Suggested Change to the PPD System

[Grandpa_01]

And I understand your proposition. As I said I do not know if it is good or bad. What about the people that want to catch the people that are at the top will it be possible. That is part of the inflation thing. To me it does not matter I started over at 0 again yesterday but to some it does. To me the points are a fun thing and a guide and that is it. But to some they mean the world.

[bruce]

One thing I would note is that to anyone who says computers don't experience declines in PPD under the current system - that clearly isn't true. I just have to look at my i7 920's PPD over the past 30 months to see that PPD has dropped on that machine. Core changes/optimizations & increases in the size of proteins and of course, drops from bigadv to regular SMP can also have a detrimental effect on older hardware which also results in lower PPD. Because people 'expect' that the ppd of their computer should remain constant any of these changes leads to quite a lot of disconcent amongst folders whenever something happens that causes any sort of change.

Perhaps the facts appear somewhat different depending on whether a person is one of the many folks who is sore about being reclassified as eligible for BigAdv to being limited to standard SMP and folks which that change didn't effect. Many of those who were effected by that change are unhappy -- and although the Pande Group has explained the reason they needed to make an unpopular decision, there is no way to make that a popular decision. I doubt that many who were
not reclassified really care.

Core changes/optimizations & increases in the size of proteins are what I'll call a red herring. The benchmark machines see the exact same core changes, optimizations, and changes in the size of proteins. The number of frames per minute on your machine may change, but the benchmark machine sees EXACTLY the same changes in cores, optimizations, and size of proteins. If your hardware is identical to the benchmark machine, you will still be seeing EXACTLY the same PPD. Both your machine and the benchmark machine are limited by some resource that's saturated. Most commonly, we assume it's a single resource that's saturated -- probably the CPUs but it's also possible that there is a different mixture of saturated resources, so the PPD will not track exactly. That's a limitation of the benchmarking process, but changing the benchmark machine is guaranteed to be favorable to some people and unfavorable to others, so it can't really be "fair" so it's best if it's no changed.

Consider the following two scenarios. Suppose the Pande Group made up for 10 years of computer growth by making a one-time adjustment. On Jan 1, every Wu would be worth 100x as many points as it's worth now and everybody who has accumulated X points would instantaneously have 100X as many points to their credit. Would anybody really care? OK, now suppose we can somehow magically change history and instead of making it happen all a once, they simply assume that we have doubled the point every 18 months (retroactively). (I think that works out about the same. Which one would make the donors happier? And who would really care? And which one is more work for the Pande Group? I think that on that basis, there's no valid reason to change anything.

Was changing the BA-8 requirement to BA-16 fair? Certainly if you were one who was excluded, then it didn't seem fair. But from a scientific perspective, there's a certain amount of work that can be done by the top 10% of the hardware pool and assigning it to the top 20% of the hardware pool (1) slows down that work and (2) creates WU shortages and (3) the work that SHOULD be being done by the second 10% of the hardware pool still needs to be done by that second 10%. The issue here is balancing FAH's total workload against the pool of available hardware, not an issue of points. Your proposal won't change the fact that the criteria that used to separate the top 10% needed to be tightened so that it still identified the top 10%, not the top 20% but at the same time, the number of projects in the top 10% of the workload didn't grow.

I certainly understand your feelings (and those of other donors). The the the balance of projects vs. hardware is a moving target, and I really don't see how adjusting the points system every few months is going to define a way to adjust the criteria that distinguishes the first 10% from the second 10% in a way that the donors consider "fair". The only thing you might accomplish is to hide the fact that some people are being forced to be in the second 10% by the growth of hardware that's in the top 10%.

Looking at projects for GPUs and for Uni/Duo/Quad/Hex processors, or for 16-, 24-, 32-, 48-core machines, absolutely nothing has changed -- and there's no good reason for it to change. [Well, that's really not true. Deadlines for the latter group have gotten shorter, more closely matching the work that they can actually produce.]

[k1wi]

I had been wondering when the truth was going to come out.

