Bigadv points change

[Grandpa_01]

Well I myself do not need a rigid point system they will all be flawed when you figure in the H factor (human) as I said before there is nothing that is wrong with the current point system. It just needs to be adjusted manually occasionally, kind of take the curve out of it. Use a point a and point b set at a 45 when the curve hits it follow the line to point b.

[ElectricVehicle]

Sorry to belabor the point, but I still don't see any specific suggestions. I know that specifics are hard, but that's the point (in a sense, "talk is cheap"). How about we avoid debating in some non-specific sense and instead debate specific, detailed suggestions? (not just the equation, but the values that go into it as well).

Vijay - base the points system on the scientific value. I can't presume to play with the formula definitions or parameters to optimize the science, your team knows that far better than I ever will. You might however want to publish a little more information about why a particular formula, value or K factor was selected based on the scientific value. While there will still be some debate, it's less likely people will see it as arbitrary.

Once you publish and implement the formulas and values, I use that to adjust my software, hardware and networking folding strategy to produce the most benefit to the science based on the points that I can see and compare between various configurations. It it weren't for bigadv, I'd still be running quad core SMP on all my computers with GPUs also. Bigadv showed me the scientific value of getting am i7-980x with over 6GB of RAM, so I saved and moved stuff around to construct such a system because the points system said there was REAL scientific value to such a system.

I have constraints in what I can do in terms of power, heat and noise. All of my folding hardware, boxen, has to use less than 1kW (1,000 watts) in total for all the boxes. I got into folding on PS3's which I originally purchased for Blu-Ray. I used to run them 24x7 for 1,000 PPD at 200 watts of power. I bought a second some months later for based on it's scientific contributions, and, errr... entertainment value at work. I have in the last year stopped both of those PS3s from folding because in my personal situation, with heat and power limits, based on the PPD, there was more scientific value in running multi-core CPUs with a couple fast GPUs in the computers I was building for shared use of folding, home and work. With budget of 1,000 watts, watts/PPD drives me to plan upgrades to my shared use computers to produce the most scientific value (Points, PPD) per watt - PPD/W.

Let the scientific value guide the formulas and values that determine points. We will follow and adjust as that changes via the points to return the most scientific value for all. I seriously question adjustments to the formulas and values that distort or interfere with the correlation between points and scientific value.

Some trajectories are more promising scientifically than others. Some projects consisting of many WU generations where one generation can't be started until all the WUs of the previous generation complete because they start from the data from the end of the previous generation may need to complete during a graduate students career at Stanford or even within a semester, so Quick Return Bonuses make a lot of sense.

When someone (Google? The NSA? EVGA? etc...) comes up with a machine that can complete WU's in minutes instead of hours, and complete entire multi-generational projects in days instead of years and they got astronomical point values - MORE POWER TO THEM!!! We all benefit! (and we'll all figure out lower cost ways by sharing information to do similar things!) If the astronomical points are not correlated to the scientific value, then adjust the points to bring them back in line with the scientific value, be that lower or higher. The points give donors feedback in how best to benefit and accelerate the science. If the points and points/watt tell me, for my particular situation, to ditch my single core celeron and PS3 that were folding in favor of SMP machines with GPUS stuffed in, and maybe a bigadv folder - so be it! And I've done it! It's all about the science.

It's all about the science.

It's all about the science.

[P5-133XL]

I'm sorry, but if the current point system reflects the science then I don't see where we can go from here without alienating the most valuable folders. The primary complaint is that it needs a cap because of the extremely large differential between the biggest folders and the average folder causing the average folder to be discouraged thinking their contribution to be relatively valueless. The fact is that Vjay just said that the point system is relatively accurate to the science. Those big folders deserve the big points.

Those that are getting enormous points, are of two categories. The server admin that is is just donating surplus CPU and those that are folding fanatics that spent a significant amount of money as well as time and effort optimizing those machines to get their PPD. The points are unlikely to be a major factor with the server admin category. On the other hand, the fanatics are much more likely to be driven by the point and are really going to be greatly annoyed when their efforts don't gain what they have been getting. Alienating the fanatics, those that produce enormous amounts of science is not a good way to drive more science.

I argue that we shouldn't try to fix that which is not broken.

I would also argue that we are not the same as when the project started. Folding has matured, but the general attitude has not. We still think in the terms of volunteering surplus CPU cycles encouraging the part time folder on atoms or P4's. However the reality is that folding needs is more full time HT-i7's rather than part time low end machines. If there is a group to alienate it is the bottom end machines rather than the top producers.

The value of the low end group seems to be just in the concept of hooking them into folding so that they can become fanatics so as to upgrade and optimize their folding capabilities. Obviously we need more people but high end people are much more valuable. Perhaps marketing folding among gamer's or enthusiasts rather than just anyone with a machine is a better strategy.

