Unbalanced Scoring

[mdk777]

FordGT90Concept,

Thanks for the reply.

As you can see in my reply to Guru, change happens.

Put a $100 NV 9600 GSO card on a couple of those machines and you will be cranking out 8000 ppd and greatly reducing your electric bill.

Speeding up the research is really the most exciting aspect of this project. 10 years ago, it was impossible, 5 years ago and it was an proof of concept.
Today, if everyone of those 200,000 cpu donors could and would run the GPU client, then 20 PLOPS of computing power could be applied to research.
This is more computing power than the top 10 supercomputers in the world today(combined).

While it is not exactly the same because, supercomputers do have the advantage of very high speed interconnects (see Dr. Pande's post on the advantage of the SMP client):
It would take results from decades to months!!

Really, instead of considering leaving, you should be excited about the prospect of 100x improvement in efficiency! This is only possible in the world of computers!
If I could invest $100 and get 100x the gas mileage in my car, I certainly would;
wouldn't you?

[FordGT90Concept]

Then what?

BOINC may be the same thing, yeah. If it is, I'll probably quit donating my spare clocks altogether. I'll probably rededicate the server to compiling maps again and call that the end of my science days. XD

Put a $100 NV 9600 GSO card on a couple of those machines and you will be cranking out 8000 ppd and greatly reducing your electric bill.

I tried folding on my 8800 GT. Mass Effect and Worms 4: Mayhem basically became unplayable when it was folding. Level transisitions were painstakingly slow. Even IE7 was getting all kinds of choppy on sites with lots of pictures/Flash animations. Ultimately, I got rid of it the day after I started it because the performance deficit is just too great.

If I could invest $100 and get 100x the gas mileage in my car, I certainly would;
wouldn't you?

If Stanford foot the bill, sure. I have no reason to upgrade computers that have sufficient graphic capacity just to fold. Doing so is a waste of money in my eye. Ultimately, that's where I disagree with a lot of people on this forum. I fold because I have a lot of spare clocks that could be put to better use. Most here fold to fold. The current score system works for those that fold to fold but not for those that just want to donate spare clocks.

[mdk777]

I tried folding on my 8800 GT. Mass Effect and Worms 4: Mayhem basically became unplayable when it was folding. Level transisitions were painstakingly slow. Even IE7 was getting all kinds of choppy on sites with lots of pictures/Flash animations. Ultimately, I got rid of it the day after I started it because the performance deficit is just too great.

I hate to point out the obvious, but you say you're willing to only donate your spare clocks.

Run the client when you are not actively using the computer and you will be making a great donation. One can only play games so many hours per day.

Also, while I don't have a NV card, I do think they have made some improvements over the initial client.

Good luck to you either way.

[FordGT90Concept]

I hate to point out the obvious, but you say you're willing to only donate your spare clocks.

Run the client when you are not actively using the computer and you will be making a great donation. One can only play games so many hours per day.

Also, while I don't have a NV card, I do think they have made some improvements over the initial client.

Good luck to you either way.

1) The software is using clocks that aren't "spare."
2) I always install as a service which means I have to kill the process before I start the game or wait 5 minutes for Task Manager to come up before I can terminate it and after I am done gaming, I have to either restart or do a manual start of the service via the service controller. Both of which are a PITA.
3) I game a lot. If I'm not gaming, I'm doing other things that involve a lot of vertical scrolling (like programming). I've dealt with programming with horrible vertical scrolling before and it simply gets untolerable.
4) I installed the 6.22 (I think) GPU client.

[JBurton57]

Ultimately, that's where I disagree with a lot of people on this forum. I fold because I have a lot of spare clocks that could be put to better use. Most here fold to fold. The current score system works for those that fold to fold but not for those that just want to donate spare clocks.

I don't think that's exactly true. Yeah, a lot of people here run computers that do nothing but fold. But that's not the primary problem here. The fundamental disagreement is that you seem to think the final arbiter of a person's worth to FAH is the time they're willing to put into the program. Most of the people here couldn't care less about time. We have two ways of approaching scoring:

1) More FLOPs = more science = more points.
2) More people = more FLOPs = more science = more points.

The thing is, most of the people here question whether or not more people means more FLOPs, or even why that step is necessary. FLOPs is the easiest metric to measure, and the one most correlated with the amount of processing (i.e., work) being done.

In the end, it doesn't matter what the scoring system is; it'll benefit people who fold for the sake of folding because they'll do whatever it is that maximizes points. So if we switched to a (time x computers) paradigm, they'd be digging up old P2 and P3 systems that hold FAH back because they take so long to crunch each WU. That wouldn't help anybody, because time is a resource. If I crunch my WU faster, that means the next guy gets the WU faster, which means the full trajectory gets done faster. In this light, it makes the most sense to align points with crunching out WUs the fastest, because people who fold for the competition are going to do whatever it takes to win (even if that means less science). People donating spare clocks have much less incentive to point-chase, simply because they're not building folding farms. What good do the points do them, anyway? At the level where points don't buy you bragging rights, it makes no sense to compare your points to somebody else's.

