PG announcement about Big Advanced Projects

[fayzacantik]

Bruce, are you saying what ever people talk FAH policy will not change ?

[HaloJones]

True democracy - one man, one vote.

Prof. Pande is the man and he has the vote.

[Jester]

I can see clearly the need to keep future Bigadv projects in the sole realm of high end hardware, and that will of course have to change with new hardware,
but surely I'm not suggesting the impossible, no guarantee's of Wu availability, or even of the beta Bigadv project itself, just some sort of actual policy
rather than just "we'll do what we like, when we like and give as much notice as we can"......
Why not have something like "while we can never guarantee a steady supply of any series of Wu's, or the continuation of any series of Wu's or project, the hardware requirement changes
in January 2012 will not be changed or reviewed before January 2013" ?
Unless of course manipulating hardware requirements is no more than a tool to steer donours towards whatever project is the current priority.

[7im]

When there is a standard schedule from Intel and AMD in regards to their hardware releases, I'm sure Pande Group will be able to provide you with a standard BA update schedule.

This is mostly hardware driven, not at all like you described... "we'll do what we like, when we like and give as much notice as we can"......

And don't forget... Pande Group can steer donors towards whatever project is the current priority any time they want. They simply change the Assignment Server settings. They have never needed to change hardware requirements to do that.

So your insinuations about what PG are doing, and why, are a little off the mark.

I'm also sure that PG does review the hardwardware needs on a regular basis, but the need for a change is so infrequent because of their long term planning. I can see how that would appear to be a random event to the casual observer.

[bruce]

My impression is that everybody is focusing on the wrong aspect of this announcement. First and foremost, the deadlines will be changed to take full advantage of the recent increases in speed for server-class hardware. If you don't have server-class hardware, then assigning you even an occasional WU will not make either you or the science happy. Consumer-class hardware will simply not be able to complete the WUs by the deadline. The 16-core requirement is secondary. It's an attempt to differentiate between server-class hardware and consumer-grade hardware and to distribute assignments that you have a chance of completing before they expire.

Then, too, FAH has an abundance of standard SMP projects that need processing and a shortage of BA projects compared to the number of machines looking for those projects.

[chriskwarren]

Sigh. Some heavy hitters and big-time contributers on my team (not just PPD contributers, but valuable advice and mentoring) are mulling their options. Hopefully it blows over and folks see the bigger picture but I get the feeling we will be losing some serious hardware from our team that will get repurposed.

[bruce]

Bruce, are you saying what ever people talk FAH policy will not change ?

I trust that Dr. Pande knows what he's doing when it makes a decision that benefits science, ESPECIALLY when he already knew it will be unpopular with some people. Do you really think that you can convince him that he needs to dumb down the scientific goals of FAH when the whole point of the change was to improve the science by taking full advantage of the server-class machines. Read what he actually said in the announcement. The purpose of BigAdv is to use the most powerful machines, not machines that were considered powerful last year or the year before and the deadlines need to be shortened to be consistent with what those "most powerful machine" are actually achieving.

Of course the policy will change. The most powerful machines will get faster every year and FAH will again realign the deadlines with those faster machines. Moore's law says that computer speed will double every 18 months and I'm confident FAH will figure out the best way to use that increased speed.

[fayzacantik]

Bruce in my humble opinion Standford need better PR and and more informative.
I always read the BA or what ever the name need done as fast as possible but there is no information why, you don't need to offer one semester class just some plain English or may be better show us in Graphical will enough to most people to understand why PG need it as fast as posible.
the last but not least PG do not like machines with heavy OC maybe they believe this heavy OC machines will give them more trouble they can take so they play safe ask what you called
the server-class machines to do the job. that why you need to put 2 things in your fuzzy website
1. Put a benchmark. please don't talk about different WU mean different point etc people need to know is their faster or not that it.
2. Put Burn test so people will not mess any more with unstable machines
the last for me this time please, please, please fixing your program I can live with more electricity bill but endless upload for my FAH is not acceptable thing

[3.0charlie]

Please define "most powerful machine", for a run-of the-mill folder that invested its own funds into that project. For me, it's my 4p rig (quad 6128s). Not an issue with this change, though ppd will be impacted. It's also my 2p rig (dual 6128s) - this is an issue for me. Return on investment - ppd / $ will not be same, since the rig will be barely capable of BAs, even though it IS a champ today in regards to ppd/w/$.