I'm sure that I will get blasted for voicing this but this really has been my take on this thread. I don't like it either but I really don't want to hide in a fantasy and I'm starting to think we all are/were.

[MtM]

Simple poll -> viewtopic.php?f=16&t=18902&start=0

Vote, briefly explain why if you feel the need, but general discussion should remain here.

I'm VERY curious how many people will vote yes.

[MtM]

Subject: point system is getting ridiculous...

With a topic title like this, and almost 9 pages of comment, I think poll is a little redonkulous at this point.

An no offense to anyone, but polls, especially forum polls, are unrepresentative of the whole, and never cover all the possible responses to as to affect the outcome, so I think they're lame.

If you want to poll, make it a simple yes/no question and answer, otherwise we'll need another 8 pages to debate the results of the poll.

But, the 8 pages see mostly the same people arguing, how do I know how many people they represent?

On the other side, the poll will likely attract a larger part of the high ppd obtaining people. They are represented on the forums more ( through registration ) then the people on the bottom layer. If it had been possible I would have set up a poll where people could vote without being logged in ( also open to abuse ).

I'm honest here, advanced math, graphs and curves isn't what I enjoyed at school, but it's mostly you who's pushing a change here if I look at the thread, I know you're most of the times right ( have to account for the miraculous occasions where you're not ) but here I'm not sure if you're really representing the majority of the people. For me the discussion is foccused on ethics not the actual math. If Vijay came up with this formula, is it ethical to ask for a change from a donor's perspective? For me the answer is a simple no. If we do this, we're putting PG in the position where they are going to get blackmailed again and again to change things if a large or very vocal group of people claim the current system is not fair in their eyes.

What was said by p5-733XL adds to what I said, if the science benefits that much more from the higher end machines, maybe it's us donors ( excluding the few who I think are already doing this or have done this ) who need an adjustment to how we see the project?

[7im]

Who said there is really that much benefit to the science that the points curve should be that high?! Now prove it.

Read some history (This is the 2nd time I've posted this link. Read it this time!). New Points FAQ The current points curve was fitted to make 1, 2, and 4 core systems somewhat match up between the older a2 fahcore, and the newer a3 fahcore.

NOWHERE does it say they accounted for 8 core, or 16 core, or 24 core, or 48 core systems. Not in that FAQ, and not in any of Vijay's posts in this thread. They did not form fit the points curve for the actual performance of those future systems. They made a curve to fit 4 core systems, and let the right hand side of the curve trail off the end of page wherever. No foresight! But here we are anyway.

Just for reference, there are people out there folding with just 11 -bigadv systems making more than 1.2 million points per day. Why is that 1,200,000 point mark interesting? That's because the 1100+ users folding under the default PS3 user account (the current #2 ranked user in the project) only makes 1.1 million points per day. 11 systems now equals 1100. Are the 11 SMP work units turned in daily by the first account worth more than the 1100 work units turned in daily by all those PS3 users?

[MtM]

Vijah did when he came up with the existing formula? Then again in this thread where he said he would adjust the already in place formula to better match donor's perceptions ( note perceptions != true value ).

I might be wrong, but that's what I'm getting from the posts made.

I know how distorted the numbers look, trust me I'm looking at stats as well. But I can't say the numbers are really distorted since I have no clue as to where the science stops benefiting from faster returns. I know Angra with his 18 bigadv nodes is having fun almost outproducing the entire 'anonymous' donor group, but who am I to say his contribution with those 18 nodes isn't represented correctly.

I understand your reference to the method chosen for the formula and like I said before I would be interested to see what you would do instead. But, again, how do you know the value of quick returns when you don't have a means to quantify the scientific value of it? That's the question no one has been able to answer except VP, who has come up with the forumula we're now putting on the chopping block to see if we agree with it based on donor perception.

I'm not saying I'm right and you're wrong, but I do wonder why you're so convinced there are so many people going to be discouraged by the big points from those people with 'mini clusters' ( Angra for instance, his nodes are really bigadv ready. )

If you want to argue about the group of people earning high points being wrong due to the i7's, that will fix itself in time automaticly when the requirements go up, and/or when new architectures are released ect. But I'm not even sure if that's relevant to your point?

How long before you get out that calc and show some examples of what you think would be fair? That would make the discussion easier to continue without going back and forth between ethics and perception imo.

Edit: while I still hold to my believe that no change is needed, your example could maybe show a formula which Vijay where able to judge. After all we have to rely on his insights. Also, ps3 work unit's don't have an QRB right? Just as GPU doesn't yet. Why not wait until the playing field levels out and then look at the overall picture. You yourself said atm gpu clients are the point pariahs, how can we judge what will hapen with Gpu wu's when QRB is applied to them ( can't say since we don't know the k value they will get ). It sounds abit premature maybe to ask for a change before the formula is applied to all project ranges?

