Folding at home points: What should be done?

FreeStream · 5 Oct 2006 at 22:24

As im sitting here, my GPU is doing 550ppd and I have two WUs on the C2d doing 500PPD each.

However, most of the time its doing 200PPD, which is hardly anything above my X2 which is 50-75% slower.

Same with my GPU, its 50-60-70 times more powerful than any CPU yet we will get hardly any points in comparison.

Rich had a good idea about saying 1 point = 1 GFlop.

I agree with that, becuase they have stats for how many flops each client is outputting, so use it for points.

Saying that people without GPUs will be left out, well tough. Technology has moved on. Some bloke with a 2ghz Celeron going to compain because some guy is getting 16x more points from a 4ghz Kentsfeild.

Points isn't the main point of running F@H, but there is competition to it.

Points would be using Flops something along like this:
1GigaFlops is 1 BILLION floating operations per second.

Lets say a X1900XTX @ 700/800 is 100 GFlops, 100 billion ops per second.
Over a minute thats 6000 Billion Operations. 360,000 Billion over 1 hour. If one GFlop = 1 point, then a X1900XTX would be getting 360 points an hour. 8640000 Billion ops a day, 8640 points a day. Seesh LOL

Now, a 2.8ghz Core 2 Duo. I think its about 2GFlops, 2 billion ops per second. 120 Billion over 1 minute, or 7200 Billion per hour. So thats 7.2 points an hour.
172800 Billion ops a day. 172.8 points a day. Per core.

Lastly, a 2Ghz P4/867Mhz G4 is about .5GFlops. 500 million ops per second. 30 Billion ops per min, 1800 Flops an hour, 43200 Billion ops a day, or 43.2 points.

In my mind, thats fair. Because the power its using = points.

What do you think?

rich99million · 5 Oct 2006 at 22:26

Concorde Rules said:
Rich had a good idea about saying 1 point = 1 GFlop.

Stop quoting me on that I wasn't being serious

FreeStream · 5 Oct 2006 at 22:27

rich99million said:
Stop quoting me on that I wasn't being serious

No dude, its a good idea, you get credit for it

VeNT · 5 Oct 2006 at 22:36

it should be based on time spent
as people with slower PCs that spend more time on the project are contributing more effort (and leccy bills) to the project.

FreeStream · 5 Oct 2006 at 22:43

VeNT said:
it should be based on time spent
as people with slower PCs that spend more time on the project are contributing more effort (and leccy bills) to the project.

But they arn't doing anywhere near the same amount of work. To do the same amount of work they have to use more power.

You cant allow for eqaul points between new and old tech, because its just not fair. Because if we have a 1k PC going spare 90% of the time = a 100 quid PC then how we going to feel?

EDIT: I can buy about 20 P3 Pcs for what I paid for my C2D, yet the C2D will output them all. So people will buy cheap slow PCs to get more points..

shadowscotland · 5 Oct 2006 at 23:18

I know it's slightly off topic but still relavant to the debate.

I was compairing my 805's out put to billy's or was it stan's c2d on a points per pound basis on the same WU - a 600 pointer if not mistaken. Any way the c2d was getting 600ppd per core to my 290ppd per core aka twice as fast. Yet my cpu cost a three times less so my cpu was better per £ spent for folding (well on that wu anyway).

Anyway my point is newest tech not nessacarly best bank for buck, folder.

Back on topic again - arn't the points alocations based on how long the test rig take to fold that WU? If your tech does it faster, you get more ppd. If slower, less ppd. That is surely both simple and fair.

Berserker · 5 Oct 2006 at 23:23

Concorde Rules said:
EDIT: I can buy about 20 P3 Pcs for what I paid for my C2D, yet the C2D will output them all. So people will buy cheap slow PCs to get more points..

But it'll cost you a fortune in electricity to run them all, so that point doesn't really stand.

I would suggest you check your sums as there's an order of magnitude error in your X1900 calculations.

Then, you might understand why it isn't so far. I think it should be somewhere between the two views already expressed. This is pretty much what Stanford do with their 'standard benchmark' PC. Faster gets you more PPD, slower less. However, they took into consideration the huge difference the GPU made.

My pockets may be comparitively pretty deep, but most of that is going on servicing debt from past excesses (including hardware to run DC clients I may add). There has to be a point at which the spending stops. I've reached that point.

