Just took a couple of weeks to get through a few SoB work units:

http://www.primegrid.com/workunit.php?wuid=144378166
http://www.primegrid.com/workunit.php?wuid=163868149
http://www.primegrid.com/workunit.php?wuid=143297745
http://www.primegrid.com/workunit.php?wuid=163723084
http://www.primegrid.com/workunit.php?wuid=163803878
http://www.primegrid.com/workunit.php?wuid=163723096

After completing them, they're credited with only a fraction of the points awarded that would have occurred had I been doing another project instead (ie. Proth Prime Search (Sieve)).

For example, with a Proth Prime Search (Sieve) WU that took 66,974.71 seconds, the credit given was 4,523.00.

http://www.primegrid.com/workunit.php?wuid=166847086

Yet with a SOB task that took 659,871.00 seconds, nearly ten times as long, the credit given was 10,449.05.

http://www.primegrid.com/workunit.php?wuid=144378166

This is on reasonably modern Intel Core2 stuff, not ancient cpu's either.

Is something really screwy in the credit calculations with SoB, or is this expected? Maybe a way to de-emphasise SoB so people don't do it or something?

I agree with your assessment of the credits. The new versions of sob give between 11,000 and 13,000 but still way below the combined amount for a pps sieve run of the same time length. I'm still not sure how credits are assigned to specific projects and what they are based on.

So, if you can get them to up the credits I'm all for ya !

RR
____________
@AggieThePew

For my account, I've disabled SoB now (of course), and suspect that's the intended result.

My guess is just that the PrimeGrid team REALLY need to update the website to make clear which projects are deemphasised like this, so people don't waste their time.

The present website's communication in regards to the crediting system is clear as mud. :( And it wouldn't be hard to improve it. ;)

sob's are just fun cause they are such a challenge to get done and even more to get validated. That's where the real fun and frustration begins :)
____________
@AggieThePew

"If you are in only for credits, it may appear as waste of time."

That seems like a pretty strange answer.

What's the point of imaginary "credits", if they're not related to each other via importance in some way from the project's point of view?

How else is a person new to this project supposed to tell which WU types are more or less important to PrimeGrid, when the only measuring stick given is these imaginary "credits"?

I can see you've been around for a while, so I'm guessing you're just not relating to a newbie's point-of-view very well. "credits" is the measure we seem to have been given for what the project considers important here.

Credit is not based on importance of difference tasks, it is based on work done (FLOPs). GPUs doing much more work than CPUs on sieving, hence much more credit per hour.

Each users has different view on what subprojects on PG or BOINC project in general are more important.

An very early post by John may also help This Post

Sieves do different work than LLRs so are awarded different amounts of points.

@VapourLock
I checked your two computers and the points per hour rate for your SoBs are around 62 and 57.
For your Sieve work you are getting 261 and 245 points per hour.

Either way the 60 odd points per hour of work is very good, you try Malaria@home, EDGeS@home, Mersenne@home or even Spinhenge@home and see how many credits you are awarded.
You will then appreciate PrimeGrid and the amounts you are awarded here.

Conan
____________

Either way the 60 odd points per hour of work is very good, you try Malaria@home, EDGeS@home, Mersenne@home or even Spinhenge@home and see how many credits you are awarded.
You will then appreciate PrimeGrid and the amounts you are awarded here.

In any case, crunch what you want (do not feel pressured to crunch it all!), when you want.

I know John said they are going to be working on adjusting the credits in the next couple of months but maybe it would be nice if someone in the know laid out how the credits (cobblestones) are actually calculated and why since I'm sure the original formula has been adjusted some over time. That might help new people and myself understand just a tad better.

Example of a perceived issue. "Credits are calculated on processing time or use". With this example I know that can't be the case since the credits for pps sieve are the same for both the cpu and gpu. So the next thought that comes to mind would be that the credits for each sub project are just an arbitrary number that was developed some how. And again that might not be accurate. So it's easy to understand how someone very new to this project might be confused.

Anyway, just a thought.

Rick
____________
@AggieThePew

Here's a primer to start your journey down the rabbit hole. ;)

Computation credit

While we don't have a 64 bit LLR application, an llrCUDA might be around the corner. It is marginally faster than LLR CPU at lower n's but really shines bright with the higher n's. An ideal match for SoB and PSP. But don't get your hopes up too much. Currently it still uses about 90% of a core to run.

Eh, little of topic, but this is very interesing subject.
Which project (llr) will definitely not be on gpu or it would be very ineffective? PPS, SGS i think, but rest should be decent, right?