"Fold for the Science!". Yeah, we all do. But would we fold as hard without the competition that the points systems created. I'm not sure. Take away this very-much capable 2p rig from folding BAs - remember, return on investment for the casual folder - means less points, less competition, less WUs folded and uploaded. My loss, your loss.

We are volunteers that invest large sums of funds, anonymously to the University. You guys need us. I'll still fold, but as hard? I doubt it.

[7im]

Less points? Sure. Less WUs folded? Not likely. The change will actually get more WUs folded. The SMP WUs that are backlogged. Less competition? Not really, because the change affects everyone currently folding BA work units.

There is no way to define the most powerful machine. As mentioned above, that is always changing with Moore's law.

The BA program has always been an experimental program, subject to change. There was a change in BA just last year. Now we have another change, and change has been stated since the beginning of the program.

I understand people's concerns about their investments. I've been folding since 2003, and I've seen several generations of my own PC hardware purchased and retired. But we knowingly invested in a system that was guaranteed to change, every 18 months or there about.

And even if the BA didn't change last week, every PC owner knows all to painfully that they are continuously on the upgrade merry-go-round every few years. 4 years ago, people were buying dual cores. 2 years ago, quad cores. Last year, hex cores. This year, Octocores. Even if BA didn't change, your existing systems would be outperformed in a few months by the next newer system to come along. Your PPD is always dropping as compared to the next cheaper faster upgrade. That's just the way it is, and always has been.

[Grandpa_01]

Less points? Sure. Less WUs folded? Not likely. The change will actually get more WUs folded. The SMP WUs that are backlogged. Less competition? Not really, because the change affects everyone currently folding BA work units.

There is no way to define the most powerful machine. As mentioned above, that is always changing with Moore's law.

The BA program has always been an experimental program, subject to change. There was a change in BA just last year. Now we have another change, and change has been stated since the beginning of the program.

I understand people's concerns about their investments. I've been folding since 2003, and I've seen several generations of my own PC hardware purchased and retired. But we knowingly invested in a system that was guaranteed to change, every 18 months or there about.

And even if the BA didn't change last week, every PC owner knows all to painfully that they are continuously on the upgrade merry-go-r ound every few years. 4 years ago, people were buying dual cores. 2 years ago, quad cores. Last year, hex cores. This year, Octocores. Even if BA didn't change, your existing systems would be outperformed in a few months by the next newer system to come along. Your PPD is always dropping as compared to the next cheaper faster upgrade. That's just the way it is, and always has been.

This statement should be framed and hung on a wall. Not that I always agree with you but Well said 7im

[Jester]

When there is a standard schedule from Intel and AMD in regards to their hardware releases, I'm sure Pande Group will be able to provide you with a standard BA update schedule.

This is mostly hardware driven, not at all like you described... "we'll do what we like, when we like and give as much notice as we can"......

And don't forget... Pande Group can steer donors towards whatever project is the current priority any time they want. They simply change the Assignment Server settings. They have never needed to change hardware requirements to do that.

So your insinuations about what PG are doing, and why, are a little off the mark.

I'm also sure that PG does review the hardwardware needs on a regular basis, but the need for a change is so infrequent because of their long term planning. I can see how that would appear to be a random event to the casual observer.