Anyway, points will get adjusted if need be. It's happened before.

Mr. Brightside · 5 Oct 2006 at 23:35

Berserker said:
Then, you might understand why it isn't so far. I think it should be somewhere between the two views already expressed.

Agreed

I'm not going to be able to get any new hardware for a year or two at least due to lack of money (and my fastest current rig is a skt478 3.0E P4) and so it would be a shame if my output became shadowed by others because of this. I'm sure there are many other contributors out there in similar situations, who could potentially be lost/alienated if the points system was changed. That said, it is of course important to credit faster hardware to some extent and for this reason I feel the current system is probably just right.

null

oceaness · 6 Oct 2006 at 00:11

It's a difficult argument.

If you consider the 20 - 40x performance increase in using GPU's then work out 20 - 40x the points a CPU gets at the moment then it gets a bit ridiculous.

Consider a 300ppd WU on a CPU. A GPU equivalent, assuming lowest 20x performance boost, would = 6000ppd thus completely alienating everyone without a X1900 series because they wont be able to compete.

The reason Stanford have kept the multiplier down is so that the vast majority of people, 178204 active CPU's, can still compete with the 284 active GPU's which already = 1/8 of all the CPU's put together (22275.5 CPU's).

So 284 GPU's = 22275.5 CPU's :eek:

I think they should work it so GPU output should just about beat a good clocked Conroe that would be fair I think.

KE1HA · 6 Oct 2006 at 00:19

My two pence worth

Some people do it for the science, and care nothing for the points. Others are competition driven, with some focus on the science. Others are purely in it for the points & comepetitoin. No matter which way they descide to do the points calculations, sombody's going to draw a short straw.

With that being the case, If I were truely interested in the science and really fixing these problems < which I personally am > I would want to put as much fire power to the problem as I could. How to do that is difficult. We are all sitting here talking about top of the line hardware, of which only a very small portion of the people doing this can probably afford or would want to put the effort into reconfiguring every 3 to 6 months.

I think this is the reason they have driven down the path of using GPU's and the likes of the Cell processor, as it clearly has a cost per WU advantage (for the science) that far exceeds the likes of the quad / multi core processors.

Not being funny, but a 250 quid game card or 400 quid playstation is allot more palletable < for the masses > than a 1000 pound processor, yet can still keep up with and in some cases out produce the Grand Master Processors. So one could attract 2x customers with 4x the output for 1/3 the cost in some cases, at least from a science standpoint that is. Afterall, that's what they are after, progress against these degenerative deseases.

I dont think there's a straight forward answer, and everyone's points of view would be valid here.
.

SiriusB · 6 Oct 2006 at 00:40

KE1HA pretty much hit the nail on the head. There is no way Stanford could come up with a points system that would make everybody happy. As it is the current system is fine. GPU clients get a 4x bonus and that is said to increase in the future.

People seem to lose sight of the fact that Folding is for the science, not how fast you can stomp slackers! Yes I know the competition and the points is the incentive to fold, but a lot of people are grubbing for more and more points for their part in folding.

One argument I saw here the other night was along the lines of people using more and more energy running CPU and GPU clients. This is a prime example of people forgetting a major aspect of folding - it is 100% voluntary. Some people seem to think you HAVE to fold using the GPU and CPU client.

I think the current system of points is absolutely perfect given the complexity of it all. As has been said - faster systems get more points, slower systems get less. It is brilliant that people go to great lengths to get faster systems just for folding, but in the end 90% of people folding probably have just one machine that is likely to be of average speed. If Stanford made the GPU WUs get something silly like 6000PPD you would have a few hundred or few thousand people in a league of their own, nobody without an X1900 could compete.

SiriusB

FreeStream · 6 Oct 2006 at 08:37

SiriusB said:
KE1HA pretty much hit the nail on the head. There is no way Stanford could come up with a points system that would make everybody happy. As it is the current system is fine. GPU clients get a 4x bonus and that is said to increase in the future.

People seem to lose sight of the fact that Folding is for the science, not how fast you can stomp slackers! Yes I know the competition and the points is the incentive to fold, but a lot of people are grubbing for more and more points for their part in folding.

