Bigadv points change

[VijayPande]

Could you make some specific suggestions for what you'd like to see us change? If bigadv is getting out of control, we can always stop that program or significantly bring it down, although of course other donors will be very unhappy with that. The challenge here is to find some new plan which can get a broad consensus, although I agree there will likely be no plan which makes everyone happy (hopefully we can make more people happy than now).

[7im]

I thought I had been specific, and using a pretty good visual aid too.

Either lower the slope of the QRB bonus curve, or cap the upper limit in some way if the function doesn't change slope well (fixing upper end blows out lower end). I don't think a computer that is 10x faster should be getting 1000x the points. And 6 months from now, the 20x computer gets 20,000x the points? Again, as the curve moves to the right, it approaches inifinity. That points model is not sustainable. Moore's Law moves right along that graph too quickly! You already have 1 million point work units, and they'll be 10 million by years end.

QRB is a must for ALL client types, or the GPU client will be dead within the year. The top teams already recommend against running SMP and GPU at the same time, and (almost) all new computers are SMP ready.

Same goes for the CPU client. If all of the CPU work units are not on the QRB system soon (a4 core), you might as well issue the EOL warning now, and stop making CPU work units by 2012.

You need to redefine the small, normal, large setting in the clients. NONE of your researchers makes small work units any more (well, GPU WUs are small, but they are not desinated that way). It should change to Normal, Large, and XLarge, with some bumps in the sizes. Less than 10 MB, 10-25 MB, over 25 MB (and even this might be too small, 10-50, and 50+ might work better)

The points gap between smp and -bigadv needs to be reduced. You have people almost burning up X6 systems chasing the -bigadv carrot. Not good for the X6, even worse for their power bill to over clock so much (diminishing returns, with exponential power increases), and they're slowing down the results. (NOT picking on AMD here, so don't even start, i7s are almost the same thing)

And you need a new way to benchmark that does not depend on multiple benchmark computers at Stanford that continue to age too quickly, and that seeminly never match my configuration any more. You should be able to benchmark my computer, understand how much science it can process in a given amount of time, and reward my computer accordingly. My least favorite thing to say, but other DC projects do it, so should FAH.

If I need to be more specific than that, just ask.

[orion]

You have people almost burning up X6 systems chasing the -bigadv carrot. Not good for the X6, even worse for their power bill to over clock so much (diminishing returns, with exponential power increases), and they're slowing down the results. (NOT picking on AMD here, so don't even start, i7s are almost the same thing)

Tim, I've been running my x6 for almost a year and haven't burned it up and neither have the ones that I know (and you know where I hang at) have burned theirs up either.

If we want to stop the x6's, the "almost the same thing i7's" and the slow 2ps from folding bigadv/bigbeta to help speed the science up then decrease the base points and shorten the preferred and final deadlines. By doing that it makes it not worth their time to fold those types of WU's. Who would want to fold for a couple of days and only get a few hundred points when they could of gotten several K on a SMP...maybe you, no me.

[Punchy]

If not fixed, we'll soon have systems that produce more points in one day than people have produced in 7 years time. And no offense to anyone, but there is NO work unit, no matter how large or how fast it was returned, should ever be worth years of contribution to the project.

I think we're already past that point. BaBa from OcUK has a system that was doing 920K ppd before the points adjustment.
http://hardforum.com/showthread.php?t=1612201&page=7

I really like 7im's ideas and think they are worth an in-depth look by donors and Stanford alike.

[Haitch]

Maybe:

( K * Deadline / WUTime ) ^ (1/3) * Base.

Use the third root rather than the second, still gives an accelerated bonus for quick return, doesn't send the values so stratospheric.

H.

[Grandpa_01]

You could use a flat rate type system WUX is worth XX points if turned in by X if turned in by Y then xx + 10% and then have a max PPD value for WUX and do the same in the opposite direction and continue to use the preferred and final deadlines. That way everybody knows what they are getting the people with faster systems will get more points (Volume + %) those with slower will get less (Volume - %). I think the simpler the system is the better off we all will be.

[Haitch]

Grandpa,

Much of the interest for hardcore hardware guys (and gals) is tweaking their system to squeeze a couple of seconds off the TPF for each WU. If you take that away by saying (for example) 22:29 TPf is worth just as much as 22:01 you'll kill a lot of the interest and competition. I say use a formula that rewards them for every second they shave off, but one that doesn't devalue the points. As 7im says, flatten out the curve.

H.

[bruce]

Grandpa,

Much of the interest for hardcore hardware guys (and gals) is tweaking their system to squeeze a couple of seconds off the TPF for each WU. If you take that away by saying (for example) 22:29 TPf is worth just as much as 22:01 you'll kill a lot of the interest and competition. I say use a formula that rewards them for every second they shave off, but one that doesn't devalue the points. As 7im says, flatten out the curve.

H.

That may be true, but I doubt it.

Many of those who like to overclock their systems will continue to overclock for maximum performance. Others will stop sooner.

Clearly, whether the PPD changes linearly or exponentially there will always be some PPD benefit to shaving a few seconds off the TPF because the next WU will start that much sooner. There will also be a scientific benefit to returning the WU 100*(a few seconds) sooner.

Is a WU which is turned in 1% faster worth 1% more PPD or is it worth much more than 1% more PPD based on a complicated computation of a factor involving increasingly large kfactors? -- and after you've answered that question, how does that compare to discouraging some new Donors by the relative insignificance of the contribution that they can make with their Dual/Quad?