I suspect that the hidden agenda for this topic has finally been revealed. You're one of the many folks who is sore about being reclassified as eligible for BigAdv to being limited to standard SMP. It's certain that many people are unhappy with that change -- and although the Pande Group has explained the reason they needed to make an unpopular decision, there is no way to make that a popular decision.

Bruce - are you serious?!

I stopped folding BA long before the changes to BA moved from 8 to 16. I folded my last BA-8 sometime in the first half of 2011, when I switched over to plain old SMP. I'm sure you can go through my WU history to see that. I have never had a problem with PG ensuring that only the top folders are able to fold BA - and I have never had a problem with PG adjusting BA so that the number of BA folder fitted the number they needed/wanted. If anything I am in the pro-adjustment group.

The quote you selected was aimed at illustrating that even under the existing system PPD changes - in my case it went down, for other users (such as with the advent of QRB) it's gone up.

Furthermore, the proposal I have suggested doesn't make any change to the proportion of points at a given time, the points curve remains 100% the same. All that changes is that it remains relative across different time periods.

[Jesse_V]

Grandpa_01, in order to catch high-ranking active folders you'd just need something which is more ahead of the average than they have. It will still be possible to smoke them in a short period of time as long as you have something really powerful, just as the situation currently is.

I think its important to make the critical distinction that k1wi's proposal does not remove that incentive. This will not completely remove the ability for newcomers to catch veterans, it will just make it more difficult, which is IMO a good thing. Remember, we have a number of variables that we're dealing with here. Computational resources and overall time both will figure into current levels of your PPD. Its important to note that the QRB curve is far more steep than the technology arms race curve. Under both systems, PPD is non-linear and you get more for quickly returning WUs. Currently, over long periods of time, the amount you earn is roughly constant, despite the fact that F@h's average computational power is increasing, leaving you in the dust after a long period of time. For the same price, eventually someone new will be able to do in a day what you do in a month, and will quickly overtake your contributions. While completely exciting and inspiring for the newcomer, (believe me, it is) this is discouraging for the veterans who are probably particularly valued by the PG for doing a lot of the early work during development periods. But you can still upgrade and leave others behind and earn a constant ton of PPD, which is why people build these powerful computers. Under the new system, your otherwise constant PPD will slowly decrease over time because just as before your system is getting more and more "meh" compared to the average as all systems do. It's a given that eventually my 560TI will become as my 240m, and as such I'll need to upgrade for other reasons besides F@h. Think of games. And here is one of the reasons people don't like the proposition: the valued few who have bought expensive systems will see their PPD slowly decline under the new system. Unfortunate as this is, that's the way computing systems go. However, because the QRB will still be in place, you'll still have an incentive to get these systems because they will smoke the average computers until it too becomes average or below.

The idea has pros and cons. Compare the current system with the new idea: which is a more comfortable idea, to earn a constant PPD and down the road have newcomers smoke you, or to earn a very slowly decreasing PPD and be passed up by others who are staying up-to-date on their hardware? Remember, under both you can still upgrade and beat your friends, it will still be a race.

I wish I could be in the elite bunch who run -bigadv and are so tied to their PPD. I'm grateful Grandpa_01 has a better understanding of points. If I understand the proposal right, I think it will be hard for some because they expect tons of PPD for their money. I mean I did for my SMP+GPU system. However the inflation is a long term problem and we need to focus on that. To me, k1wi's idea seems fairly intuitive and makes sense, and though it may be a bit strange at first will solve a lot more problems down the road. If its all about delivering PPD based on the QRB and F@h's average PPD, then I support that.

[k1wi]

For full disclosure, according to HFM.net my last big-adv project was P6901 (R23,G0,C26) completed on the third of May 2011. I understand that the change to BA-16 was announced in November 2011.

I stopped BA8 voluntarily for a number of reasons:

1. At the time there weren't many users folding uni-proc clients in the section of the forums that we don't discuss here (if you want to see the hit on PPD that running 4x uni-proc clients makes instead of 1x8 on SMP, I'm happy to show it). I thought that dedicating a fast-clocked and powerful computer to these work units was in the best interest of the project. That situation has since changed, particularly with many A4 projects being uni-proc and multi-proc capable, so I'm back running -SMP 8.