I can definitely agree that it's disheartening to see that GPUs crush all other clients right now, but that's just the name of the game. From everything I've read or heard, the GPUs really are just that much more powerful. They're the way of the future for supercomputing in general. Eventually I think the points will go down somewhat. The beta clients tend to score high points in general, due to their additional annoyance factor (i.e., sometimes taking up more resources than they should, slowing down other processes, crashing, etc.). But this isn't the first or last time it'll happen to FAH. As others have said, I remember when the PS3 client came out. Nobody wanted to have a CPU any more. Everybody wanted to throw away their desktops and buy a PS3 to run Linux. And when the SMP client came out, single-core processors were *so yesterday*. Now GPUs are crunching, and we're not even impressed by the PS3. Tomorrow Larrabee will come out, and we'll tear out our hair wondering why we bought these stupid, useless 9800GTXs.

The more things change, the more they stay the same.

[mdk777]

3) I game a lot. If I'm not gaming, I'm doing other things that involve a lot of vertical scrolling (like programming). I've dealt with programming with horrible vertical scrolling before and it simply gets untolerable.

Now you're just being contrary out of reflex.
I too spend a great deal of time on my computer, but you have to sleep don't you? get out of the house every now and then?
If you have so little computer time to spare,

WHY DO YOU CARE ABOUT THE POINTS?

Come on now, I tried to positive, but if you don't want to contribute, that certainly is your choice.
Again, best of luck to you either way.

[Guru]

It's too bad that people would rather waste more storage on arguing about the necessary solution instead of implementing a solution.

Yes, it *is* too bad you have wasted our time. You have tried to prove that you knew better than the professors and students of one of the top Universities in the U.S., plus the combined efforts of thousands of donors who are computer engineers, yet cannot accept that you have failed.

It's funny that you say that. I have two degrees from two of the top Universities in the U.S. and I graduated with honors. By your logic, that gives me just as much authority to speak on the matter, (and it does by my own as well...)

Some of us are here because we have family members who are dying, or have died, of incurable diseases.

Then stop trying to hinder progress by insisting that the software and the way it is scored can not be improved. Are you really so ignorant as to suggest that improvement is not possible and that things should just stay the way they are?

Your attitude of entitlement is insulting to everyone involved. Unless you would like to visit your local hospital's cancer ward and explain to everyone that their cure will have to wait a little bit longer, so you can change the system to suit your own personal needs, then leave this project now. By your own reasoning, we'll instantly replace you with a GPU anyway.

LOL I'm not the one saying that we should pull the plug on everything that isn't getting high numbers! You are, along with many others!

Don't let the door hit you on the ass on the way out.

I'll hold it open for you on your exit.

[Guru]

Example solution

Device 1
CPU = XXXX
GPU = XXXX
Score = XXX
Time Contributed = XXX
WU's Completed = XXX
Average WU Size = XX GB
Minimum WU Size = XX GB
Maximum WU Size = XX GB

Device 2

etc....

What's wrong with that solution? Why not have a fair comparison?


You've got a PM. You need a lot more detail than that. Unit labels? Is that score PPD, PP$, PPwatt. Is device 1 equivalent to one client, or one computer with one/multiple clients? You *might* have a rudimentary beginning there, but you need to flesh it out A LOT more, and then show sample data from a few systems so people can better understand. X's don't mean anything. Spell it out so an idiot like me can understand it. That's what's wrong with it. -7im

[FordGT90Concept]

I don't think that's exactly true. Yeah, a lot of people here run computers that do nothing but fold. But that's not the primary problem here. The fundamental disagreement is that you seem to think the final arbiter of a person's worth to FAH is the time they're willing to put into the program. Most of the people here couldn't care less about time. We have two ways of approaching scoring:

1) More FLOPs = more science = more points.
2) More people = more FLOPs = more science = more points.

The thing is, most of the people here question whether or not more people means more FLOPs, or even why that step is necessary. FLOPs is the easiest metric to measure, and the one most correlated with the amount of processing (i.e., work) being done.

...

There is a way to satisfy both parties here. Donations are always based on time or money and both are lacking from the existing system. It shouldn't be too hard to implement a time system and it will satisfy the slow and steady group. That's all I'm really saying. If my time is not valued by the Pande Group, I'll go somewhere else.

I too spend a great deal of time on my computer, but you have to sleep don't you? get out of the house every now and then?

I turn the computer off at night that has the 8800 GT. I rarely leave the house being self-employed and all...

[mdk777]

FordGT90Concept,

One last try,
I don't speak for the Pande group, but

It shouldn't be too hard to implement a time system and it will satisfy the slow and steady group.

The entire reason d'etre of the project is to do what "slow and steady" science cannot.

A million ancient philosophers working for a million years with their abacuses, are not going to get the job done.

Therefore, by definition, this project does not want to encourage "slow and steady"

The points are expressly designed to reward the fastest;allowing the most science to be done in the least amount of time.

There is no great good in encouraging people to run their 386 machines, (not one, not a million),they are just too slow!

This might go against your anti-elitist feeling, but at some point you have to realize that the amount of benefit you are generating is not really worth the electricity.

If you are running a 386, I urge you to shut it OFF NOW!

