Case in point: http://www.primegrid.com/workunit.php?wuid=253188475

So don't do as I stupidly did and abort a task because it appears to run too slowly.

I should elaborate that the change in N means that the tasks will take about three times as long as the N=262144 work.

Don't know if the wu I linked to was just a part of a test batch or if all new work will be from the higher N. We haven't reached the GeneferCUDA b limit on N=262144.

Also, credit has been bumped to 8,000. This applies to N=262144 reissues as well. Get them while you can...

Does it mean we are finished with 262144 at ~750k (and no other prime) and starting with 524288 at 110k on BOINC GFN?
I see PRPNet is gonna finish at 500k with 262144 and at 110k with 524288 in about a month.

On another note - I'm usually not the one to complain about credit but...
262144 takes about 2680 secs on my GTX580, gets 3600 credit
524288 takes about 8190 secs on my GTX580, gets 8000 credit

262144 does about 116k credit a day, 524288 does about 89k a day, that's 1/4 drop.
I'm not sure how much farther it will drop as b goes upward.
____________
My stats

Across my cards I am seeing the same thing as Honza. Credit adjustment is an increase of about 2.2 times higher, but 524k work is consistently about 3.2 to 3.5 times longer. If the desire is to push the GFN search, then some adjustment will need to be made or GFN will likely miss out on much crunching power that will stay with the PPS sieve work.

Also, is this proportional increase in length likely to remain consistent at each increase in b level? I remember Michael Goetz noting that his 90 minute b=262144 work would translate to about 8 days for the b=4191304 work, which would be quite consistent with the 3.2 (6.5 days) to 3.5 (9.4 days) ratio noted above. If so, the future credit assignment should be fairly straightforward as we move up in b level.


____________
141941*2^4299438-1 is prime!

Also, is this proportional increase in length likely to remain consistent at each increase in b level?

Does it mean we are finished with 262144 at ~750k (and no other prime) and starting with 524288 at 110k on BOINC GFN?
I see PRPNet is gonna finish at 500k with 262144 and at 110k with 524288 in about a month.

262144 does about 116k credit a day, 524288 does about 89k a day, that's 1/4 drop.

Nice title, John.

Very appropriate!
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Nice title, John.

Nice title, John.

Thanks. Bestowed by the powers that be...and much appreciated. However, I think I lose it if I stay. ;)

Nice title, John.

Thanks. Bestowed by the powers that be...and much appreciated. However, I think I lose it if I stay. ;)

[edit] Discussion is ongoing for a second GFN project to be implemented where N=4194304 work would be offered.

Will there be 2 sub projects or just one with all the work from the short and long versions going to the same sub project?

So now the 64K question for you Mike. Have you switched over to the (wr) units yet :)

Will there be 2 sub projects or just one with all the work from the short and long versions going to the same sub project?

Although GFN is in production now, credit is still beta. Data is being gathered to help pinpoint a more reasonable level. Once that's established, a nice formula by Michael will be used to very accurately assign credit to each b/N task.

Neo, I'm going to play devil's advocate here. You are saying that the current credit (8000) is too low because it's less credit/hour than you were getting with the 262s at 3600 credits.

Middle school algebra is all you need to know that this is correct.

However, what's to say that the 3600 you were getting was "correct"?

Consider this:

You can run these WUs with GeneferCUDA or with GenefX64, Genefer80, or Genefer. Of the three CPU builds, Genefx64 is the fastest.

On my hardware, GeneferCUDA can run a 262 in about 90 minutes. I can also crunch the same WU with Genefx64 in about 15 hours. It's the same WU, so I should get the same credit for both. If you're running on a Mac and crunch this on your CPU, you do in fact get the same 3600 credits, as you would expect.

So far so good, but here's the kicker: Is there any other CPU app, either here or anywhere else, where you can get 3600 credits on a CPU so quickly?

(My computer is a failrly slow Core2Quad, by the way. I'm not running a water cooled unlocked Sandy Bridge i7, so lots of computers can certainly do it faster than 15 hours.)