I've been pondering all the comments made in this topic and I have to admit that I don't have any answers that are better than the ones that have already been proposed. It's a very complicated question with some extremely complicated suggested answers.

Suppose someone has a big system. -- say 48-cores in today's environment. Now suppose that a specific overclocking tweak can improve the completion speed by 2%. That's equivalent to adding one more core. (Putting aside the fact that there's no such thing as a 49-core machine, and even if there were, it wouldn't run Gromacs. We can still talk about it in theory.) How much increase in PPD will that 2% speed improvement produce?

Hopefully, somebody with a 48-core machine can answer the question.

Now the real question. How do you compare that 2% improvement to the PPD that could be generated by running that "extra" cpu-core on a 48-core machine with what could be earned if that "extra" cpu-core is run as a new uniprocessor client? I'm not going to pretend that the two options are equivalent, scientifically speaking, but realistically there should be some kind of comparison that makes sense to a rational person and I'd be interested in hearing it.

[Haitch]

People fold for different reasons.

Some fold for the science, some fold for the competition or the challenge of tweaking.

I fold over on the [H]orde - donors are spending thousands of dollars to overtake a competitor that has passed them - tell them that any 22:xx TPF is worth the same and the howl of outrage will be deafening.

What is so hard about coming up with a formula that rewards every second shaved off the TPF, but doesn't blow out to insane values ?

[Haitch]

Bruce,

And as for your question about the the 2% value - surely that is for Stanford to answer. How does getting that completed WU back 2% faster help their science? Does it help 1%, 2% 10% ? I can't answer that - but what I have read over and over is that the points are relative to the scientific value. If those really fast returned units are really valued that much for the science, what is wrong with the points reflecting the science contribution ?

H.

[Grandpa_01]

Grandpa,

Much of the interest for hardcore hardware guys (and gals) is tweaking their system to squeeze a couple of seconds off the TPF for each WU. If you take that away by saying (for example) 22:29 TPf is worth just as much as 22:01 you'll kill a lot of the interest and competition. I say use a formula that rewards them for every second they shave off, but one that doesn't devalue the points. As 7im says, flatten out the curve.

H.

Ahhh now we a delving into an OCers brain By nature that is dangerous territory. I am one of those idiots myself. I myself will always push to the max (self motivated). You can place as many steps as necessary to entice or reward people. But somewhere there needs to be a cap of some sort otherwise Mores Law will overtake the point system and create huge gaps. I really do not see anything wrong with what 7im suggests I am just suggesting a simple system that is easy to adjust to Mores Law and maybe stop point inflation. I do sympathise with those that have folded from the beginning and there contribution is starting to look pretty meagre because of point inflation.

[Haitch]

Grandpa - yep - very dangerous territory. A stepped method will always repulse them as that extra tweak is not rewarded, the formula needs to accommodate them but it needs to cap them. But I can't devise a method that is reasonable.

H.

[k1wi]

The stepped/capped method does not stop you from having a higher ppd - it might cap the points awarded for each work unit, but completing work units an hour faster will still result in completing more work units and thus earning more ppd, it just limits the exponential increase in bonus the really fast machines earn points.

[Leonardo]

I don't offer a unique solution, but can voice some sentiments. Although I do my best to remain analytic and consider Folding@Home a charitable contribution - hardware purchases and power bill, I can no more deny the emotional draw of F@H than I can deny human nature.

I do sympathise with those that have folded from the beginning and there contribution is starting to look pretty meagre because of point inflation.

Indeed. I have been Folding nearly from the beginning - within just a few months of the project going public. Yes, I understand that offering bigger rewards to encourage certain behavior (quick turnaround on -bigadv) has benefited science through obtaining desired results - more -bigadv units returned faster. Yes, I completely understand that competition can bring about the best results sometimes, and I'm fully engaged in the competition. With that said, inflation is bad now and will quickly become a monster. We can preserve the positive emotional experience of greater reward for higher production without risking losing the mid-range and modest donors, but as has been said, Moore's Law has introduced an inflation variable, that I think, no one had really pondered when devising QRB.

As I wrote above, I don't have a magic solution for us, but I believe a good start would be to significantly flatten that green line on the graph, or soon, we'll lose a large number of Folders, who feel that their contributions don't count for much anymore. If the goal is to rely on the top 5% of Folders and gradually dismiss the rest of the donors, just continue with the present bonus system.

[7im]

You have people almost burning up X6 systems chasing the -bigadv carrot. Not good for the X6, even worse for their power bill to over clock so much (diminishing returns, with exponential power increases), and they're slowing down the results. (NOT picking on AMD here, so don't even start, i7s are almost the same thing)

Tim, I've been running my x6 for almost a year and haven't burned it up and neither have the ones that I know (and you know where I hang at) have burned theirs up either.

If we want to stop the x6's, the "almost the same thing i7's" and the slow 2ps from folding bigadv/bigbeta to help speed the science up then decrease the base points and shorten the preferred and final deadlines. By doing that it makes it not worth their time to fold those types of WU's. Who would want to fold for a couple of days and only get a few hundred points when they could of gotten several K on a SMP...maybe you, no me.

It's an anechdotal example. I did say almost!!! burning them up which is very true.

Yes, I do know where you hang, but I don't think people would typically run so hot if they were not chasing that carrot. And I've read about the few that choose not to chase over there also.

And yes, I have recommended they shorten the deadlines to cut off the non 8 core systems.

Finally, I'll refer you back to the folding best practices about running -bigadv on 6 core systems, and let everyone else be your jury.