Folding Forum

And I never said "top teams" actually followed the documents. I only said they flaunted them.

You, however, assumed they did follow the document so that a "top team would equal donors", so you're more guilty than I am.

??

However, I will agree with others that this thread has become pointless a long time ago.

mdk777 wrote:

And I never said "top teams" actually followed the documents. I only said they flaunted them.

You, however, assumed they did follow the document so that a "top team would equal donors", so you're more guilty than I am.

??

However, I will agree with others that this thread has become pointless a long time ago.

I disagree there have been some good points made from all sides of the fence mixed in with the useless things (some of which are mine (Useless) but it has also allowed some of us to do some venting (Myself included). Many folders are passionate about there folding for one reason or another whether it be e-peen or project related it really does not matter. As long as the science is getting done and we are not slowing down the progress of the science. Whatever Stanford decides to do it is all right with me I am sure they will do what is best for the science.

I did go do a little fishing and crabbing last weekend and when I came back 2 of my rigs were down. I had to close the windows when I was gone so the computer room got pretty hot. Even when I was angry about the way things have been happening and the lack of clear guidance from Stanford it still bothered me that 2 of my rigs had been down for a couple of days and I was slowing down the science. So what did I do I ordered 2 more pumps 2 more radiators and CPU blocks plus all the accessories to liquid cool those 2 rigs.

If I had not been allowed to vent my frustration I am not sure I would have done that.

I was talking to some people at work today about folding and what it was and what it was all about and one of the people ask me why I did it. He said the amount of money I had invested and the amount of money I spent on electricity seemed a little foolish to him. I looked at him and asked if he had ever helped anybody out or donated anything that had made him feel really good about himself. His reply was yes. I then asked if the amount it cost him made any difference or what he received if anything mattered. His reply was no it did not the feeling he got far outweighed anything else. My reply to him was you have got your answer to the question, I feel really good about the fact that I am doing something that may help somebody have a better life some day. And no matter how confused I get or how frustrated I may get that will never go away.

So in conclusion I wish to say THAHK YOU VJ, KASSON and ALL THE OTHERS at STANFORD for letting me be part of this great endeavour. I may rant and kick and scream from time to time but that is just the (H) factor, also thank you to those on the forum that allowed me to kick at them and get it out of my system. I will fold as long as I can and may we continue to make advances in science.

Grandpa

Nice post, Grandpa.

Those who wish to critique one another, please find a private room somewhere and let this thread return to topic.

sortofageek wrote:Nice post, Grandpa.

Those who wish to critique one another, please find a private room somewhere and let this thread return to topic.

I concur. This topic has turned into people saying something and then everybody else arguing about whether the words used were right or wrong. Sure, everybody occasionally says words that are subject to interpretation (or misinterpretation) but quibbling over what words were used and whether using some different words would be a better way to say something has nothing whatsoever to do with the original topic. If you guys waste another page with that sort of trivia, I'm going to lock the thread.

The purpose of this forum is to help folks who are having problems with FAH. Many of the most recent posts in this topic not only contain almost zero on-topic information, none of it can be called particularly helpful. If you've posted anywhere in this topic more than twice, read over your posts and I'll bet that your first two posts really said everything that you needed to say and the other N-2 are pretty much unnecessary as are (almost) any future posts that you might make in the remainder of this topic.

(That's not anything scientific. Sure, there are probably exceptions. I didn't actually count anybody's posts and analyze them. I leave that to each of you, but I'm really feeling like it's time to draw this discussion to a close.)

Grandpa_01 wrote: A 2600K at 4.4Ghz takes 53 min. to do a frame of a 6903 guess what allot of the 2P 12 core server boards will not do that and if the drop the deadline which is the simple fix they are gong to have some 12 core machines incapable of folding them that are going to be assigned them.

If I might return, for the nonce, to the subject of the thread.

The interesting quote above gets to the heart of the issue, for me at least.

PG and its associates put the projects in place. Each project has its design - how the work units are going to be parceled out to runs/clones/gens. How "parallel" the WUs are going to be, and how quickly they are seeking answers to support conferences, academic schedules and publications. And, at the end of the day, getting the science to point to understanding how to manage or cure the terrible diseases related to protein folding (and misfolding).

So - why should it matter what kind of hardware is applied to the problem? If the issue is speed of return because of overall project schedules and serialization of individual projects, then greatly shorten the deadlines to reflect what the real project turn around needs to be. Maybe adjust points and k-factors accordingly. Reward those who bring the biggest iron to the parties that need fast return.

