GPU72 and BOINC a match made in ....

[rebirther]

Originally Posted by KEP

I support running everything above 96.83M to 120M to 78 bit. Let's hope that we can use BOINC to push the wavefront of even Category 3 away from "just in time" to actually "sufficient"/"ahead"/"far far ahead"/"never gonna be able to catch us"

Reb, I suggest, that we for the time needed, take up following battleplan:

1. Everything n>96.83M at 77 bit to 78 bit
2. Everything at 76 bit to 77 bit, then everything that remains n>96.83M from 77 bit to 78 bit
3. Everything at 75 bit to 76 bit, then everything that remains at 76 bit to 77 bit and then everything that remains n>96.83M from 77 bit to 78 bit

I know that you and I would both like to go "breadth first" and most likely we can do that from n>110M to n<=150M (maybe n<=200M), once we get far enough ahead of any wavefront. Can you agree on the above suggestion?

hmm, ok, I wish if we can fix the apps, is there anyone who can compile the mfaktc mac app? I could still need a tester on this opencl AMD linux app if anyone is running it.

[bayanne]

Ask The Judger, he arranged for one to be compiled for me.

[chalsall]

Originally Posted by rebirther

ok, I will add 10k for 77-78 in the next batch so every one is happy then I will go back to 73-74

And the first batch are off!

[Uncwilly]

Originally Posted by Uncwilly

The BOINC entry on the PrimeNet Top TF. They went up 10 places in the last hour.

90	SRBase	70,643.912	7,926	111	↑639	↑609	↑618	↑105

Their attempt vs success ratio looks right.

And they are in the top 20. It looks like the will move up another 4 or 5 before sun up in London.

19	SRBase	741,830.335	48,763	656	↑705	↑686	↑688	↑3

[pinhodecarlos]

There’s only sun down in London not sun up...lol

[chalsall]

Originally Posted by KEP

Reb, I suggest, that we for the time needed, take up following battleplan... Can you agree on the above suggestion?

So, the work to 78 is beginning to be submitted. Sweet!

Rather than do what you've suggested KEP, may I suggest a more targeted approach?

I have set up the "super-secret fetch URL" to prioritize assignments based on the needs of the various wavefronts. So, Reb, if you next fetch 10,000 going to 75, and then once those mostly complete do the next batch to 76, then 77, and then finally 78.

This will give breathing room for Cats 2, 3 and 4 (Cats 0 and 1 are fine for the moment, and the P-1'ers will be shortly). We can then examine where we might need another run or two. Then breadth-first will be fine again (for a while).

This is quite impressive stuff. Why didn't we do this years ago???

P.S. Now I'm going to have to do a bunch of work on the reports and graphs to handle this new normal. Not that I'm complaining...

[pinhodecarlos]

Originally Posted by lalera

hi,
i think that credit should be given for computing power
for mfaktc the credit you get should be 4 to 5 times higher (estimated)

Agreed. It’s all about credits at the end, not project love. One reaction has been posted on the forum by a guy who has 13 RTX 2080 cards complaining about credit per wu.
I would add my GPU laptop but earning 7200 points per day is not worthy, prefer to run other project.

This is a difficult position. Think there was a stats page showing credits per wu per BOINC project but I can’t remember where. Would be nice to see what’s SRBase position regarding other GPU projects and adjust credit accordingly, to get closer to the highest payers.

Here’s more. One year ago NumberFields project introduced a GPU client with fixed credit to align CPU and GPU apps even if the latter was 30x faster. One year after power deployed there has dramatically decreased.

[KEP]

Originally Posted by chalsall

So, the work to 78 is beginning to be submitted. Sweet!

Rather than do what you've suggested KEP, may I suggest a more targeted approach?

I have set up the "super-secret fetch URL" to prioritize assignments based on the needs of the various wavefronts. So, Reb, if you next fetch 10,000 going to 75, and then once those mostly complete do the next batch to 76, then 77, and then finally 78.

This will give breathing room for Cats 2, 3 and 4 (Cats 0 and 1 are fine for the moment, and the P-1'ers will be shortly). We can then examine where we might need another run or two. Then breadth-first will be fine again (for a while).

