Lower PPD with several projects

[MeeLee]

my throttled system (excluding monitors) consumes around 260W at the plug.

unbelievable ...
Mine has 65W with Ryzen 2600 and 2080 Super in throttled mode.

It's not unheard of to see those 250+W numbers on a 3900x or 3950x with PBO enabled.
The 105W TDP is just for the CPU, not the memory controller, Infinity fabric, or USB hubs (essentially north/South bridge-part built into the cpu).
Then there's the motherboard trying to compete with other board manufacturers, by increasing power delivery to the CPU, to keep it running past it's boost speed.
Just CPU stress tests can lead to 260W at the wall, 200W with fine tuning and 2133MHZ ram (you can lower voltage on the CPU and SOC by a lot.

That being said, if you are just GPU folding from a higher end Ryzen 3000 series CPU, you may have purchased the wrong CPU.
Not only doesn't CPU speed above 3,5Ghz matter at all for GPU folding, the insane core count is just a waste if you're only feeding 2 or 3 GPUs.

[uyaem]

Agreed, I locked my CPU to 3800 and could take it down by 0.3V (to 1.05V) with no issues ever.
PPD vary, depending on project, between 210k and 350k.

@Ichbin3
I find 65W at the socket for Ryzen 2600 and 2080 super impossible... you almost have a perpetuum mobile there... I think there's a digit missing (should probably be 265W? for your system?)

[Ichbin3]

Maybe I got the word 'throttled' translated wrong and we are not at the same page?
My 65W means - my system has nothing to do. CPU --> 0%, GPU --> 0%.

[uyaem]

Maybe I got the word 'throttled' translated wrong and we are not at the same page?
My 65W means - my system has nothing to do. CPU --> 0%, GPU --> 0%.

Ah okay that makes sense, this value is ~100W for mine.
By throttled I mean "reduced power", as in limiting both CPU and GPU to get more "science per Watt"

[lafrad]

Looks like the PPD across the board is far lower recently: https://folding.extremeoverclocking.com ... ary.php?s=

My CPU's seem to be the same, but my GPU's are 1/2 the PPD as they used to be. Initially I thought it was my system, but the workload and software must have shifted on the GPU side...

[Ichbin3]

I can't confirm that from my side.
My client-type is beta and there has nothing changed in the daily PPD of my GPU.

[Neil-B]

There are a number of important projects under test that are vary small and don't fully utilise fast GPUs and therefore this is impacting the high end, high PPD, high WU throughput GPUs ... This impact could show both on individual installations and overall stats.

[Ichbin3]

I can't confirm that from my side.
My client-type is beta and there has nothing changed in the daily PPD of my GPU.

I have to revise that, PPD have gone worse even on beta.
Projects up from 226545 atoms like the 16600 give a decent PPD, but it seems there are not so many of them anymore.
Will that change in the forseeable future?
Actually, I'm thinking of giving up on my 2080TI, because it is running so low PPD that it makes no sense and is a huge wasting of power.
The 2080 SUPER is giving nearly the same profitability but is much cheaper.

[JohnChodera]

> There are a number of important projects under test that are vary small and don't fully utilise fast GPUs and therefore this is impacting the high end, high PPD, high WU throughput GPUs ...

Just a quick note here: On the 134xx project series to support the COVID Moonshot open drug discovery effort, we're still refining the GPUSpecies restrictions to make sure that WUs with small systems don't end up on fast GPUs they can't take advantage of. 13414-5 is the latest series here, but we're still seeing some RUN-to-RUN variation we're attempting to track down and mitigate. You should see more refinements as we make further progress here.

Thanks so much for bearing with us!

~ John Chodera // MSKCC

[MeeLee]

I can't confirm that from my side.
My client-type is beta and there has nothing changed in the daily PPD of my GPU.

I have to revise that, PPD have gone worse even on beta.
Projects up from 226545 atoms like the 16600 give a decent PPD, but it seems there are not so many of them anymore.
Will that change in the forseeable future?
Actually, I'm thinking of giving up on my 2080TI, because it is running so low PPD that it makes no sense and is a huge wasting of power.
The 2080 SUPER is giving nearly the same profitability but is much cheaper.

If you're running Windows, use MSI Afterburner, EVGA XOC, or Zotac FireStorm, to adjust the power curve on your GPU.
At 160-180W the 2080Ti still performs significantly fast on fast WUs, but saves a lot of power on the lower atom count WUs.
I wouldn't take that GPU much below 160W, where it'll become slower than a 2080 at similar wattage.
All the other GPUs, 2060 to 2080 Super, you can lower all the way to 125Watts.
While they do suffer a performance penalty, still a 2080 Super is going to be faster than a 2060 at 125W; and even faster than a 2060 at 160W.
The 2080Ti might on low atom count, be able to be dropped down to like 140 watt minimum, before serious throttling issues occur, but at those wattages, serious throttling will occur on high atom count as well.