[FordGT90Concept]

There is no great good in encouraging people to run their 386 machines, (not one, not a million),they are just too slow!

So is something not better than nothing?

This might go against your anti-elitist feeling, but at some point you have to realize that the amount of benefit you are generating is not really worth the electricity.

That's for me to decide, not you or anyone else.

If you are running a 386, I urge you to shut it OFF NOW!

The oldest computer I keep operational is an Intel Pentium 2 (266 MHz I believe). Would I plug it in to fold, no, as it is not worth the electricity and heat generated. I only keep it to play DOS/Windows 95 games like Earth 2140.

[mdk777]

So is something not better than nothing?

NO,NO,NO!!!
Time is what this project is all about!
If all you can donate is the widows mite, then god bless you.
But that is not going to get the WU done in a timely fashion.

That's for me to decide, not you or anyone else.

And THAT is EXACTLY why the point system is working as intended!!!

It allows you to make that informed decision!!

[JBurton57]

There is no great good in encouraging people to run their 386 machines, (not one, not a million),they are just too slow!

So is something not better than nothing?

In a sense this is true. Pandegroup has said in the past that if a client computer cannot reliably make the "preferred" (first) deadline, then it shouldn't be used for folding. Computers that take too long to fold hold up the next person in line. Each successive WU from the run folds the protein a little more. So in a sense this is a balancing act whereby we do want the most FLOPs possible, but the speed of those FLOPs is still important.

In a project like Einstein@Home it's a little different, because there it's mostly about fast fourier transforms on a huge dataset. If we're missing part of the dataset it's okay because the project is so massively parallel. You can do the next part while you're waiting for the old one. That little bit isn't (generally) holding anybody up. But for FAH a lot of the parallel processing happens because each protein needs to be replicated so many times.

So yes, a 386 that can't fold its WU fast enough (and causes the assignment server to send the WU out only to assume it's been lost) is damaging the project because somewhere out there is an AMD Phenom ready to crunch it out. FAH must balance wanting to include all the computers it can with wanting to get back the WUs in a timely manner to re-construct the data and assign the new WU. In general, however, I don't think it's an issue for most people. I believe that an old Athlon 1800+ can still complete standard CPU client WUs on time. A Pentium 2 is almost surely considered too slow to work on the project. I don't know about Pentium 3s.

[Clintonio]

I just today re-discovered F@H. Last year I was on my AMD 6000+ Dual Core (3.0Ghz) and I could barely make a single project finish because of my PCs online time. And that was a 500 one.

This year I purchased two 9600GTs on a whim (What can I say, I had £1000 laying around for my new £800 laptop, so I spent the excess), and today I am folding at what feels like 100 times the speed.

Now, do I care about points? No. To me they're just a little reminder of how much I have done, not a way to beat everyone else or get the most points. I understand that many people are competitive by nature, but I prefer getting things done, being efficient and just helping.
Right now I'm playing a video game on that laptop I purchased, which itself has a CUDA enabled 9500GT that I could use (But I won't, my laptop needs to have a good lifespan), while this PC I am typing on is chugging away at F@H. For the sake of feeling like I've not got an idle PC, and for convinience, I also have my 5 chat clients and 20 Opera tabs open on here due to their low footprint

[FordGT90Concept]

In a sense this is true. Pandegroup has said in the past that if a client computer cannot reliably make the "preferred" (first) deadline, then it shouldn't be used for folding. Computers that take too long to fold hold up the next person in line. Each successive WU from the run folds the protein a little more. So in a sense this is a balancing act whereby we do want the most FLOPs possible, but the speed of those FLOPs is still important.

In a project like Einstein@Home it's a little different, because there it's mostly about fast fourier transforms on a huge dataset. If we're missing part of the dataset it's okay because the project is so massively parallel. You can do the next part while you're waiting for the old one. That little bit isn't (generally) holding anybody up. But for FAH a lot of the parallel processing happens because each protein needs to be replicated so many times.

So yes, a 386 that can't fold its WU fast enough (and causes the assignment server to send the WU out only to assume it's been lost) is damaging the project because somewhere out there is an AMD Phenom ready to crunch it out. FAH must balance wanting to include all the computers it can with wanting to get back the WUs in a timely manner to re-construct the data and assign the new WU. In general, however, I don't think it's an issue for most people. I believe that an old Athlon 1800+ can still complete standard CPU client WUs on time. A Pentium 2 is almost surely considered too slow to work on the project. I don't know about Pentium 3s.

You're basically telling me there is a flaw in the system then. WUs should be prioritized and high priority WUs should end up on the fastest hardware available to calculate it. The Pentium IIs should only receive low priority WUs. As a contributor, that really isn't my problem. If I make my Pentium II available for folding, Pande Group has to decide how to best utilize it, if at all. The software/service itself should notify the contirbutor after X number of days of not being assigned work that it is better to just no longer contribute via that hardware. Again, I just make the time available, I don't have anything to do with how it is used.

Now, do I care about points? No. To me they're just a little reminder of how much I have done, not a way to beat everyone else or get the most points.

You only gets points if a) it was folding and b) the work was turned in.