A good argument could be made that 3600 credits is just too much, compared to other CPU apps. If you've got a good counter-argument for that, now's the time to speak up. THAT is how you'll get the credit raised.

If 3600 is too big for the CPU, then it's too high for the GPU also since they're processing the same WU. With CPU apps, there's lots of comparisons that can be made to decide what's "correct". With GPUs it's much, much harder to determine how to set the credit other than "I want more!". Therefore, if you can set "fair and correct" credit for the CPU, then that should be done, and GPU credit is then set so that it matches the credit for a CPU WU.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Even though I really like to find a larger prime I'm not going to crunch at unfair credit compare to a different range.

In the meantime, we'll continue to gather data as quickly as possible so that a "reasonable level" of credit can be offered.

That's not too hard ;-)
all on the same PC with stock GTX570 (Gigabyte)
times are averaged values from several WUs
Cullen/Woodall (sieve) 127 sec - 364 cr - 10330 cr/hour
PPS (sieve) 915 sec - 3371 cr - 13262 cr/hour
Genefer 262144 2880 sec - 3600 cr - 4500 cr/hour
Genefer 524288 9060 sec - 8000 cr - 3179 cr/hour

Regards,
Mike

I understand the next level (1048576) will be paid even less than the current one.

I'm not sure I could find at least one GFN prime, but if I do not do this now, later I will have to spend much more time to achieve the same ruby badge than it is now.

Oh, excuse me, I forgot we planned devide N between boinc and prpnet.
But I remember Johh said below

Yes, N=262144 is suspended for now in BOINC. N=524288 will be searched until a prime is found and then the next N will be inserted...and so on.

There were 2 points on credit I was trying to make (despite being in a sick headed fog at the time). One, of which Mike answered, is being the per hour credit at each N being linear when permanent credit is assigned. And I like your bias thought as longer WUs mean more wasted work/time when something does go wrong and that would be a (very?) small help in psuedo-compensation.

The other was that if credits are in beta then temporary estimates should be updated as quick as possible to known run times according to previous 'beta credit'. I know the current numbers are temporary, but that doesn't mean there shouldn't be some consistency. Unless the 8000 is an intermittency number, bracing us for the possibility of an even lower permanent number at that N.

I thought this was a GPU project and that the Mac BOINC CPU version was temporary as N=20 or higher is GPU ONLY territory, and a Mac GPU app would eventually get developed.

As far as other projects (MW, Moo, DistrRTgen SETI, etc.), I'll get some data but as you know there is big differences in nVidia and ATI/AMD GPUs at each project (and within each project) and I think taking data just from nVidia would be wrong as an AMD (opencl?) app may come along as well. AMDs 7000 series GPUs with GCN and nVidia on their 600 series are both going to push opencl (there's still CUDA for niche programming), so there might be a GFN opencl apps for both in the (very?) near future. Probably averaging ALL GPUs and not just nVidia would be prudent. Some data from other projects to come.

NM*

Added:
I decided to crunch someGFNs, mixing GFN and PPS_Sieves on my 460 for now. My bi-polar got the best of me yesterday I guess. But it still always gets in my crawl when crunchers credits are lowered (or run times increase without proper credit increase), however temporary it may be. Should have seen the riot at MOO a couple of months ago because WU times increase 5% but the credit didn't. After a couple weeks Teemu inserted some custom code to give just the slightly longer WUs more credit (plus the Santa bonus).
____________
Largest Primes to Date:
As Double Checker: SR5 109208*5^1816285+1 Dgts-1,269,534
As Initial Finder: SR5 243944*5^1258576-1 Dgts-879,713

I thought this was a GPU project and that the Mac BOINC CPU version was temporary as N=20 or higher is GPU ONLY territory, and a Mac GPU app would eventually get developed.

Looks like I've accidentally started long discussion about credit. Guess it would come sooner or later anyway.

Other projects like MilkyWay, Collatz, Dnet, Aqua, Einstein, distrrtgen, DrugDiscovery, Moo and SETI must have been discussing the same issue, but I can't remeber how it was (it is not so important to me after all).