[bruce]

I can't confirm that from my side.
My client-type is beta and there has nothing changed in the daily PPD of my GPU.
I have to revise that, PPD have gone worse even on beta.

For years and years, there has been a persistent rumor which is untrue that using the beta flag gave you higher PPD because the projects run faster.

Every project goes through Beta, then Advanced, and finally full FAH. The speed does not change.

[Ichbin3]

The 2080Ti might on low atom count, be able to be dropped down to like 140 watt minimum, before serious throttling issues occur, but at those wattages, serious throttling will occur on high atom count as well.

Sure it does.
Still - from my research - @140W my 2080 Ti has the best effiecency - means the best PPWatt.
140W is the peak, before and after the PPW drops, at lower and higher wattage.

[MeeLee]

I can't confirm that from my side.
My client-type is beta and there has nothing changed in the daily PPD of my GPU.
I have to revise that, PPD have gone worse even on beta.

For years and years, there has been a persistent rumor which is untrue that using the beta flag gave you higher PPD because the projects run faster.

Every project goes through Beta, then Advanced, and finally full FAH. The speed does not change.

Probably was true, when core 21 was the standard core, and core 22 was only in Beta.

The 2080Ti might on low atom count, be able to be dropped down to like 140 watt minimum, before serious throttling issues occur, but at those wattages, serious throttling will occur on high atom count as well.

Sure it does.
Still - from my research - @140W my 2080 Ti has the best effiecency - means the best PPWatt.
140W is the peak, before and after the PPW drops, at lower and higher wattage.

You and I are on opposite teams here.
Check your GPU frequencies when you're folding, and see how much they drop at what wattage.
And take into account that you could have been running a 2080 or 2080Super at 140W, where a 2080Ti is only marginally faster, but almost twice the price.
All my research indicates, that running a 2080 Ti under 160W isn't recommended.
I get 4 to 4,4M PPD out of a 2080 Ti at a mere 200-220W.
At 140W, you should at least get 3.1M PPD out of a 2080Ti if you compared the results.
I highly doubt you could actually get 3M PPD at 140W. You'll need at least 20W higher.

[Ichbin3]

You and I are on opposite teams here.

Not necessarily.
Maybe we are talking about different topics.
It seems you want to maximise your output and cost of power is not so important, maybe because it is cheap where you live.
May way is about cost-efficiency - means the highest output (PPD) relating to the cheapest cost (power + depreciation of the GPU)

First this hits the 2080Ti, no doubt.

Cost-efficiency means for me to hit the peak of the PPDPW-curve, also because power here is nearly the most expensive in Europe as I wrote before.
This is at 140W. Of course I could fold at 200W and would get a decent more PPD but I decided not to because of what I wrote.
In addition, at higher power the fan has to work more and the GPU gets noisy, more than I want to have it.
At 140W I can set it manually to 1,400 rpm and the temp stays at 64°C.

Check your GPU frequencies when you're folding, and see how much they drop at what wattage.

Yea, I can see that and it matches the output of PPD.

Code: Select all

2080Ti, project 13414, 89k atoms
			
GPU	average PPD	Freq	PPDPWatt
130	2672000	    1425	20554
135	2842000	    1455	21052
140	2928000	    1470	20914
145	2980000	    1500	20552
150	3035000	    1530	20233
160	3100000	    1590	19375
180	3274000	    1635	18189
200	3430000	    1695	17150
220	3577000	    1740	16259

where a 2080Ti is only marginally faster, but almost twice the price.

Not here.
I paid €679 for a new 2080SUPER which I returned.
The 2080Ti was about € 905 unused and with warranty from a shop at ebay.

And take into account that you could have been running a 2080 or 2080Super at 140W, where a 2080Ti is only marginally faster

Also not my findings.
I don't have data from a 13414@2080SUPER, but the 14467 has more atoms and should even perform better.

Code: Select all

2080SUPER, project 14467, 192k atoms	
	
GPU W	PPD	PPDPWatt
105	1342760	12788
110	1624006	14764
115	1678281	14594
120	1844994	15375
125	1981595	15853
130	2042271	15710
135	2064308	15291
140	2098526	14989
165	2177989	13200
200	2243710	11219
230	2289595	9955

I get 4 to 4,4M PPD out of a 2080 Ti at a mere 200-220W.

Than you are a lucky one.
I havn't seen this kind of projects for a while.
The largest and fastest performing one for me now is the 16600 and @220W I could get merely 4 Mill.