If you termed "casual observer" as anyone not in Pandegroup or the DAB I'd agree,
and I'm sure there would be a lot less angst in this thread if a "standard BA update schedule" or even an estimate, had been made public in the forums a few months back, or at least something in the form of the readily available Intel roadmap as a guide,
most "informed insiders" have been able to look way ahead at what's coming........
Intel's big price cut on 6 core 12 thread Cpu's had a lot of folders snapping them up even though they were still a lot more than the slated $200 Cpu, the new 8 core 16 thread minimum (in reality) has relegated them straight to smp,
if of course there's going to be something done about misrepresenting core numbers, tighter deadlines won't do it either as reducing deadlines enough to shut out say a 3.6g Hex core Cpu
will shut out lower end server class rigs like dual quadcore 2 Ghz Xeons that meet the hardware requirements,
don't worry 7im your words from long ago still ring true, "don't make hardware choices on current projects", a little more knowledge about future directions in folding would sure help though..
I've also read a few comments about this increasing the gap between "the have's and the have not's" (for want of a better term) and maybe it's something that could be looked at by adding a third level,
For sure the top tier "server class" hardware running the toughest of Bigadv, but how about slotting in something a little less demanding for "enthuisiast level" hardware instead of lumping a lot of dedicated
folders in with "consumer level hardware" and "casual" folders...
As far as letting assignment servers handle project priorities it seems Bruce's comment doesn't support that:
Then, too, FAH has an abundance of standard SMP projects that need processing and a shortage of BA projects compared to the number of machines looking for those projects.
Not 100% in context I know, but the inference is there.

[chriskwarren]

I think the main issue here is a marketing one. I don't blame PG for wanting to change the hardware requirements for certain types of science.

They had a couple of basic choices;

1. make changes that will reduce the PPD available to certain people (ie. the bigadv folders), or

2. they could have increased the PPD available to other folks (ie. GPU and SMP)

You could have gone with option 1 and got the desired result (redistribute the work to be done), or they could have gone with #2, which would have done the same, but would have resulted in 'PPD inflation'. The choice was to 'take away' from one group, or to 'give more' to another. Relatively speaking, when you compare everyone who folds, the relative end effect is the same logically for both options. Logically, option 1 is the best, as it is simpler to implement, and helps fight PPD inflation.

Problem is, any marketer will tell you that 'taking away' (or being perceived as such), is never the way to go. It brings out a negative emotional response. In this case, they are taking away from the 'heavy hitters' and leaders of the various teams, and the negative response from these relative few will have an effect on 'regular' folders as they follow the leaders.

Logically, I understand why PG did this. But I think the smarter emotional approach to the problem they faced (or intended goal) was to simply give more points to the other hardware and donors out there, and simply made bigbeta projects 16 threads and sent the newest, baddest hardware to those projects.

[7im]

And there are long time leaders and folders that would also come out against the #2 "points inflation" for 3 big reasons.

1. Points are based on scientific value (always on the Science, and recently added timely return, also scientifically valuable).
2. Arbitrarily inflating points not based on science devalues past contributions, causing long time folders to want to leave the project.
3. You can't throw points at every adjustment that is needed. We'd all be making bazillions of PPD, and it would be meaningless...

If anyone can accurately predict when Intel will cut prices, and then predict if the economy is good enough that people will snap up the cut priced chips in volume, and then predict if they will be folding with those upgraded systems, and then how many will choose to overclock enough to make the deadlines, then PLEASE come consult for Pande Group (right after you give me the next lotto numbers)! I just don't think that kind of thing can be accurately predicted, even if you already knew that PG makes quarterly reviews (<-- made up interval, I don't know how often they do it, just that they do it). It wouldn't make any difference.

But I do agree with the concerns about the appearance of developing classes of "have"s and "have nots" folders, but then I've always known that I personally cannot compete with large businesses that fold on corporate servers, and that I can't compete with huge computer labs folding on college campuses. Then again, you couldn't touch an 8 core system for under $2500 when BA WUs first started. Now you can put together a 16 core system for about a $1000. Seems like the "haves" might be more reachable now. Call me pragmatic.

[Jester]

Ok pragmatic,
But trying to build "affordable" Bigadv rigs would go against the stated intentions and goals of the Bigadv project,
if we all try to be "have's" by doing so then the bar would need to be raised higher still to avoid the current situation of demand outstripping suppy....
sort of a "catch 22" ?