Folding Forum

Hey people,

Although not quite a noob, I am relatively new to FAH having only been folding since early spring.

I have been exploring the forum and am somewhat perplexed by some of the angst about points per project/platform.

Right now, points say put more nVidia GPU's on-line, run SMP even if you only have two cores, run mono cpu clients only as a last resort to squeeze the last 130ppd out of a rig already running a GPU client.

As for the PS3, at 900 ppd and its 180 watts power requirement, it is becoming increasing irrelevant for those who are running enough FAH to be concerned about paying the electric bill. Add the fact that the PS3 can't even browse the web while running FAH* and it is hard to justify even turning it on (understanding that it is a tool for getting additional machines/people involved in FAH).

*you actually can but try to load a large page like Kakao stats and it crashes

While discussions about the relative processing power of nVidia versus ATI are interesting in their own right, the hard adherence to benchmark-based point determination misses the fact that some science is better performed on different platforms--This is as assumption , as it has been stated here, but the points don't reflect it.

The question "Are you interested in ppd or science?" is (or should be) nonsensical as the points should reflect the science needs of the day and not some arbitrary GFLOP number or WU completion time. Otherwise, we might as well give out points for watts consumed and the cost of hardware to consume them as it would more accurately reflect the actual donations being made by members of the community.

I know that there are some seemingly arbitrary points bonuses for Beta clients and quick turn WU's and that is as it should be. Those point bonuses should probably be extended to the Beta WU's and cores as well.

I guess my main point is that the Pande group should not be bound by previous point determinations or benchmark hardware performance when it stops being the right thing to do and the community should encourage such flexibility.

Anyway, enough spouting off, it has been a pleasure reading many of your posts and using the forum to get my new multi GPU client and core affinity settings correct as well as stretching my Windows knowledge and getting my computer hardware knowledge more up to date.

Best regards,

IF you have a chance, go read this posts.

viewtopic.php?f=47&t=4812

We had some good ideas there, and Vijay is aware of the last thread and reviewing it with his team.

It is a very complex problem & it is actually being looked into.

Thanks EvilAlchemist for the link, that thread happened while I was out of town and I had missed it.

Excellent thread and shows some of the social engineering challenges that the Pande group has to deal with. It is clear that they feel pressured by the community to resolve perceived point imbalances and it is up to us to allow them the scope to make changes without staging a revolt.

later,

I would like to know if there is any two WUs that do the same scientific work (folding the same molecule, etc.), the first in mono CPU client and the second in GP2 client, so we can compare directly the speed difference related to clients.

I am still wondering why there is no measure of simulation time provided for by the stats.

Obviously there is a MASSIVE db there with all the user results in it. Now the average simulation time in ns is known for each unit (and possible more granular than that, obviously we know that some units mysteriously became more complex half way through), so these values could be assigned into the stats as a total ns simulation time donated.

Yes, I know that this is not a fair measure of the scientific value of the work, but could be helpful to see how adding new clients adds to the amount of simulation going on.

Mitsimonsta wrote:I am still wondering why there is no measure of simulation time provided for by the stats.

Obviously there is a MASSIVE db there with all the user results in it. Now the average simulation time in ns is known for each unit (and possible more granular than that, obviously we know that some units mysteriously became more complex half way through), so these values could be assigned into the stats as a total ns simulation time donated.

Yes, I know that this is not a fair measure of the scientific value of the work, but could be helpful to see how adding new clients adds to the amount of simulation going on.

Points are awarded (correctly, in my view) based on the scientific value of the task. It appears that many folders think the points should reflect the computational difficulty of the task. That seems like a reasonable viewpoint for folders because they provide the resources required to complete the tasks.

I don't see that the scientific value of a simulation is only related to the computational difficulty of performing it. Timing in relation to other simulations being run in parallel, velocity impact on simulations that depend on the current task etc, also have an impact on the "scientific value" of a task, but have little to do with computational difficulty. Dealing with those factors would greatly increase the complexity of, and therefore, arguments about, the point system

I don't envy Standford as they try to sort this out.

This thread again? We just went through this last week. And the week before that. Go to your local soup kitchen or rescue mission, and tell them they're doing it wrong, too.

Maybe Stanford could separate the points based on client? The whole idea of using a single metric here is like comparing apples to oranges.
I mean I currently have 1.2 million points which might seem a lot.
But it could be separated as:
900,000 SMP points, which is decent but nothing special compared to other long-time SMP folders.
270,000 GPU2 points, which is quite meager compared to other GPU2 folders.
30,000 CPU points, which is a joke compared to long-time CPU folders
0 GPU points..
0 PS3 points..

That way it would still be a challenge to catch a team-mate who's got 100,000 CPU points. You wouldn't be able to "cheat" by buying a 8800GS and letting it run for 3 weeks. No. You'd need a Core 2 Quad to run 4 CPU clients for half a year!

Edit: fixed the point values for various clients..

Foxery wrote:This thread again? We just went through this last week. And the week before that. Go to your local soup kitchen or rescue mission, and tell them they're doing it wrong, too.

I agree.

But on the other hand, I think I read somewhere the points are awarded as an extra stimulus to start/keep people folding.
Apparently, people are more inclined to participate if there's some kind of competition going on.
So if people start to get disappointed for not being able to compete against the newer/faster processors, the system of awarding points might become contraproductive.
It would be missing the point, so to speak.

Why are we arguing over points? I think there is a worthwhile discussion here for us to calmly put forward ideas to make the points system better for all concerned.

I do like the idea of splitting each contributor's points by client - it gives a much nicer picture of what you have actually achieved. Anyone with millions of standard client points should have everyone's respect as they did heaps of science to gain those sort of points. You could even break the standard client down into the various cores (Tinkers, Gromacs, Amber, QMD, Gro33, DGromacs etc).

I agree that it would be nice for the stats to break down a user's points into CPU, SMP, GPU, PS3 etc. However, the user's total points should remain the most important figure. My points score is the only way I can compare the work I've done to that of another member. A few of us in my team are doing a 'race to one million', and this kind of thing would be impossible if there were many different kinds of points. The current system may not always be fair but at least it's a common foundation that everybody can relate to.

Mitsimonsta wrote:Anyone with millions of standard client points should have everyone's respect as they did heaps of science to gain those sort of points.

On the other hand, the SMP client requires a much greater time commitment (systems have to be on most of the time, and the client quite often needs prodding.) IMO this justifies the points boost over the standard 'set it and forget it' client.

Mattus wrote:I agree that it would be nice for the stats to break down a user's points into CPU, SMP, GPU, PS3 etc. However, the user's total points should remain the most important figure. My points score is the only way I can compare the work I've done to that of another member. A few of us in my team are doing a 'race to one million', and this kind of thing would be impossible if there were many different kinds of points. The current system may not always be fair but at least it's a common foundation that everybody can relate to.

Mitsimonsta wrote:Anyone with millions of standard client points should have everyone's respect as they did heaps of science to gain those sort of points.

On the other hand, the SMP client requires a much greater time commitment (systems have to be on most of the time, and the client quite often needs prodding.) IMO this justifies the points boost over the standard 'set it and forget it' client.

Not quite so in the earlier days.
There were some tough times then, also, and making more than a few points per hour was huge.