Folding Community Board

[Kilrah]

My graph pretty much represents the weather at my place 

 

[marknd59]

I seem to be doing a lot better than most people for some reason and I have no idea why.

 

 

[explosivetnt]

I haven't seen those huge dips that others have. And the ones there were, have been caused by me.

[Cabanaj]

Just out of curioisity, how many have the advanced tag in their advanced settings? I removed it out of instinct (and was curious since I was almost at 100M, decided to try it).

Since then I have got less very low yield WU, and averaging what is "normal" for my card.
If my theory about project 169XX is right, after GOTSpectrum's  explantion, that would really explained the drop in "WU" production. Theses units take a long time, have low yield (in points), the perfect combination  to see this kind of drop(?).

I know we should care more about the science then points,but at the same time, theses kind of "economic" systems must be kept balanced to keep donors' trust. If you rent time on a supercomputer, or other similar scarce ressources (Radiotelescope, etc) You must do your homeworks before it's your turn to optimize your calculations / search. When your alloted time is over, your SOL if you had no results. I humbly think it should be the same here, or remove the "economic" system that are points, and just count working units to show progress (which is a good thing to keep interest of your donors).

On my side, I'll put back the advanced tag, just to try and see for myself how much this really affect WU production on my side at least.
PS: I have tried switching my PIA server, and noticed that I was receiving WU I was not receiving, or rarely, before. Coincidence? Or Is there a "regional" distribution also? I sure would like to know, don't know if there would be a valid way to verify. Because statistically, it could also be explained why this happened when I tried it.

[GOTSpectrum]

3 minutes ago, Cabanaj said:

-snip-

The issue with moonshot units is that they are doing a number of things with the moonshot project, so the individual units vary far more than typical and that means unfortunately the test scenario of running a limited number of units on a set test bench isn't applicable to all the units that are released. Is it hard to strike the right balance between the number of units tested(as it limits the speed which new units can be released) vs giving accurate PPD for the work done.

 

As long as it affects everyone equally then it is still fair imo.

[Cabanaj]

8 minutes ago, GOTSpectrum said:

The issue with moonshot units is that they are doing a number of things with the moonshot project, so the individual units vary far more than typical and that means unfortunately the test scenario of running a limited number of units on a set test bench isn't applicable to all the units that are released. Is it hard to strike the right balance between the number of units tested(as it limits the speed which new units can be released) vs giving accurate PPD for the work done.

 

As long as it affects everyone equally then it is still fair imo.

Agreed, but my theory is more toward the project 169XX, which is cancer related. This project in particular is even more "odd" then the moonshot project. I know they are working on a better distribution system (theses units would be send to "older" cards, cards that do not produce more calculation when used fully).  The thing is, you can not choose to allocate your card toward project that would fully use them. And that also could (maybe) explain so much drop in production, that people are seing theses days, less then a few days after the release of theses advanced WU. Just thinking "outloud" to try and understand this drop wich is big, even considering the variability of the Moonshot Units (The worst thes have done to my prod is make it in the likes of 990 000 ppd a day, compared to <600 000 for the 169XX project thats is more then a 50% drop).

[marknd59]

I have been seeing a big difference in estimated PPD with WUs for the last few days. For example few hours ago I had an estimated PPD of 3M with 1 pair of WUs and a little later 1.7M estimated on the next WUs. I figure it all evens out in the end.

[Kilrah]

1 hour ago, Cabanaj said:

and just count working units to show progress

That's even worse since you have WUs that take one hour and some 4...

[TVwazhere]

2 hours ago, Cabanaj said:

I know we should care more about the science then points,but at the same time, theses kind of "economic" systems must be kept balanced to keep donors' trust. If you rent time on a supercomputer, or other similar scarce ressources (Radiotelescope, etc) You must do your homeworks before it's your turn to optimize your calculations / search. When your alloted time is over, your SOL if you had no results. I humbly think it should be the same here, or remove the "economic" system that are points, and just count working units to show progress (which is a good thing to keep interest of your donors).

I just fold man. I dont care if it's for COVID, Parkinsons, or the new moonshot WU's. As long as ITNOS is contributing and I'm getting points (and I'm not plumeting out of the top 100) then I'm good. 

