M40 Again!!!!

[GP2]

Originally posted by PageFault
Experience at ars indicates 80 - 90 % will vanish within a few weeks of this type of recruitment. [...] Of course the same thing is happening with TPR right now. Of the M39 casuals, perhaps two or three remain - most never completed the first test.

Yes, but why should this happen?

The default install on Windows XP should just install Prime95 as a service, so you can install it and forget about it, even if you lose interest. If new users stop returning results, that means they actually took the time and went to the trouble of uninstalling it... why? Why not just forget it and leave it?

What was different two years ago? Well, Windows XP wasn't around, so maybe the default install that was available at the time didn't install it as an automatic-startup service, but as an ordinary program that you had to manually restart after every reboot... does anyone remember? George?

This time we'd really like to see much better new-user retention. Teams can play a vital part in this... Team Prime Rib wants to maintain its lead over curtisc? Free-DC wants to move up in the rankings? Recruit and retain some new blood.

Also, this forum wasn't around two years ago. Hopefully, it (and the New to GIMPS forum) can make a major difference.

A high rate of new user attrition should not be accepted as inevitable. They were drawn to GIMPS for a reason, so they have some initial interest in the project. And they've installed the ultimate set-it-and-forget-it background application. A few might even be motivated by the $100K reward for a 10-million digit prime.

A high rate of attrition this time around should not be accepted as inevitable, in fact it should surprise us, and if it happens we'll need to figure out why. Maybe even go so far as to do an e-mail survey (for those who clicked on "accept newsletters") asking why they went to the time and trouble of uninstalling Prime95.

[GP2]

Another question we'd need to ask is, would a screensaver version of the program lead to better retention of new users... or maybe even a higher rate of users trying it in the first place?

[Prime Monster]

Originally posted by GP2
Another question we'd need to ask is, would a screensaver version of the program lead to better retention of new users... or maybe even a higher rate of users trying it in the first place?

Probably - if you could manage to get a kind of cool factor to the screen saver.

PM

[GP2]

Originally posted by GP2
A high rate of attrition this time around should not be accepted as inevitable, in fact it should surprise us, and if it happens we'll need to figure out why. Maybe even go so far as to do an e-mail survey (for those who clicked on "accept newsletters") asking why they went to the time and trouble of uninstalling Prime95.

Maybe they saw some program seemingly "taking 100% of the CPU time" and figured that uninstalling it would make their computer work faster. Maybe they set their P-1 memory limits too high and their computer slows down and swaps during stage 2 (that's an argument for making P-1 a separate type of work, just like trial-factoring, so the average user will never do it).

If a high attrition rate happens again, we need to datamine the reason for it. Something as simple as changing a few text messages during setup, or the wording of a few dialog boxes within the program itself, might make a difference.

[Prime Monster]

It will be interesting to follow.

I do agree with you that getting the "professional" DC teams involved would do a certain amount. The client is optimal from their point of view, very stable and configurable, highly optimized and with some very useful features. Just what should attract the teams.

There are two big events that will happen shortly in the DC world that could effect us quite a lot. One is the end of Seti@Home I and the other is the "near" completion of ECC2-109.

If there are anything to learn from this event, then by correcting and tweaking things before these two upcoming event, we might get a higher retention rate next time.

PM

[eu-Robert]

That article also has a cute little graphic :
http://www.spiegel.de/img/0,1020,311769,00.jpg

Now look at these top 11 lines of mersenne40.txt :

12597689545033010502049430957482431145599341608535183595225467012565498768908\
35156022124009680282853613254412715832332548115046530107631671237352586512237\
97623392168097752904174121031793027766749883270131702229942984844394149386141\
46923615146405320384930131677411867193308775658535744726248719065403711481011\
86423521460887061584240946931461144886371568165702926779681963275230120108755\
67866137704610554959335850058929413976069101429767323404583564854828291042054\
10261521824675460358635318886299052948972072378456299284969747785194967449947\
26335777846008407305422706323723085738620878680121763732494460717640640582013\
15319242434805555451075151859547651927121451554795786023853642248011842056001\
81869208528967063366243443423796336048257400497290787589379577456413984625396\
47505572610833665825447782677536558540045164852919840692279825348018826398757\


Well, that certainly explains what file they used for their graphic, doesn't it!
A good thing the (77-digit per line) decimal expansion of M[40] is not copyrighted... (can it be?)