2. I decided that even though my client was, at the time, well within BA deadlines, PGs indication was that my i7 was better suited to running regular SMP.

I don't believe that my proposal changes the incentive associated with BA & QRB at any given point in time, because the underlying formula is the same, it's just being frozen in time.

Unless you can suggest otherwise bruce?

[bruce]

I edit that several times and the system logged me off a couple times during the process. Upon re-reading it, I've changed to tone considerably. The facts don't really change, but certainly the attitude -- and I'll apologize for that. I certainly didn't mean it to be an attack on any one person, but rather a different way of looking at it. The fact that I did quote only k1wi makes it seem more personal and I regret that.

Please accept my apology, K1wi.

No, I'm not suggesting that you alter any of your WU selections. AFAIK, the only thing that was out of balance was BigAdv.

[k1wi]

And my position remains the same: keep the QRB points curve, keep the benchmark system, retain the incentive toward faster computers, just remove the source of point inflation and therefore remove the exponential growth in absolute PPD both absolute points and relative absolute point differentials.

I have even said that PG should retain the points premium of BA (is it 20% or 50% at the moment?), in fact, my proposal probably makes it easier for them to shift the curve as required because it builds into the points system an expectation that PPD can change.

My proposal gives PG MORE ability to adjust PPD in order to match their scientific value. If they really wanted to take the proposal to it's fullest extent, and the more I think about it the more I really think they could, they could easily use the quarterly update to make adjustments if and when they are needed.

No hard feelings bruce - I'm just trying to propose the best points system I can.

[Amaruk]

I think ChelseaOilman might have plenty of company.


k1wi has repeatedly proposed a baseline of 100 PPD. A conversion factor of 100:1 results in an i7 930 @3.8 getting 130 - 200 PPD.

Obviously, achieving the results k1wi is after will require an even larger conversion factor, but 100:1 makes the math much easier for a quick sanity check.

So 100:1 it is.


FWIW, these are all clients I'm currently running.


CURRENT PPD:

2800+ (CPU) = 125 PPD

PS3 = 1,100 PPD

9850 (CPU) = 5,500 - 6,600 PPD

9800 GT (GPU) = 5,600 - 6,300 PPD

GTX 295 (GPU) = 6,000 - 8,000 PPD (per core)

i7 930 (CPU) = 13,000 - 20,000 PPD (SMP)

dual 5620 (CPU) = 55,000 - 70,000 PPD (SMP)

dual 5620 (CPU) = 100,000 - 135,000 PPD (bigadv)

dual 5645 (CPU) = 102,000 - 112,000 PPD (SMP)

dual 5645 (CPU) = 165,000 - 218,00 PPD (bigadv)

quad 6174 (CPU) = 135,000 - 142,000 PPD (SMP)

quad 6174 (CPU) = 355,000 - 500,000 PPD (bigadv)


PPD AFTER ADJUSTMENT:

2800+ (CPU) = 1.25 PPD

PS3 = 11 PPD

9850 (CPU) = 55 - 66 PPD

9800 GT (GPU) = 56 - 63 PPD

GTX 295 (GPU) = 60 - 80 PPD (per core)

i7 930 (CPU) = 130 - 200 PPD (SMP)

dual 5620 (CPU) = 550 - 700 PPD (SMP)

dual 5620 (CPU) = 1,000 - 1,350 PPD (bigadv)

dual 5645 (CPU) = 1,020 - 1,120 PPD (SMP)

dual 5645 (CPU) = 1,650 - 2,180 PPD (bigadv)

quad 6174 (CPU) = 1,350 - 1,420 PPD (SMP)

quad 6174 (CPU) = 3,550 - 5,000 PPD (bigadv)


Obvious question: How does slashing EVERYBODY'S PPD provide a scientific advantage?

...in order to catch high-ranking active folders you'd just need something which is more ahead of the average than they have. It will still be possible to smoke them in a short period of time as long as you have something really powerful, just as the situation currently is.