[P5-133XL]

"We set up the QRB system with a reasonable plan for how the science connects to time. To first order, that's a good estimate of the value of science vs time."

A good estimate is a good estimate is a good estimate. Because you don't like it, does not make it untrue. A mathematical proof is way beyond what is needed. I got all I needed when our glorious leader said that the current point system is a good first order approximation of the value of science vs time. Why should I distrust Dr. Pande's statement to the point I need absolute proof? Answer: I don't!

[MtM]

From the poll thread ->

I'm reluctant to reply as requested because all of this has already been gone over in the other thread. I voted yes because the exponential increase in points at the top end is unsustainable and will lead to the departure of contributors running systems with fewer than 2 sockets.

In addition, I'm embarrassed that I can get 1.75x the ppd that I previously was, just because I happened to get a 6903. I didn't do anything special to get it and it doesn't deserve that high of a reward. It makes my prior contributions all that much less meaningful.

The last paragraph is not about the point system but about adjusting the base points of a single project which is out of line, that happened before the QRB was put into working as well. It's therefore not relevant.

Then the first paragraph you're giving an opinion and disguising it as a fact. Why is it unsustainable? Why would it discourage folders with less then 2 sockets?

[Punchy]

Like I said, all of this has already been covered in this thread. There are posts from folks representing discouraged less-than-2-socket users here, and it's quite easy to find them on other forums.

The reason that only a few people continue in this thread is that they are the only ones with the patience to continue debating the same points. I tend to agree with almost everything 7im has said here, I don't have the time or patience to participate in the arguments, and since he has the patience to continue, I'm happy to let him do so.

Basically you have now invalidated your poll by requesting that people voting "yes" must reply.

Lest I be accused of not contributing to this thread, let me add that a simple solution would be to return to the 10x bonus cap. While I understand that Dr. Pande believed the formula appropriate at the time it was created, I don't believe the area above 10x was considered at that time - perhaps the reason that the 10x cap was originally enforced.

[7im]

...

Lest I be accused of not contributing to this thread, let me add that a simple solution would be to return to the 10x bonus cap. While I understand that Dr. Pande believed the formula appropriate at the time it was created, I don't believe the area above 10x was considered at that time - perhaps the reason that the 10x cap was originally enforced.

More than perhaps, and likely for more reasons than just this. Knowing why the 10x cap was lifted (and not just simply raised) might also help quell the debate some.

[MtM]

You still have not really answered the questions, and I did mention a vote either way needed validation with a post of what one voted and a short version of why, I am not trying to push either way believe me. I believe it should reflect the scientific value and already admitted more then once that a> I don't feel comfortable with the 'human' factor and b> I have no view on what value time has in regard to the value of a work unit.

I'm defending the original formula as I believe it was chosen because it does match the science as close as possible, and I feel that if it is changed now there will be an everlasting tendency of people working to get it changed again and again.

That being said, the post 7im made though lengthy was worth waiting on, and I hope he doesn't mind if I quoted and continued here as well:

I voted yes. And my reasons cannot be summed up in one sentence. I'm not hear to debate my reasons, but simply list them.

First, I do not want to scrap the current points system as some would imply. I would like to tweak it to make it work better. Here are my reasons for voting yes.

* All CPU work units NEED to transition to the QRB system, or they might as well stop sending them out. People will eventually stop folding these without bonus points. They are already stopping now.
* All GPU work units NEED to transition to the QRB system, or people will stop buying them, and simply fold -bigadv. (most home computers going forward will be multi-cored systems, so most will fold SMP by default. Even the V7 client has changed from a default of a CPU slot to a default of a SMP slot for the install.) And one of the top folding teams even recomends not mixing GPUs with SMP clients to get the best performance.
* The bonus points for the new 12 core -bigadv work units has not been defined. They seem to net more points than regular -bigadv work units, but NOTHING has been published on that yet.
* Many people have complained about the points gap between -smp and -bigadv, and has lead people to chase the -bigadv carrot with systems that are marginal for the -bigadv task. Normalizing the difference where one stops and the other begins would alleviate that problem a bit.
* The WU size setting seems to be completely ignored lately. Everything is Big or Fast. Small, Normal, Big mean nothing if everything is Normal or Big. If Small no longer applies, get rid of it, or change the classification. Change Small WU to less than 10 MB instead of less than 5 MB, etc.
* The current QRB system was graphed to fit actual performance numbers for 1, 2, and 4 core systems. I would like to see a small revision in that graph to also form fit to ACTUAL performance numbers from 8, 16, 24, and 48 core systems. That has not been done.


And maybe a completely different debate, but many have asked for this as well. A benchmark system that runs on your own computer instead of some abstract hardware at Stanford that never seems to match your own system's performance. That way we don't get any more P6701s, or P2684s whose performance is deemed outside the normal. But that is a major undertaking. We can at least tweak the current system with the above bullets to make it better, and see if we still need a major change later.