Oh, by the way, congrats to George et al for their superb achievement!!

Originally posted by TauCeti
spiegel.de Student entdeckt Rekord-Primzahl

[PageFault]

It is inevitable. This is a a reality of the game. Some of you non-pro's might have a hard time to see this, as you have vested personal interests in running prime95. The majority doesn't. - I certainly didn't, all I wanted was a good client and a challenge - dnet couldn't deliver the second part. They see something "kewl" in a newsflash. Three days later, they haven't found M45 yet, or the cure for aids, or whatever. Three more days and it's gone.

This is a typical cycle, one the ars folk know all too well. Today we had front page pimpage, resulting in dozens of new machines showing up on the team account (this in itself is amazing, testament to having readily available quality faq's on the team site). Not a single newbie post yet in the dc arcana. But, I don't fret - of those dozens, I want the guy who has maybe twenty machines in a work lab, or a good job and time for a virtual hobby. We will attract this cruncher and retain him - of that I am sure. At the first call to arms, this individual will be spellbound and acting with credit card in hand.

As an example of what recruitment can do, this time around TLC found over a million seti wu in a two week period - amazing. Last time, they did the same, and found morphie - hello team, have 200 P4's ... new leader of the pack ...

Maybe I'm making this sound sleazy, but it's the way team based DC is. Been there, seen that same thing all over, every time. The lone individual doesn't even have this much behind the effort. I took place in the hottest statsrace evar - three months of hell, seeing both incumbents drive each other to goals thought impossible even a week before. A stupid glitch ruined it overnight - a team bigger than gimps folded overnight.

The only hope is to draw the individual in here and addict them.

Originally posted by GP2

A high rate of attrition this time around should not be accepted as inevitable, in fact it should surprise us, and if it happens we'll need to figure out why. Maybe even go so far as to do an e-mail survey (for those who clicked on "accept newsletters") asking why they went to the time and trouble of uninstalling Prime95.

[GP2]

Pagefault, you may be right about the motivations of people who are in it for loyalty to a team like Ars Technica or Free-DC, and who readily hop projects or divert resources to other projects for gauntlet challenges and so forth.

But my guess is that the influx of new users is largely composed of people who read the mainstream media stories and who are new to distributed computing... not hardcore crunchers drawn in temporarily from other projects. Thus they must have some specific interest in GIMPS in particular, rather than just distributed computing in general... and most mainstream news stories never even mentioned the $100K EFF prize, so I'm guessing that wasn't a major motivation either.

I took place in the hottest statsrace evar - three months of hell, seeing both incumbents drive each other to goals thought impossible even a week before. A stupid glitch ruined it overnight - a team bigger than gimps folded overnight.

I keep hearing indirect allusions to the stats fiasco that befell one of the other distributed computing projects... could you provide the details for our benefit? Can we try to ensure that such a thing doesn't happen in GIMPS?

[Xyzzy]

IIRC, it was G@H, and I think the fiasco was they reset the stats or maybe came along later and "readjusted" them causing people to lose tons of work... For some reason I think it had to do with sneakernetting and duplicate results...

[Xyzzy]

[Dresdenboy]

Hmm... it's time for an further optimized Athlon FFT. Seems I have to spend more time on this.

Theoretically the Athlon could come close to a P4 per clock. It has to be implemented. Maybe my new implicit transpose (making use of modern processor features) could help here.

More on that later..