This is quite impressive stuff. Why didn't we do this years ago???

P.S. Now I'm going to have to do a bunch of work on the reports and graphs to handle this new normal. Not that I'm complaining...

That is a good suggestion, with a targeted approach. But as pinhodecarlos notes, there is complaints about these long tasks and the credit given at BOINC. I'm not running a GPU under BOINC at the moment, so I can't tell if the credit is fair or not, but it seems that we might loose some users, to these high bit leveltests. Maybe your suggestion, of a targeted approach will suit those wanting to run short and those wanting to run long workunits. At least taking 10000 to 75, should get no one complaining - except maybe the credithunters.

To be honest I don't get it, why most projects don't use BOINC at all. It really shouldn't be hard to set up BOINC and make things working, if you have already made all the initial programming of independent software. Thankfully now GIMPs is partially running on BOINC and it appears that in a few months (maybe weeks) all wavefronts will be way behind Trial Factoring to optimum depth.

So now a question, is cat 2, 3 and 4 actually doing work that is poorly trial factored and with low or no P-1 at all?

[kracker]

I just noticed you're still using mfakto 0.14 - I would highly recommend 0.15, even though it's technically a pre-release there hasn't been any (new) issues...

[rebirther]

Originally Posted by kracker

I just noticed you're still using mfakto 0.14 - I would highly recommend 0.15, even though it's technically a pre-release there hasn't been any (new) issues...

We could also need a compiled linux app and mac.

[chalsall]

Originally Posted by KEP

That is a good suggestion, with a targeted approach. But as pinhodecarlos notes, there is complaints about these long tasks and the credit given at BOINC. I'm not running a GPU under BOINC at the moment, so I can't tell if the credit is fair or not, but it seems that we might loose some users, to these high bit leveltests. Maybe your suggestion, of a targeted approach will suit those wanting to run short and those wanting to run long workunits. At least taking 10000 to 75, should get no one complaining - except maybe the credithunters.

Frankly, I find it somewhat interesting that we're constantly told that BOINC will be the great savior for GIMPS, to stay ahead of the FC'ing wavefronts. But then we're told "oh, we don't want to do any of the heavy lifting. Takes too long...

This is the nature of the work.

To put on the table, perhaps going to 78 is a bit too far (literally). It is "optional", but perhaps not achievable. Not the end of the world -- even if no TF'ing was going on, GIMPS will still find the next MP. It will just take ~5% longer.

Originally Posted by KEP

To be honest I don't get it, why most projects don't use BOINC at all. It really shouldn't be hard to set up BOINC and make things working, if you have already made all the initial programming of independent software. Thankfully now GIMPs is partially running on BOINC and it appears that in a few months (maybe weeks) all wavefronts will be way behind Trial Factoring to optimum depth.

You have to understand the history of GIMPS. It is the longest still-running distributed computing project. It has generated a fairly strong community of mathematicians, programmers, and general geeks, who enjoy interesting "driving problems". And, understand the (clearly defined) goal(s).

BOINC seems to be a very different culture. Less focused on clearly defined goals, and more on getting credits on ranking boards in a false economy. No disrespect intended by that statement; whatever rocks one's boat.

Originally Posted by KEP

So now a question, is cat 2, 3 and 4 actually doing work that is poorly trial factored and with low or no P-1 at all?

Because several "big guns" stepped up, all the Cats are now receiving candidates close to optimally TF'ed (read: 77 bits). Cat 4 (known as the "churners" because very few assignments are actually completed) are receiving candidates without P-1 done.

Somewhat ironically, less than 24 hours after Reb took 10,000 candidates at 77 to go up to 78, Ryan Propper started doing his massive P-1'ing thing again. So I'm having to release candidates TF'ed at "only" 76 bits for the P-1'ers to work. No problem -- we'll bring the survivors back in to TF to at least 77 before re-releasing back for the FC'ers.

Lastly... If the run-time is really such an issue for BOINC, why not take LaurV up on his offer to provide a version of mfaktc which breaks the bigger jobs down into sub-jobs?