[rkv_2401]

 

5 hours ago, Cabanaj said:

-snip-

 

I know we should care more about the science then points

 

-snip-

I humbly think it should be the same here, or remove the "economic" system that are points, and just count working units to show progress (which is a good thing to keep interest of your donors).

That’s a good point you raise. You can do that, then. Just count the WUs per donor as progress.  The “point system” is somewhat flawed, IMO, as it isn’t reflective purely of amount of work done, work units processed, or amount of time (and electricity!) donated to the project  - I had more points in a team with 140 WUs done than the previous guy, who had completed 13500 something. I don’t see myself as having contributed more than them. I personally evaluate myself on the number of WUs done - that’ll remain fairly consistent throughout the years and with scaling of hardware. 
 

Any economic system is only as good as the amount of trust people have in it(IMO).

And considering that there are no real ‘value’ to points in the first place, I’m sure you can tune out of that and just count WUs done as a measure of progress. I only see points as being important if you’re mining curecoin or folding for a team that pays you for the points you score them, and of course, competitions.

[GOTSpectrum]

1 hour ago, rkv_2401 said:

 

I had more points in a team with 140 WUs done than the previous guy, who had completed 13500 something

 

One of the reasons for people having stupid high WU count in the past is NaCl folding, which isn't a thing anymore as google stopped support NaCl apps a while ago. These were tiny WUs that folded in the browser and you'd chew them stupid quick. The points reflected the tiny nature of the WUs. 

[Cabanaj]

Well maybe some good news to folks not receiving GPU wu ? 
I have put back the advanced client-type tag, and I have been receiving non-stop WU from project 16600. If I read correctly what Lars' extension, it has few entries still in the Database (at least for the AMD RX5700XT)

They seems to be quick to be done (1.5 hour)  and doesn't affect the production for the day. 

This is the third one I have since I restarted folding at 12 PM.

Thought you guys would like to know.



edit:
yep, all fresh:
https://foldingforum.org/viewtopic.php?f=24&t=35806

Quote

 

COVID-19 GPU (OPENMM_22) project 16600 moved to Advanced

by xhuang » Fri Jul 17, 2020 10:47 am

CPU (GRO_A7) project 16600 to Advanced and to FAH in 2 days
Project descriptions here: https://apps.foldingathome.org/project.py?p=16600

<stats-credit v="8200"/>
<timeout v="1.6"/>

 

Wait a minute... Gromac? nah... must be a copy-paste error...

 

[Kilrah]

1 hour ago, rkv_2401 said:

I had more points in a team with 140 WUs done than the previous guy, who had completed 13500 something. I don’t see myself as having contributed more than them.

Again some WUs are small, so you can/could rack up a high WU count with comparatively little work.

[Cabanaj]

2 hours ago, rkv_2401 said:

I only see points as being important if you’re mining curecoin or folding for a team that pays you for the points you score them, and of course, competitions.

Well I will admit I had this in my mind when I started to talk about the "economic" argument. But after a quick reconsideration, I guess the curecoin folks would'nt activate the client-type = advanced tag, especially if there are units that are "sub-optimal" for the return on investment (running cost). 

At the end of the day, I never purged a WU, and let them complete. I will admit being a little annoyed seing the drop in "production" (I'm only human). 
I was more concern on the effect it could have to see points overall dropping so much, for so many users (and team) at the same time, on the folding effort.

[Kilrah]

Note that on these low yield WUs the running cost also reduces, it's just that they're not able of using all the capabilities of powerful cards so their power consumption also goes down.

 

I'm running one of those 1.4M PPD WUs right now (normal on my card is 2.1), but my card is also only drawing 170W compared to 250 when it's on a "normal" WU.

[Cabanaj]

@Kilrah If i can believe what Asus GPU Tweak II was telling me, it was using full clock. For 4.5 hours, giving < 600k ppd. (Project 16909, so we can compare similar project)

I have just seen a good suggestion on the official forum:

Quote

"It may be good if next versions of OpenMM and FAH_cores would be modified for crunching such small atom count projects parallel in pairs, 2 simultaneously at each GPU. Or even 3 or 4 in parallel with powerful GPUs with very high CU counts."

I won't pretend I know enough about F@H or GPU and how they work, to judge if it's feasable. But if it is, that would be great for everyone. Modern card would crunch theses WU  at their "normal pace"  and researcher will have more data, faster.