And, do it regardless of what kind of hardware, how many cores. If the deadlines accurately reflect the desired behavior to support project schedules - then those who try to fold on inadequate hardware will miss those deadlines. And the enforcement mechanism is that those who sign onto that category of WUs who can't meet the deadlines will be penalized by losing their QRB bonuses if they drop below the 80% success rate.

As grandpa's quote says, if older 12-core servers can't make the deadlines as well as souped up i7 2600ks, or whatever will come out in the next quarter or the quarter after that - then shouldn't the timelines decide?

So - some of the changes that might be needed:

a. More categories. We have adv, bigadv, bigbeta. Do there need to be more shredouts?

b. Assignment servers don't look for hardware characteristics [this is probably the weak part of the argument. ]

c. There has been discussion of the servers keeping stats for the individual user systems - what the average TPF by project for that user/passkey/slot by project. That might be the objective filter - whether a souped up 6 core AMD, or a 5 GHz 2600k, or a dual Xeon 16 core system. Can they meet the minimum TPF for the project? [this won't happen overnight]

d. enforcement of losing the QRB privileges for any cherry picking or dumping of WUs - or running on systems that can't make the minimum TPF. [this may be in fact happening, but I can't see how dumping assigned WUs doesn't count against your 80% success rate.]

Will this help motivate the folding community any more? For me, at least, it seems objective. If my system can meet the TPF deadlines that the project has set, then it can fold. If it can't, then there are lots of other opportunities out there. It should still incentivize bringing the big iron to the party - if a 6904 can be completed in under a day, it should be appropriately rewarded.

Not a complete, or maybe even a practical, solution, but my two cents to try to get the conversation back on some sort of track late on a Tuesday night.

And again Grandpa's wisdom shines forth 3+

I have to agree with pretty much all of that GreyWhiskers thank you

We should have a small incentive (10%?) to move up a category of work unit, smp 10% more than uni,
bigadv 10% more than smp and big bigadv 10% more etc etc

Frame time should be the only thing that gets you into the next class up, core/thread counting is going to
become less relevant and should be phased out.
Moving to frame time also removes any incentive to manipulate the system,it also takes into acount
different cpu architecture as that moves fwd /changes.(cores/threads/shared fpu's and apu's etc)

As GreyWhiskers said, set the deadlines so you get the amount of folders on a project that are
required and regularly review(3/ 6/12 months) the deadlines so as processing power increases you maintain
the balance.(But please let us know this is happening)

To do this you would have to bring in another flag so people can opt in/out of a type if they cannot make
the deadlines (-bigbigadv)

This would also increase the speed of the other work unit types as fast rigs are moved to them.

The new 6903/4 points are so far out of whack with points/deadlines its silly ,i get a 70% increase in ppd
folding these over 6900/1's and 100% over 2684's.

The QRB needs to be extended to all classes of work unit and points for the GPU need to be set at a level
that represents the usefulness of the data to the project relative to the other types, if this means GPU points
climb dramatically so be it.

I have always optimised my purchases to make the most points because as far as i see this is the only metric
we have for the science we are doing.

For ref i mainly fold bigadv and yes i have an overclocked 48 core rig running Linux.


+1 for Grandpa putting this all in perspective

I'd prefer to eliminate the QRB than to apply it, as it is currently calculated, to all clients.

The exponential QRB will always make the value of future work, on ever faster hardware, significantly greater than present work. On my team, we're seeing how this affects folding hardware purchases in that a number of donors, myself included, are postponing their purchases, awaiting the next generation, because the QRB is so rich.

Ok i can understand the need to re-datum the curve(benchmarking) but not the
nature of the curve if it represents the benefit of time to PG.

VijayPande wrote:
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.

Instead of spending money on a dedicated benchmark rig could a benchmarking
work unit not be sent out every 3/6/12 months in each class to work out the
lowest/average/peak performance and then re-datum according to the results?

The bigest problem i see is that PC life cycles are getting longer for the average user,
most people i know could run a fast P4 single core for what they use it for.

Now when your average home user is maybe running a dual core rig and will be for
three or more years, how do we encorage participation at this level to foster the
next generation of folders?

ChasR wrote:I'd prefer to eliminate the QRB than to apply it, as it is currently calculated, to all clients.

The exponential QRB will always make the value of future work, on ever faster hardware, significantly greater than present work. On my team, we're seeing how this affects folding hardware purchases in that a number of donors, myself included, are postponing their purchases, awaiting the next generation, because the QRB is so rich.