One argument I saw here the other night was along the lines of people using more and more energy running CPU and GPU clients. This is a prime example of people forgetting a major aspect of folding - it is 100% voluntary. Some people seem to think you HAVE to fold using the GPU and CPU client.

I think the current system of points is absolutely perfect given the complexity of it all. As has been said - faster systems get more points, slower systems get less. It is brilliant that people go to great lengths to get faster systems just for folding, but in the end 90% of people folding probably have just one machine that is likely to be of average speed. If Stanford made the GPU WUs get something silly like 6000PPD you would have a few hundred or few thousand people in a league of their own, nobody without an X1900 could compete.

SiriusB

True, but then why should GPU users be bogged down to a limit PPD because some people can't compete. again this situation of a 2Ghz P4 Vs. a 4Ghz Kentsfield, is that not fair?

It has to be a scaleable points system, new tech gets more points because its more powerful and in some cases uses less power.

Or we could abolish the points system all together and just see how many workunits someone can produce instead

Edit: Added a 0 to correct X1900XTX scores, well, I'd be happy with that LMFAO!

SiriusB · 6 Oct 2006 at 10:16

I understand the sentiment that those with newer and faster tech should get more points, but the fact is we do!

As has already been shown something like a Core 2 Duo all by itself can output far more points than say an entire farm of P3s!

I think it would be somewhat unfair if not only did you get faster output but even more points.

SiriusB

Berserker · 6 Oct 2006 at 10:18

I have no problem with faster output = more points, despite what it's doing to my team rank (can you say 'decimation'). But 20-30x more points? No, I don't think so.

shadowscotland · 6 Oct 2006 at 10:21

just a side issue (yes another one) any one considered what would happen if sanford ran out of Wu's?

Since C2D came out and now the new GPU client total GFlop output for all folders worldwide is up - significantly up

Every month more people will swap old hardware for new (or add new to existing farm

)

with kentfield & PS3 soon to arrive theirs another boost. And let's not forget what ever AMD have up their sleave's.

This will either give stanford more 'weight' when bidding for work (abet other DC programmes will also benifit) allow them to take on extra work, (if any is available), but will defernatly give scientist better turn-around times for their results. So the question is will science feed the demand for WU that the new tech has caused? Or will the supply (WU need for science) having neither positive or negative feedback to the demand (folders world wide) fast run out and therefor lose folders to other DC projects?

In essance I hope the marketing boys at stanford are watching the current upward trend and drumming up more 'business'

It's possable that there is already a huge backlog and theirfore no problem.
Or that the earlier returned results will directly feedback into more science being done.
Or that some projects were to big for CPU to handle in a normal folding time frame (under a week) so can now be run on the GPU client.
- yes I know some DC project tie up cpu's for months, but that's not normal here. (well it is if you have a 800 celeron but still

)

You never know the new tech with its excerlent ppd might convince members with old tech, (p2 even p3) and increased electric bill to stop folding as carbon emmisions from power used outweights science gain and therefore reduce demand.

Edit: to that end (as it's been bothering me for some time now) I'm removing the folding clients on machines that don't even acheive 100ppd on good WU - these are the machines that also have older PSU so even worse power wasteage.

VeNT · 6 Oct 2006 at 11:37

don't we already get more points for performance on WUs that use more ram than others?
people should get more points for useing more hardware, but not so much that its going to put a sque on the whole scores.
btw I'm gonna get a 1900 with my next system I hope!

sculptor · 6 Oct 2006 at 19:21

Personally I feel the points awarded for Gpu folding should be increased.
My reasons for this are as follows:

Folding using the Gpu is far less user friendly as it slows down the GUI.
Cpu folding atleast allows the donator to use thier pc whilst the client is running. A fast Cpu or a dual Cpu is easily capable of running a game at the same time, memory permitting.
So basically the donator is giving full resources to Stanford whilst the pc is folding.

Folding using the Gpu will work out to be more energy efficent as long as the points atleast remain the same.
If the points were increased (after all the X1900XT is a top end card) I think it would encourage donators who are interested in points to upgrade thier pcs.
If donators did upgrade thier pcs they would save money in the long run by having more energy efficent hardware.
Donators who are folding using older hardware are already at a dissadvantage compared to the X2s, Opterons and Conroes.
Many donators using older hardware are happy that they are contributing to Stanfords cause and may not necessarily be concerned with points.
I feel these donators would not be interested in upgrading thier pcs apart from the energy efficently angle.