[Kilrah]

Here clock is full, utilization also shows "100%" but that doesn't mean it's used at full power.

[rkv_2401]

11 hours ago, Cabanaj said:

At the end of the day, I never purged a WU, and let them complete. I will admit being a little annoyed seing the drop in "production" (I'm only human). 
I was more concern on the effect it could have to see points overall dropping so much, for so many users (and team) at the same time, on the folding effort.

Yeah, that's one of the reasons I have slowly moved away from the points system. I have a laptop with a 940M, with a theoretical peak of 880GFLOPS. That takes 2 days to complete a WU, and only nets me 30-50k points at the end of it, so 30k PPD. My 1060(4TFLOPS) does a similar WU(a WU from the same series, with the same base credit) in 4 - 6 hours, and it nets me 150k credit with a PPD of 350-400k. I don't feel like my contribution on my desktop is 10-15x more 'valuable' than the WU I ran on my laptop, even though it is significantly faster. So I measure my personal progress with WUs done. The certificate you can get also talks about WUs done and not points, IIRC.

 

12 hours ago, GOTSpectrum said:

One of the reasons for people having stupid high WU count in the past is NaCl folding, which isn't a thing anymore as google stopped support NaCl apps a while ago. These were tiny WUs that folded in the browser and you'd chew them stupid quick. The points reflected the tiny nature of the WUs. 

Oh, I didn't know that. I agree with your point, though. Even in this time, CPU WUs get processed a whole lot quicker than GPU WUs, and are nowhere near as computationally intense as their GPU equivalents. Hence their lower points.

 

11 hours ago, Kilrah said:

Note that on these low yield WUs the running cost also reduces, it's just that they're not able of using all the capabilities of powerful cards so their power consumption also goes down.

 

I'm running one of those 1.4M PPD WUs right now (normal on my card is 2.1), but my card is also only drawing 170W compared to 250 when it's on a "normal" WU.

Yep, that's a thing. There are WUs that often make my card draw 70W and offer slightly less points, and there are large WUs that push my card up to 100W and offer a boatload of points.

[explosivetnt]

33 minutes ago, rkv_2401 said:

I don't feel like my contribution on my desktop is 10-15x more 'valuable' than the WU I ran on my laptop, even though it is significantly faster.

But isn't a faster WU return part of the thing that both researchers and we are striving for? And IMO the higher point return for procesing a unit faster is just another incentive to use more powerful gear which in return would allow the generational changes between WU happen faster.

[GOTSpectrum]

4 hours ago, explosivetnt said:

But isn't a faster WU return part of the thing that both researchers and we are striving for? And IMO the higher point return for procesing a unit faster is just another incentive to use more powerful gear which in return would allow the generational changes between WU happen faster.

Yep.

 

WUs are released sequentially. In other words when WU is revieced more WUs can be generated from the results of the data created. This is why a rapid turn around is important for the FAH network and in turn why we have the QRB system in the first place. 

[rkv_2401]

19 hours ago, explosivetnt said:

But isn't a faster WU return part of the thing that both researchers and we are striving for? And IMO the higher point return for procesing a unit faster is just another incentive to use more powerful gear which in return would allow the generational changes between WU happen faster.

I do mention that in my post. Yes, faster = better, but I, personally, don’t value it as 15x or more times better. Yes, it is an incentive to use more powerful gear(although maybe they’d prefer donations rather than an end user purchasing new hardware purely for folding) + finish a WU in one go, and to finish WUs well before their deadline.

 

It’s just that a) I don’t like to have to micromanage, I’m donating my computers SPARE power, and I shouldn’t have to wait for a WU to finish before I, say, start playing a game or let a family member use the computer.

 

b) I would find it demotivating if a person with a 2080Ti’s contribution was 1000x more valuable than mine with a 940M, and then there are slower GPUs and CPUs that only produce 3000-5000PPD. It begs the question, what’s the point of stressing out my hardware (and paying the electricity costs) if someone’s contribution (in processing the exact same WU - the 940M received the same 1174x series WUs, with the same base credit, as my 1060 and the rest of us did in March) is 1000x-10000x more valuable than mine? 