I think general approach was using calculation #2, simply because: CPU app is born, fair credit during beta-testing is established. Later on, some comes up with optimalized app...if it eventually becomes stock app, credit is lowered to be in par with other CPU apps/project.
Later on, GPU apps is developed. (After a long, long discussion)...it gets the same credit as if being done on CPU. It is the same work after all, just done more efficient. This is the case of GFN.

It might be a bit different where there is GPU app straight away (GPUGrid). Still FLOPs count can be used to establish some credit.

Let's go chasing primes :-)
____________
My stats

All people in this thread ( except John and Michael) are little sarcastic.
I look at this thread, and see same answers I got on another credit thread.
We, all of us: using our computers, pay our electrically bill, spend time on Primegrid, and all we get is BOINC credits ( and in this case badges in different colors)
I will not start any discussion since many of you thinks that "I talk too much".
I am just glad to see , that others also hunting color badges or credits.

We all do that. Everyone will be happier with more credits, everyone, right?
It is not story that I hunt credits, all of us do that, because credits are only thing we get from this project. And please dont say to me: if you not satisfied with number of credits per WU , then PG is not project for you, and you can leave us.
That is not answer, that is rude.

Let say for GeneferCUDA , on 262144 my card do it for 1 hour.
If next level will do for 4 hours then I expect 4*more credits.
I dont care because someone crunch that with CPU: make it equal, rise credits base for cpu, and all problems is solved.
Is that hard to do?
Look around yourself: this is 21 century, we make calculation with GPU ,and before only few years we all can just dream about it. Adapt credits scheme to present time, to powerful CPU and GPU.

And last thing: CPU before few years at 3 GHz and CPU today, at same frequency dont crunch same time, technology is changed.

P.S I dont wont to start any war, I think that is time do adapt PG to current date state. Look at users: many of us have latest CPU-s and GPU-s. We dont crunch on CPU that is few years old... We are specific in many ways...
____________
92*10^^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^⁴⁹⁸¹⁰¹-1 CRUS PRIME
2022202116^¹³¹⁰⁷²+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie!

One thing you should think about:

Credit isn't a real currency, of course, and any BOINC project admin can simply change a few lines in a config file and arbitrarily increase the credit their project hands out. It costs nothing to do that.

If you increase the credit, more people will crunch at your project instead of at others' projects.

Since it's free and easy for an admin to do that, why don't they?

Think about that for a bit. The answer has nothing to do with server capacity.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

I dont know reason, John send me to some pages, to write at those pages about my concern for credits.
It is oblivious there is some credits scheme are same for all Boinc projects. ( there must be some rules that admins of project respect)
____________
92*10^^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^⁴⁹⁸¹⁰¹-1 CRUS PRIME
2022202116^¹³¹⁰⁷²+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie!

I dont know reason, John send me to some pages, to write at those pages about my concern for credits.
It is oblivious there is some credits scheme are same for all Boinc projects. ( there must be some rules that admins of project respect)

I think "pissing contest" is the term you're thinking of.

I think "pissing contest" is the term you're thinking of.

We, all of us: using our computers, pay our electrically bill, spend time on Primegrid, and all we get is BOINC credits ( and in this case badges in different colors)

The credit = 3600 was a beta credit just to get the GFN project up and running.

After some calculation we stayed at 2500 but we changed N and we got 3.2*2500

We are still working on a more fair credit because b increase time.


Could we drop this discussion to we have got a better system working.







Lennart

Noticed something nice while trying to figure what the credit for the WR tasks would be. If the factor 3.2 for each n change were to be retained, we'd get:
3.2^3 * 8,000 = 262,144 = 2^18

Was this by design? :)

Maybe a better system, like one Michael mentioned, will be used, however.

Just one more point that may be relevent...

AQUA, when they were active, added credit to long work units on the basis that they were tying up your computer for long periods of time, and gave extra credits as "compensation".

8 or more days is a very long time to tie up a gpu, and I believe some "compensation" consideration should be worked into them. In fact, imho, the longer a wu, the more crefit per hour it should pay (and that includes SOB & PSP too).