I think its important to make the critical distinction that k1wi's proposal does not remove that incentive.

My math sense is tingling.

Have to run some numbers to be sure, but my guess is the opposite will be true.

[k1wi]

Hi Amaruk,

I haven't suggested reducing everyone's PPD by 100:1 - that is a lot of slashing and not something I would even consider. I think MtM mentioned it once? At best under a 3 monthly system I imagine it'd be a reduction of around 4% each quarter. Never a one off of ~99%? The idea of the proposal is that it keeps the proportions the same, it just keeps the absolute differences small enough that it doesn't discourage folders.

I have attempted to explain my proposal by saying that if the 'average' computer gets 100ppd in one time period, the 'average' computer in the future should also get 100ppd. I don't know the 'average' starting point should be - it could easily be current 'average'/'median', where we say "lets keep the status quo and not have things get further out of hand". If we took the current values then any drop in values would be limited to the increase in computational power over 3 months, which as I said would be around 4% I believe.


The reason I took a bit of a break from pushing this discussion is because I want to really lock down the formula and provide some real world results. Unfortunately, the project I chose to create my model (8004) is no longer on the psummary so I'm going to have to use a new project and values to create some real data and between research and teaching my time is pretty short at the moment

[v00d00]

Why dont we just scrap points completely and move to workunits completed as the main currency. Or maybe we move to a seperate system, where points awarded for GPU get a different certificate and are counted on a different system to points awarded on say BigAdv. If people can only fight within the areas they can fold then it evens things out. If you fold BigAdv, then points arent comparable to GPU or CPU clients, because they are seperate areas, with a seperate stats table. If SMP users want to screw with their points system, then let them, as the points system of SMP has no relevance to BigAdv, due to it being a different area.

In the end what do those points stand for? Nothing whatsoever. They are a method of manipulating the ego of users, a psychological tool for recruitment, a psy-ops tweak for the weak minded, and it works, as we have an ever expanding number of folders joining us.

What really shows contributions is the number next to workunits completed. That is what shows your contribution to science, which is the important part of FAH, the bit where we help to cure diseases. As opposed to the petty squabbles of a handful of people who complain because they managed to hack the system for a while to allow them to do something they were never supposed to do and are now crying because someone is patching the exploit.

You dont run the project, i dont run the project, and this thread is pointless and a waste of bandwidth. PG will do what they see is right, regardless of you, me, obama, god and anyone else that cares to raise their voice. Like ive said in the past if you dont like the system, no one is forcing you to run the software.

[vbironchef]

Why dont we just scrap points completely and move to workunits completed as the main currency. Or maybe we move to a seperate system, where points awarded for GPU get a different certificate and are counted on a different system to points awarded on say BigAdv. If people can only fight within the areas they can fold then it evens things out. If you fold BigAdv, then points arent comparable to GPU or CPU clients, because they are seperate areas, with a seperate stats table. If SMP users want to screw with their points system, then let them, as the points system of SMP has no relevance to BigAdv, due to it being a different area.

In the end what do those points stand for? Nothing whatsoever. They are a method of manipulating the ego of users, a psychological tool for recruitment, a psy-ops tweak for the weak minded, and it works, as we have an ever expanding number of folders joining us.

What really shows contributions is the number next to workunits completed. That is what shows your contribution to science, which is the important part of FAH, the bit where we help to cure diseases. As opposed to the petty squabbles of a handful of people who complain because they managed to hack the system for a while to allow them to do something they were never supposed to do and are now crying because someone is patching the exploit.

You dont run the project, i dont run the project, and this thread is pointless and a waste of bandwidth. PG will do what they see is right, regardless of you, me, obama, god and anyone else that cares to raise their voice. Like ive said in the past if you dont like the system, no one is forcing you to run the software.

Well that pretty much sums it up. I can't argue with that statement.

Just wondering if it would be possible that a dollar amount donation could be given a points component to the donation that could be added to a persons overall score?