I don’t think there’s anything wrong with the system as it’s implemented. It’s actually really good, it incentivises just the right thing for enthusiasts to focus on(a quick turnaround, the faster you send it back the quicker more WUs can be generated, as @GOTSpectrum mentioned above) and it’s excellent for competitive purposes. It’s a great benchmark, as well. An enthusiast can buy a bigger, better card and see the enthralling rise in performance thanks to the QRB. 

I just don’t personally subscribe to it on a day by day basis. I’d rather use my computer freely, and I’m okay with losing a few thousand - ten thousand points here and there. 

[mattheginger]

1 hour ago, rkv_2401 said:

I do mention that in my post. Yes, faster = better, but I, personally, don’t value it as 15x or more times better. Yes, it is an incentive to use more powerful gear(although maybe they’d prefer donations rather than an end user purchasing new hardware purely for folding) + finish a WU in one go, and to finish WUs well before their deadline.

 

It’s just that a) I don’t like to have to micromanage, I’m donating my computers SPARE power, and I shouldn’t have to wait for a WU to finish before I, say, start playing a game or let a family member use the computer.

 

b) I would find it demotivating if a person with a 2080Ti’s contribution was 1000x more valuable than mine with a 940M, and then there are slower GPUs and CPUs that only produce 3000-5000PPD. It begs the question, what’s the point of stressing out my hardware (and paying the electricity costs) if someone’s contribution (in processing the exact same WU - the 940M received the same 1174x series WUs, with the same base credit, as my 1060 and the rest of us did in March) is 1000x-10000x more valuable than mine? 

I don’t think there’s anything wrong with the system as it’s implemented. It’s actually really good, it incentivises just the right thing for enthusiasts to focus on(a quick turnaround, the faster you send it back the quicker more WUs can be generated, as @GOTSpectrum mentioned above) and it’s excellent for competitive purposes. It’s a great benchmark, as well. An enthusiast can buy a bigger, better card and see the enthralling rise in performance thanks to the QRB. 

I just don’t personally subscribe to it on a day by day basis. I’d rather use my computer freely, and I’m okay with losing a few thousand - ten thousand points here and there. 

I agree with most of that. It’s too disconnected from the actual computation taking place, and is extremely discouraging to people with lower performance hardware. 
 

I can understand the concept of a QRB, but surely it should be a garnish rather than vastly disproportionate to the base credit. 
 

Call me silly, but I have a significant amount of old hardware that I would be running if it weren’t for the pitiful points they earn, relative to my 2070s (primarily because of the QRB characteristics). 
 

It may be sort of irrational relative to the cause, but the point system disincentivizes it. Unlike say World Community Grid etc.. 

 

Extra note: Fah seems only really practical for enthusiasts, unlike WCG, which is much more practical for everyday people to contribute to (where you can start/stop easily without penalty etc).
 

 

[Kilrah]

The points/QRB also make sense from an energy efficiency standpoint. Your old card may be providing results to science calculations, but it's doing so with abysmal energy efficiency. Makes sense to reward / encourage newer hardware that can be literally 10 times more efficient and discourage use of old stuff that's wasting so much energy for doing the same thing.

[mattheginger]

55 minutes ago, Kilrah said:

The points/QRB also make sense from an energy efficiency standpoint. Your old card may be providing results to science calculations, but it's doing so with abysmal energy efficiency. Makes sense to reward / encourage newer hardware that can be literally 10 times more efficient and discourage use of old stuff that's wasting so much energy for doing the same thing.

I understand the efficiency argument, but surely that's not Fah's main concern. Maximal computing power should be. Not everyone is an enthusiast who's going to upgrade their hardware for the purposes of this contribution. 

 

When I began contributing at the end of 2018, I had an HD7870 which has roughly a fifth of the raw compute performance of a 2070s, and yet it only gets around 5% of the points that the 2070s does, for the most part because of the quick return bonus. Cards like that hold valuable compute power that could be harnessed, but the points system makes them appear nearly worthless.

 

A 1050Ti has a third the compute power of my 2070s, but only pulls a 7th (~250k PPD) of the points... etc... That's a power efficient card, and yet people may be discouraged from using it. I acknowledge the purpose of a QRB, but it just seems to distort everything in it's overkill nature.

[chaozbandit]

All this talk of efficiency, and here I am wondering what people are doing with their old bigadv rigs. Always though it would be fun to own a 4P board back in the day of cpu qrb.