Unbalanced Scoring

[FordGT90Concept]

I have the following processors folding:
-AMD Athlon 64 4000+ (1 x 2.4 GHz) San Diego
-Opteron 180 (2 x 2.4 GHz) Denmark
-2 x Intel Xeon 5310 (4 x 1.6 GHz) Clovertown

They add up to about 1000 points/day (11 processors). A fellow team member is folding on just an 8800 GTS 320 MiB and gets about 3000 points/day. In order to equal his one video card, I would need to run approximately 6 Intel Xeon 5482 processors (4 x 3.2 GHz, ~$1400 each) which adds up to upwards of 620 watt/hours in power consumption excluding banks of memory, motherboard, etc.

Yeah, GPUs are more suited for floating point operations but that doesn't mean CPU-only folders should be disowned as they are actually spending far more on their equipment and operational costs than the GPU folders.

This really needs to be addressed. I am inclined to take my spare clocks somewhere else.

[7im]

Hello FordGT90Concept, welcome to the folding forum.

The primary factor for assigning points is the scientific production of the client. I understand your concern about cost of the equipment, but if cost were the only concern, then only people willing to spend a lot of money would do well in the project, and that's not really a good incentive.

I think you would agree that the incentive should be to do the most science for the project to help find the cures. And the incentive in this example is the points awarded for completing work units.

I do know how you feel, as I still have several Pentium 4 systems folding at 110 PPD, while people with new hardware seem to be getting a lot more points. But that's not a problem with the point values, that more accurately reflects the technilogical advancements in processors over the last few years. They really are just that much faster at producing scientific results, so they do get a lot more points.

That's not what you wanted to hear, but I hope can now better understand the situation.

[Foxery]

I have the following processors folding:
-AMD Athlon 64 4000+ (1 x 2.4 GHz) San Diego
-Opteron 180 (2 x 2.4 GHz) Denmark
-2 x Intel Xeon 5310 (4 x 1.6 GHz) Clovertown

...

Yeah, GPUs are more suited for floating point operations but that doesn't mean CPU-only folders should be disowned as they are actually spending far more on their equipment and operational costs than the GPU folders.

"Disowned?" The point value of your CPUs has never been descreased; new technology really is that much faster.

[FordGT90Concept]

The problem isn't the technology, it is in very expensive CPUs becoming obsolete overnight compared to relatively inexpensive GPUs. Even a Harpertown processor, in terms of flops/sec, can't hold a light to a typical GPU but is it not the effort that counts? There's really no sense in me wasting hundreds of watt/hours folding when they really don't yield any benefits by comparison, is there?

The point system heavily favors NVIDIA GeForce 8, 9, and GTX 2 series GPUs over all other hardware available. If it takes over 30 CPUs to equal one upper-midrange 8800 series graphics card, there is really no sense in folding on just CPUs anymore.


I think the most fair way to calculate points is to base it on number of processors folding. Make it the effort that counts more so than the work accomplished. As said somewhere in the FAQs that the points are rewarded to "pay" for power consumed, per processor is the best way to stick to that mentality. It is easy to estimate that there is about 100 watts consumed per hour of a typical processor be it GPU or CPU. Why base it on performance at all? People by expensive computers to do tasks other than folding. You ought to be grateful to every single person that offers their spare clocks to you and reward them accordingly to do so. The point system, as it stands, has the wrong motives at its core. It is more based on capitalism than on the generosity of the people.

[7im]

The project suceeds through all types of donations, large and small. It's a team effort, not an individual effort. However, if your only criteria to participate in a project is to be able score in the top 100 folders, then this isn't the project for you. Folding@home didn't become the most powerful distributed computing project by sitting back and only developing one type of client for on type of processor. FAH is very progressive compared to any other project, and will go where the performance is available. Unfortunately, that means we either try to keep up with the latest hardware, or try to be content with the contributions that we can make.

Competition is fun, but most of us have a more philonthropic nature, and donate for the better good of us all, great or small.


P.S. The initial performance numbers from the GPU clients is a little distorted. The current work units are very small, and the NV hardware outperform the ATI hardware on small work units. That will change dramatically when the clients progress to larger work units. For the time being, an NV purchase may be the best bang for the buck. But for anyone who has been folding a while can tell you, the leader of the pack changes often because technology also changes quickly.

[FordGT90Concept]

Folding@home didn't become the most powerful distributed computing project by sitting back and only developing one type of client for on type of processor.

Funny you mention that because x64 support is still non-existant (10 of the 11 processors run on XP/Server 2003 x64 Edition). I have an 8800 GT and 8500 GT but I will not run beta software on them (especially my server). As far as I am concerned, a lot of "sitting back" is occuring.

Competition is fun, but most of us have a more philonthropic nature, and donate for the better good of us all, great or small.

Why do you think I got involved in this in the first place? My problem is that now, great monetary contributions in terms of CPU expense have become small in comparison to relatively minor GPU expenses. There is no longer any incentive for me to remain folding on my 11 processors (over $1500 in initial purchase costs, CPU only) just because they can't keep pace with a single GPU (under $300 in initial purchase cost).