[Nathan_P]

But WU count can be manipulated as well - its human nature to be competitive, people can say that they are not but i don't think that there are many on here who never look at their stats.heres an example:-

I run 3 dual hex machines - under combined Bigadv/SMP they earn about 240k PPD with an average of 3 to for 4 WU a day, now under the WU counts i could push that higher by running 12 uni-proc clients on each machine giving me about 28 WU a day (assuming 24hrs per WU). Now does this help the project more or my "wu score"? Certainly it will help with the pile of uni proc that are out there but what about the SMP/Bigadv units? these are now not worthy of running as people race to complete as many WU as possible. Or how about a gpu farm that does lots of GPU specific Wu but leaves the cpu units for others - is that fair?

The current system isn't perfect, not by a long shot but i've yet to see a better alternative despite some noble efforts. The only sure fire way is to remove the QRB.

Oh and an up to date benchmark machine for the larger WU might be a good idea as well!

[Jorge1950]

Hi all. It is insisting that the change of hadware going to be forced by the PPD. This is false, for fevers of computer science. And for those who are fevers, more sooner than later the shrink.

I ask all the participants, how many times have gone 12 months without major hardware environment changes? Change is the norm, with or without FAH. Always seek the reason we founded for change, in fact the reason is you.

Over 13 pages have highlighted this aspect. Thirty-five years of walking in the computing world, as a user, teacher, consultant and developer, I say that human nature, will propel us to sacrifice ourselves for having the best.

Español:
Hola a todos. Se está insistiendo en que el cambio de hadware va a ser forzado por el PPD. Esto es falso, para los fiebres de la informàtica. Y para los que no son fiebres, màs pronto que tarde la contraerán.

Yo pregunto a todos los participantes, cuantas veces han pasado 12 meses sin hacer cambios de hadware importantes? El cambio es la norma, con FaH o sin èl. Siempre buscaremos la razòn que nos fundamenta para el cambio, en realidad la razón es uno mismo.

A lo largo de 13 páginas se ha destacado este aspecto. Treinta y cinco años de andar en el mundo de la computaciòn, como usuario, profesor, consultor y desarrollador, me dicen que la naturaleza humana, nos impulsará a sacrificarnos por tener lo mejor.

[7im]

Mental exercise...

Assuming 1 point equals 1 WU when the project started in 2000, and the processing power doubles every 18 months, we go from 1 point to 256 points in 12 years. Then add in 3x speed up when SSE was added 2003ish. 2x speedup with Intel and AMD doubled the SSE bit-width on core 2 chip technology few years later. We end up at 1536 PPD, just assuming Moores law, and 2 basic chip improvements.

Big WU bonus (2x for SMP, a project based bonus) = 3072 PPD.

4 cores? 12,288 PPD
8 cores? 24,576 PPD
64 cores? 196,608 PPD
128 cores? 393,216 PPD

See, we really aren't that far off the PPD we should be getting just with natural technilogical advance. Ya, sure, I picked convenient numbers, and they fit nicely, they also have the benefit of being accurate.

Anyone see a problem with PPD going forward?

[vbironchef]

Mental exercise...

Assuming 1 point equals 1 WU when the project started in 2000, and the processing power doubles every 18 months, we go from 1 point to 256 points in 12 years. Then add in 3x speed up when SSE was added 2003ish. 2x speedup with Intel and AMD doubled the SSE bit-width on core 2 chip technology few years later. We end up at 1536 PPD, just assuming Moores law, and 2 basic chip improvements.

Big WU bonus (2x for SMP, a project based bonus) = 3072 PPD.

4 cores? 12,288 PPD
8 cores? 24,576 PPD
64 cores? 196,608 PPD
128 cores? 393,216 PPD

See, we really aren't that far off the PPD we should be getting just with natural technilogical advance. Ya, sure, I picked convenient numbers, and they fit nicely, they also have the benefit of being accurate.

Anyone see a problem with PPD going forward?

I for one don't.

Just wondering if it would be possible that a dollar amount donation could be given a points component to the donation that could be added to a persons overall score?

Instead of upgrading equipment just give a cash donation instead and be given points for it. What do you think?