Depends on what GPU (+single core CPU) you are willing to put on LLR.

[back to topic]
imho there could be a fairly and easy way for counting points:

1. Choose a PG-refferersystem (i.e. a Q6600 with a CUDA and ATI card)
2. crunch all different tasks and publish the time of that WUs as refferer in a sticky thread
3. give i.e. 3000 point for every 24h independly of the task because time is energy and energy is money

The result would be: All task would be worth to crunch and very long tasks give higher badges at once. (i.e. genefer long gives about 300.000 atm) If some like more credits in time, he has to pimp up his soft- and hardware, so its faster than the referersystem and reach more than 3000 points in 24h.
i.e.: One core of my 8150 needs aprox 27h for pps sieve. If I set it into relation to the refferer system it could be slower or faster and so I would have a direct comperison, how fast my machine realy is. Same mechanics for CUDA/ATI: If someone doesnt want to use GPUs (because of the price or energy costs) he will reach +/-3000 point per 24h with a singelecore instead of 300.000 with a modern GPU. (all in comparison to the fixed refferer system). At least this fairly system would make SOB as interesting as all other tasks.

Badges are important, too. Its the game-effect or "olympic affair" wich build the motivation to get them. (Hey, who said that science must be dry and not funny? ;))
Perhaps some new badges could motivate more ... i.e. for people who crunch with very old hardware, because they spend a lot of energy and time in PG (in relation to their points) and have much patience. So there could be badges for i.e. Pentium-4 or ADM64 CPUs and older stuff wich people like me never can reach with my newer hardware as long I dont reactivate an old machine. ;) It should be no problem for the dbase to filter it. The score of this "oldschool-badges" could be: Years of PG-Membership * so far reaching points with that old machines. (i.e. 3 years * 10.000 points so far = score 30.000) That would be fair, too, because not everyone can buy the newest hardware. (or think about all the Atom Netbooks out there ;)

Last but not least: Science. It is important to explain, what the WUs computes, so also laymen can understand it. PG has a WIKI and thats fine for people who want to learn more about primes.

best regards,
ritch
___________________________
Sorry my English, I'm still not a native speaker ;)
____________
best regards,
ritch

101*2^382407+1 <-my lucky number

For over a year I have been using a spreadsheet to find mean run times, credit, credit per day/hour, time/WU to the next badge and such like.

I could post the results but they are never going to be accurate, some projects I have not run recently so credit and run times will have changed, some were run on one machine, some on another and some on both, PPS sieve is running on a GPU etc.

However I can say that credit does vary from project to project, SGS I found the worst paying.

This is not a complaint, just an observation of the way things are.

SOB has other advantages -- You can load work and forget it for many days (ideal if you have a rarely connected dial-up or if you are going away and do not want to leave the connection running); you cannot get SOB badges without running it and if you find a prime it will be big.

Whether you run it or not is up to you, credit rate, I expect, will be only one of the factors in your choice, but if that is your main aim then I agree SOB is not for you. If you want fast jobs, quick crediting or rapid wingman confirmations then it is not the one for you either!!
____________
Member team AUSTRALIA
My lucky number is 9291*2^1085585+1

Hi Dave

Because of that my suggestion is a fixed refferer system as I wrote before. PG would only have to collect the results of that PC, publish them in a list and declare it as an official refferer to compute the points for all other machines in comparison to the time they need. As long someone use the same machine as the refferer system he wil get aprox the same points per day. If someone uses better, newer, faster machines he will get naturaly much more points in time. At least it would be a general question of the core-number, OC, OS and some advantages features of CPUs.
That would imho be fair for all crunchers and all projects - and the refferer could be a mid-range Q6600 using one of 4 Cores, ATI and CUDA tasks for example.
At the end of this, you will allways find a wingman (quickly) in all projects (SOB included).

PS: I crunch mostly all WUs I can get because my cores must have something to do while the GPUs produce the points for rankings and RAC. ;)

best regards and greetings to the other side of the world,
ritch (team fv-verlag-group Germany)
____________
best regards,
ritch

101*2^382407+1 <-my lucky number

Seems like all of my SoB WUs validating now are hitting the 17000 credit cap. Predictably, here are my two thoughts:

1. Are a lot of people seeing this?

2. Should the cap be reviewed?

Actually I seem to remember getting far fewer crdits in the past, but I can't remember how much work they involved. Different PCs too, anyway.