For the time being, an NV purchase may be the best bang for the buck. But for anyone who has been folding a while can tell you, the leader of the pack changes often because technology also changes quickly.

I would never, ever purchase a peice of hardware for the sake of folding. I have been folding well over two years and went from last of my team to first (422,190 points) mostly because of that Clovertown server. Now, my $4000 server is liable to be passed up by a single $300 graphics card. I might as well quit folding on the computers because they are no longer point-competitive. The messages F@H is sending me is that my contributions are not worth the personal cost of keeping all 8-cores running at full bore 24/7.

[Pette Broad]

I'm in the same boat, lots of Cores (around 30) but being outscored by newbies in my team with one GPU. Although it's not a big issue for me I can see that it would be for others.

But for anyone who has been folding a while can tell you, the leader of the pack changes often because technology also changes quickly.

Good point, maybe spells the end of the dominance of The Corporate Folders.....


Pete

[v00d00]

Last time i looked, it was the users choice on whether to fold.

The newer clients use more resources, cost more, and therefore produce more points. It is just how it is.

I bought two quad cores last year and they knock out about 6000 PPD, but if i enable my GTX260 i can do about the same per day with ease. The difference is in the projects. GPU is good for certain things, SMP for others. So i will continue to fold both, as i want Pandegroup to get data for all diseases, and frankly points are pointless. People get so hung up on them and the competition side, which is nice, but it doesnt buy you a meal, or get you a trophy at the end of it. Bragging rights maybe, but chances are you would be branded foolish if you chose to brag about such a thing.

v00d00, a person who does buy hardware specifically for Folding 24/7.

[FordGT90Concept]

People get so hung up on them and the competition side, which is nice, but it doesnt buy you a meal, or get you a trophy at the end of it. Bragging rights maybe, but chances are you would be branded foolish if you chose to brag about such a thing.

People expect to be compensated in one way or another--an incentive to get involved in the first place. The new points system basically belittles all CPU contributions basically making their contributions pointless. The system needs to be corrected to not favor any type of technology but instead, to favor the generosity of those contributing.

[v00d00]

If you have an idea on how to do that without severely annoying everyone, we are all ears. In fact a similar thread to this has appeared many times in the past and still no one can solve the problem.

In reality it will probably never be solved.

But your idea may help, so please tell it to us.

[FordGT90Concept]

I think something as simple as "Score += Active Processor Count" would be effective enough, calculated upon every WU completion. It gives an incentive to dust off even the oldest of computers in addition to not favoring any specific technology. GPUs produce more WUs but usually there aren't as many. CPUs produce fewer WUs but there are far more of them out there. In the end, they balance each other out favoring the actual people that make an effort to contribute more so than hardware.

The only difficulty is in defining an "active processor" as GPUs especially are bound to go multi-core in the near future (Larrabee especially). I think that should be the only part that is routed in hardware. Each core should not count as an active processor, a physical chip should. If it is a quad-core, it will effectively produce four times the WUs as a single core chip thusly, appropriately compensating the user. The same goes for video devices. Each card should be considered a processor and not the individual cores inside. Again, the number of WUs produced reflect on the power of the GPU.

That simple formula favors the user contribution more so than the hardware but the hardware performance still does have a direct impact; however, GPUs may have to be weighted in order to attempt to make them point-competitive with comparative CPUs. For instance, if a high end GTX 280 produces 4000 more WUs in the same period as a 3.2 GHz Yorkfield, the GTX needs to have a 0.125-0.25 multiplier as to not slaughter the CPU contributions. Call it the GPU-to-CPU ratio. Other ratios may have to be introduced over time as new hardware comes out but the objective of that ratio remains the same: attempt to remove hardware type from having a major impact on scores. Reward the user, not the computer.

Using a "Score += Active Processor Count" system, an older computer would not contribute much in terms of WUs but it would contribute via bulstering the higher WU CPUs and GPUs.

[Foxery]

I have been folding well over two years and went from last of my team to first (422,190 points) mostly because of that Clovertown server.
...

People expect to be compensated in one way or another--an incentive to get involved in the first place. The new points system basically belittles all CPU contributions basically making their contributions pointless.

I'm worth 100 Million Internet Dollars! </SouthPark>

Honestly, if you hate curing diseases that much, what do you want other people to do about it? The CPU clients still make meaningful contributions to the research. Technology always moves forward, and Stanford has embraced it with a flood of Windows, Mac, Linux, GPU, and PS3 code. Like I said, new hardware really is that much faster. Computers always get replaced year after year. Maybe you can explain why owners of newer machines should be penalized instead?

This topic is a dead horse that has been beaten into a creamy paste already.

[FordGT90Concept]

The CPU clients still make meaningful contributions to the research.

The point system sure doesn't show it.

Technology always moves forward, and Stanford has embraced it with a flood of Windows, Mac, Linux, GPU, and PS3 code.

These things are besides the point...

Like I said, new hardware really is that much faster. Computers always get replaced year after year.

Moore's Law still generally applies but instead of thread execution rate doubling every 18 months (like it did prior to dual-core processors), the number of cores is doubling every 18 months.