I feel Stanford should consult with the top people of the top teams regarding a fair points system for Gpu folding.
It need not take long for this consultaion to take place as Im sure along with myself you would rather have the scientist doing science than sorting out a points system.
If each team set a date for all its members to raise thier point of view and then the top people from each team presented the results to Stanford.
A poll on the fairness of the points awarded to Gpu folding would probably be
a good place to start.

Personally I feel this is a great opportunity for Stanford to encourage more energy efficent folding.

Tom

FreeStream · 6 Oct 2006 at 19:31

sculptor said:
Personally I feel the points awarded for Gpu folding should be increased.
My reasons for this are as follows:

Folding using the Gpu is far less user friendly as it slows down the GUI.
Cpu folding atleast allows the donator to use thier pc whilst the client is running. A fast Cpu or a dual Cpu is easily capable of running a game at the same time, memory permitting.
So basically the donator is giving full resources to Stanford whilst the pc is folding.

Folding using the Gpu will work out to be more energy efficent as long as the points atleast remain the same.
If the points were increased (after all the X1900XT is a top end card) I think it would encourage donators who are interested in points to upgrade thier pcs.
If donators did upgrade thier pcs they would save money in the long run by having more energy efficent hardware.
Donators who are folding using older hardware are already at a dissadvantage compared to the X2s, Opterons and Conroes.
Many donators using older hardware are happy that they are contributing to Stanfords cause and may not necessarily be concerned with points.
I feel these donators would not be interested in upgrading thier pcs apart from the energy efficently angle.

I feel Stanford should consult with the top people of the top teams regarding a fair points system for Gpu folding.
It need not take long for this consultaion to take place as Im sure along with myself you would rather have the scientist doing science than sorting out a points system.
If each team set a date for all its members to raise thier point of view and then the top people from each team presented the results to Stanford.
A poll on the fairness of the points awarded to Gpu folding would probably be
a good place to start.

Personally I feel this is a great opportunity for Stanford to encourage more energy efficent folding.

Tom

Yep, but they can't really do that either because how do you calculate the power use of every PC out there?

Mines different to every other config like mine, voltages, amps, watts all different.

The only continuous thing between us all is the Flops rating the software uses. People will buy better hardware which folds more with less power.

So there we go:
Points system using Flops = less powerful hardware being replaced by powerful, more efficent PCs..

Edit: I agree about loosing your whole PC because the GPU client stops you from doing anything 3D and slows down the GUI. I actually don't mind because the 2D speeds are still ok.

Berserker · 6 Oct 2006 at 19:43

Concorde Rules said:
So there we go:
Points system using Flops = less powerful hardware being replaced by powerful, more efficent PCs..

And a select few giving everyone else the smack. If you got the points you calculated for Folding, why (as someone who is mostly interested in the points these days) even bother continuing? I might as well just give up as my points contribution would be utterly meaningless. Shall I just go and turn all my PCs off then? That's not going to help Stanford.

sculptor · 6 Oct 2006 at 20:29

Concorde Rules said:
Points system using Flops = less powerful hardware being replaced by powerful, more efficent PC..

Whist a points system based on Flops is a good idea, but I think it is too soon to implement.
As Berserker explained what would be the point of continuing folding, and Im certain there would be many others who share his point of view.

Stanford have created a client which is 20 - 40 times faster which also means it is 20 - 40 times more efficent. The problem lies with the fact that it uses the Gpu.

My fist thoughts about the Gpu client were that I could atleast half the number of pcs I am running (14 at present) by installing an X1000 series card in seven of them and running the Gpu client.
At the time I did not realise I would loose one of the two Cpu cores on each pc.
Now I think 7 pcs running the Cpu client on one core and running the Gpu client aswell would use far less electricty than 14 pcs.
As far as points is concerned I would probably be at about the same level.

Whilst it is easy for me to do this, someone who only has one pc and whishes to use the Gpu client will encounter all kinds of problems not associated with Cpu folding. For this reason I feel the points should be set at a higher rate.

If for any reason a fair points system cannot found the only answer I can see is to have different leagues for Gpu folding.