This is precisely why I think the base points should be adjusted periodically. One thing that is not taken into account in the current system is that there is more "value" to a WU completed a year ago than the same WU completed today. Since it's not currently accounted for, it causes two things:
1. Unfair devaluation of prior contributions
2. Postponement of folding hardware purchases.

The QRB exacerbates the problem.

There certainly is precedent for time-based value, as in the time value of money (PV vs FV etc). I'm sure the same arguments could be made for the research data - if enough people postpone their upgrades, enough work doesn't get done, project completions are delayed.

I could see a few ways to do this. One would be to pick a system of a certain dollar amount, or the greatest desktop market share, periodically, and benchmark the same project with it, then adjust points so that the PPD remains constant with the previous system. Another would be to use published benchmark data to estimate the increase in computational speed year-over-year and adjust the points annually.

The idea is to make it "equally difficult" to generate the same amount of points from year to year. Right now if someone wants to reach the top 100 folders, they might see that the best thing to do is wait 2 years, saving all their money that would otherwise go to power, cooling and hardware, then splurge on a huge system that easily could reach top 100 in a year. Personally I am much prouder of my 20 million points done mostly with P4 and dual-core processors than with my 90 million done with dual Westmere systems; those 20 million took much longer and much more money to complete. Shouldn't there be a way to better reconcile past and future contributions?

ChasR wrote:I'd prefer to eliminate the QRB than to apply it, as it is currently calculated, to all clients.

The exponential QRB will always make the value of future work, on ever faster hardware, significantly greater than present work. On my team, we're seeing how this affects folding hardware purchases in that a number of donors, myself included, are postponing their purchases, awaiting the next generation, because the QRB is so rich.

I do believe that is the purpose of the QRB, (to give us a guide as to what our next purchase should be) I believe it is working. Each of us have to decide on our own what we are going to do. You will have next to nobody building and folding on top end rigs if there is no insentive to do so. Everything has to move forward I know that my next purchase will most likley be a multi socket rig from what I am currentley seeing. If it were not for the QRB I most likley would not do this (I would have no reason to). As far as the exponential curve goes that is another problem. I still believe that the best way to deal with it is to manually adjust it when it needs to be adjusted. There needs to be some kind of a guide line set up for manual adjustment for when points excead scientific value.

Sorry Grandpa, I have to disagree, at least to an extent. If you build a rig twice as fast you get 2x the ppd. That has proven to be enough incentive for donors to build and buy the fastest rigs they could afford in the past. While it will take more incentive to get people to move to radically powerful multisocket rigs, I don't think it takes a 10x bonus yielding 100x the ppd. I actually do see the need for a bonus. I just favor a smaller, linear, capped bonus. Fix the QRB and the balance between clients can be maintained for a longer time period.

BaBa wrote:Ok i can understand the need to re-datum the curve(benchmarking) but not the
nature of the curve if it represents the benefit of time to PG.

VijayPande wrote:
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.

Actually, yes, the nature of the curve does need to be reviewed as well. 2 years ago when they made this estimate i.e. best guess, they didn't have 24, 48 or 64 core systems.

They have new sets of performance data, and should revisit thier "plan" to make sure it is still "reasonable" because many people have questioned that continued reasonableness. If the curve is still reasonable, after they review it, then fine, keep the current curve if that's the best estimate, and they can confirm that estimate. But we all know technology develops quickly, and if they don't reconfirm as new data comes in, then eventually the curve becomes less than optimal.

At this point, all I ask is a confirmation the old curve is still the best curve dispite indications to the contrary.

Deadlines adjusted for 6903 and 6904 'Big Bigadv' work units.

http://foldingforum.org/viewtopic.php?f=24&t=19180.

ChasR wrote:Sorry Grandpa, I have to disagree, at least to an extent. If you build a rig twice as fast you get 2x the ppd. That has proven to be enough incentive for donors to build and buy the fastest rigs they could afford in the past. While it will take more incentive to get people to move to radically powerful multisocket rigs, I don't think it takes a 10x bonus yielding 100x the ppd. I actually do see the need for a bonus. I just favor a smaller, linear, capped bonus. Fix the QRB and the balance between clients can be maintained for a longer time period.

I think we pretty much agree. That is why I said manually adjusted, as far as the ammount of point diffrence should be determined by Stanford. What if they had a really important protein to fold and needed to do it super quick and needed people to upgrade to get it done. I can tell you the bigger the carrot the faster you are going to get what you want / need.