Maybe you can explain why owners of newer machines should be penalized instead?

I will answer your question with another question: Why should owners of older machines be penalized by not folding on a GPU? As it stands now, it is GPU folding or bust. A 30 CPUs to 1 GPU ratio makes CPU folding completely unworthy of the power it consumes.


Edit: In the end, I don't care how it is resolved as long as it is resolved. The few people I know that are veterans of F@H are deeply annoyed by the recent changes. The entire point system needs to be overhauled. Another idea to play with is execution time--perhaps the great equalizer:

score = total processor time (probably in hours or days) running the clients

As such, it completely excludes hardware from the equation and relies solely on user contributions based on processor time donated. SMP clients would multiply their processor time by the number of cores used. The single-threaded and GPU clients give their time as is. The best solutions are sometimes the simplest.

[JBurton57]

I hate to be the bad guy here, but who says we should dig up every old processor and plug it into FAH? Now don't get me wrong; I like the project. I think we all do, or we wouldn't read these forums. But in my opinion, the project chiefly works because it favors a bigger, but ultimately otherwise useless hammer. FAH alone is not a good reason for me to dig up my old Pentium 166 and see if it can run a client (though, honestly, I doubt it could make the deadlines). Nor is it a good enough reason to go out and buy hardware. And to some extent or another, it's not good enough to let my computer run all night if I were going to power it down.

So what is it good for? For using up the clock cycles that would go to waste. For folding in the background while I'm word processing, looking at spreadsheets, checking my e-mail, or surfing the web. It's for when I'm reading these forums, and my processor is running tons of null ops because there's just no work to be done.

In this sense we need not encourage everybody to install FAH, because there are electricity costs, and ultimately we must weight these costs against the benefit of FAH. Is a few more points, or even a few more academic papers, worth tons of pollution? Most US power plants still burn coal. Some burn oil. There's a lot of nuclear power out there, and wind, and hydro, but we all have to think about the big picture.

The point system as it stands is harsh. But it is justly harsh. If my 65nm processor can crunch numbers faster and with less watts than my Pentium 166, then why not reward it for doing so? And if my GPU can crunch out the FLOPs even faster with proportionally less power, why not reward it for doing so? At some point FAH cannot be about trying to build a distributed computer with the highest conceivable processing power. If that's the case, FAH will eventually turn into one giant Prescott machine. Performance/watt must somehow enter the equation at some point. I'm not saying that the point reward system is designed to do so (I doubt it is in any conscious way) but it's a valid argument in favor of the current system.

It's true that CPUs aren't given a lot of points right now, in comparison to GPU folders. But then, not everybody has an nVidia 9800GTX. Or even an 8800GTS. Or even an 8600GS. We forget this when reading the forum, when people are building folding farms and buying 9800s just to fold with, but the average folder out there probably doesn't have a DX10 video card at all. He's probably using a Celeron P4 or Athlon64. Probably has a single core. Can't do SMP WUs. His video card is likely something like a GeForce 7600GS. This computer was probably built 3-5 years ago with mid-grade or budget parts. So, in this context it does seem fair to reward the people who have the latest hardware. They have, after all, paid for it. The problem is, we're used to thinking of 1000 points/day as "barely getting by." The average user out there is probably getting something more like 150 points/day, and neither knows nor cares how many points he has.

[FordGT90Concept]

If my 65nm processor can crunch numbers faster and with less watts than my Pentium 166, then why not reward it for doing so?
...
Performance/watt must somehow enter the equation at some point.

Because I firmly believe it is the effort that counts. Yes, individuals must weight the cost/benefits of running older hardware but, for those that only have older hardware, why not stand and be counted? Those older processors have just as many idle clocks for the software of their era as modern processors do. They too, could be more productive during menial tasks.

And if my GPU can crunch out the FLOPs even faster with proportionally less power, why not reward it for doing so?

Because the current system disowns all the CPU contributors out there.

...
So, in this context it does seem fair to reward the people who have the latest hardware. They have, after all, paid for it.

That is running on the assumption that most computers are designed for F@H which is simply not the case. Most people buy GTX 280 graphics cards and QX9770 to play their games. F@H or other distributed computing projects are an after thought. Well under 1% of F@H users have computers built to fold. The problem that boils up there is an inequation:

hardware ? users

Pande Group is ultimately going to have to decide which is more important and fill in that question mark with the appropriate symbol. Is performance/competition more important or is sheer numbers? Do you appeal to the 99% of users that do it for the cause or the 1% that do it competitively? The inequation has been and remains to be "hardware > users;" however, the divide was narrow because there was only CPUs doing the work and they progress steadily with the years. Now that hardware includes GPUs which slaughter CPUs, the inequation that has been relied upon no longer is acceptable. In order to keep the inequation working, CPUs need to be brought up approximately 30 times or GPUs torn down by about 30; otherwise, expect CPU enrollment to decline and GPU enrollment to increase. If that ratio ever gets to the point where you are losing approximately 30 more CPUs than you are gaining in GPUs, the points system has effectively diminished overall productivity. That is something to keep a very close eye on.