You're amazing, but in a good way. I agree wholeheartedly!

I'm not against changing the system ( maybe you want to rub it in that I didn't choose a good topic title? ) but I am against changing the QRB formula unless it's about the scientific value being different then what it was when put into place.

Edit:

I have to add something..

When people talk about ' I know people who .... ' or ' there are people representing .... ' let them sign up and take part in the discussion. It's useless to represent something with just an opinion, and then hiding behind the fact that there are more with the same opinion without validating that opinion first!

You can validate your opinion with stating why you think that way, it can never be wrong as anyone is entiteld to their opinion. But debating an issue can no be done based on opinions, there is a need to get deeper and without going there any debate is useless.

And as to the 10x limit being removed:

What is there isn't a cap on the increase in value from a quick return, what if it's really 10 times the value between doing it in one hour and in two, but 40 times if you run it in half an hour ( random numbers used ).

Should the points then not incite people to run the high core count multi socket systems if they can?

And when gpu and cpu are also included, as 7im already said it would level out abit as with any system.

It's way to soon to be so negative in my opinion.

I have yet to see someone quit who been folding and active on this forum. Maybe I don't look close enough though.

[MtM]

Also your notice of my bad choice of words has forced me to change the poll thread title to indicate it only covers changing the QRB formula. For those following the last page of posts in this thread that was probably as obvious as it seemed to me when I made it, but that was a very wrong assumption on my end. My apologies.

And, you should change your vote as well 7im, you don't seem to want to change the formula, just finish the roll out of the new system

[7im]

I do want to change the forumula, even if just slightly. The curve was only form fitted to match scientific production from 1, 2, and 4 core systems. The FAQ does not say that 8 core or higher systems were even considered. That is an oversight at the least the 10x cap probably covered. And a mistake not to come back and address it when larger systems came online.

[ElectricVehicle]

I'm sorry, but if the current point system reflects the science then I don't see where we can go from here without alienating the most valuable folders. The primary complaint is that it needs a cap because of the extremely large differential between the biggest folders and the average folder causing the average folder to be discouraged thinking their contribution to be relatively valueless. The fact is that Vjay just said that the point system is relatively accurate to the science. Those big folders deserve the big points.

Those that are getting enormous points, are of two categories. The server admin that is is just donating surplus CPU and those that are folding fanatics that spent a significant amount of money as well as time and effort optimizing those machines to get their PPD. The points are unlikely to be a major factor with the server admin category. On the other hand, the fanatics are much more likely to be driven by the point and are really going to be greatly annoyed when their efforts don't gain what they have been getting. Alienating the fanatics, those that produce enormous amounts of science is not a good way to drive more science.

I argue that we shouldn't try to fix that which is not broken.

I would also argue that we are not the same as when the project started. Folding has matured, but the general attitude has not. We still think in the terms of volunteering surplus CPU cycles encouraging the part time folder on atoms or P4's. However the reality is that folding needs is more full time HT-i7's rather than part time low end machines. If there is a group to alienate it is the bottom end machines rather than the top producers.

The value of the low end group seems to be just in the concept of hooking them into folding so that they can become fanatics so as to upgrade and optimize their folding capabilities. Obviously we need more people but high end people are much more valuable. Perhaps marketing folding among gamer's or enthusiasts rather than just anyone with a machine is a better strategy.

I'm sure that I will get blasted for voicing this but this really has been my take on this thread. I don't like it either but I really don't want to hide in a fantasy and I'm starting to think we all are/were.

We certainly don't want to create any kind of class or caste system amongst folders, but maybe we need two different, non-interchangeable pools of points, for lack of better terms, I'm thinking of "Home" and "Enterprise" or "Home" and "Folding Farm". Better terms for these are welcome!

Single core CPU WUs, and maybe up to SMP 4 core would score points in the "Home" category.

SMP 5 or more cores, bigadv, Quick Return Bonus (often computers running 24x7), other currently big point WUs / configurations would go into the "Folding Farm" category.

I leave it as an exercise for the reader to figure out where GPU units would go! High power GPUs that generate loads of points are available in the home budget, but they overpower the points from single CPU clients. I'm thinking GPU WUs would probably go in the "Folding Farm" category. Or the GPU WUs could be split based on a key GPU architecture number, like number of shaders, say up to 200 shaders is "Home" and over 200 shaders is "Folding Farm" category. The rough basis for the division between "Home" and "Folding Farm" category is the capabilities of a $150 Video card in 2011 as the base line, a price point that is close to a home versus enthusiast video cards. (I know there isn't equivalence between ATI/Nvidia, but that may just have to be, or the shader cutoff could be different between the two.)