp            : 61 C
        GPU Shutdown Temp           : 100 C
        GPU Slowdown Temp           : 97 C
        GPU Max Operating Temp      : 89 C

[uyaem]

How long did you all measure to get to the PPD?
I feel like for something reliable you'd need a week per setting to be on the safe side - for me the variance in projects on Core22 is +/- 15% depending on the project.

[Ichbin3]

For sure it depends on the projects.
I used only ONE project with a bigger TPF (14467) and let it run with the different settings for at least 5 min.
Every setting could run so at least two or more frames.
For every frame I took the best shown vallue in this row which appears at least for two cycles and stayed stable.
I changed from 105 to 140 to 110 to 165 and so on to get a better view to the changings.
I did this twice with this project to get better reliable data.
My system is Linux mint, so there are not so much interferences like in Windows.

I did this with 3 other projects too and the result was nearly the same - 125W seems to be the most economicaly setting and the best balance between power usage and PPD-output.
We are paying 30 Eurocent here in Germany for power, for me it is most important to have a look at this.

[PantherX]

...I used only ONE project with a bigger TPF (14467) and let it run with the different settings for at least 5 min.
Every setting could run so at least two or more frames.
For every frame I took the best shown vallue in this row which appears at least for two cycles and stayed stable...

Please note that the time is meaningless... it is the Time Per Frame (Time taken to complete 1%) which is very important for this comparison. You mentioned that you waited for 2% so I am assuming that you calculated the actual time from the log file instead of using FAHControl, right? FAHControl is accurate after 3% to 5% of the WU has folded since is started. If you paused and unpaused, then you need to wait another 3% to 5% hence, the suggestion to use the actual log file to ensure that you're getting the accurate results.

[Ichbin3]

I am assuming that you calculated the actual time from the log file instead of using FAHControl, right?

No. I used the shown TPF in FAHControl.

If you paused and unpaused,

I did not. I simply changed the setting of the GPU.

Please note that the time is meaningless... it is the Time Per Frame (Time taken to complete 1%) which is very important for this comparison.

This I do know.
You can see, that it is called TPF in my table, no? Time Per Frame ...

you need to wait another 3% to 5% hence,

I did.

Every setting could run so at least two or more frames

Guys, I don't want and I don't need to defend my position, If you don't like my results, just ignore them or do your own measurements.
After that you are welcome to come back and than let's (argue) discuss

[PantherX]

...I don't want and I don't need to defend my position, If you don't like my results, just ignore them or do your own measurements.
After that you are welcome to come back and than lets argue

Apologies if my post came cross as terse or "argumentative", that wasn't my intention at all I just wanted to ensure that you were aware of some of the limitations of FAHControl so you can work around them. Given your post above, it seems that you did but that wasn't a clear-cut case in the original post, hence my comments.

I personally don't like to argue, instead having a healthy and logical discussion supported by facts and data is more my thing