GPU to 72 status...

[storm5510]

Also, while this was created under Linux, it should be possible to make it work under Windows without too much effort. Feedback on what is needed to do so appreciated.

I just pulled in the spider and changed what was needed. I am using Strawberry Perl under Windows 10 Pro and have written a few short scripts with it. So, I understand parts of the spider. There should not be any language differences which would keep it from running with Strawberry. I will simply have to try it and see what it does. Just to be safe, I plan to make a copy of my results and place it in another folder.

[storm5510]

I tried it and here is what it did:

20200727_003523 INFO: Submission spider starting...
20200727_003523 INFO: Attempting to log into Primenet. This can take a little while...
Use of uninitialized value $Line in pattern match (m//) at C:\mfaktc\spider.pl line 147.

The third line above, there were hundreds of these. I had to Ctrl-C to stop it. This may be because I had to comment the "exit" command to keep the window open so I could read it. I had the mfaktc console open and ran the spider from there. It flashed a second console for an instant. I needed a workaround.

Line 142: my ($Line) = $_;

Line 147: if ($Line =~ /no factor for (M\d*)/ || $Line =~ /(M\d*) has a factor/ 
      || $Line =~ /(M\d*) no factor from/) {

Line 142 is the declaration. It is different from what Strawberry would use. If it has no immediate value, then Strawberry might expect "my $Line;"

Line 147 tries to use $Line but cannot. "Uninitialized" it says. This would point back to line 142. Perhaps Strawberry does not understand the declaration.

[KEP]

Originally Posted by chalsall

Also, tangentially... Although Reb and KEP demonstrated empirically that BOINC could provide a serious amount of GPU resources, I simply haven't had the time to take the GPU72 BOINC system past alpha.

I'm glad to read that you are going BOINC. SRBase is still doing TF but it is breadth first and there is no way to change that. Is there none of the excellent guys that have set up projects previously who can help you get BOINC going, well before november?

At november SRBase will most likely have cleared a big chunck of the 73-74 bit level and next year, I expect (if the firepower remains) that SRBase will have cleared all n's to at least 76 bits (hopefully more). With your BOINC project there will at least be a breadth first and much more emminent wavefront first Really looking forward to see your BOINC project

An advice in regards to BOINC, remember badges as seen at Primegrid and remember to have enough badges, such that you can maintain the badge chasing users

[chalsall]

Originally Posted by KEP

Is there none of the excellent guys that have set up projects previously who can help you get BOINC going, well before november?

If there are, please step forward!

I'd provide an independent virtual machine, and handle the GPU72 side of the IPC. The VM, Software stack, IP, DNS, "trust", etc is already set up. I just need someone who's done BOINC before to set up the work-types and the badges. (And then extensive beta testing, of course.)

Originally Posted by KEP

An advice in regards to BOINC, remember badges as seen at Primegrid and remember to have enough badges, such that you can maintain the badge chasing users

Appreciate the knowledge. And yup, already modeled. Also, the credits need to be modeled to reward wave-front work, while at the same time not being inflationary.

[chalsall]

Originally Posted by storm5510

Line 142 is the declaration. It is different from what Strawberry would use. If it has no immediate value, then Strawberry might expect "my $Line;"

Line 147 tries to use $Line but cannot. "Uninitialized" it says. This would point back to line 142. Perhaps Strawberry does not understand the declaration.

Hmmm... Perhaps... Or, $_ doesn't have a value. Can't determine without looking at the full script, and possibly running with test data.

I don't have any time at the moment to drill down. But I'll program to look at that over the weekend on my Wincows development box.

[storm5510]

Originally Posted by chalsall

Hmmm... Perhaps... Or, $_ doesn't have a value. Can't determine without looking at the full script, and possibly running with test data...

I tried changing the declaration to my $Line; Doing so had no affect. Perl can be more than a bit ambiguous with its error reporting. Just because it says there is an error on line x, it might not be on that particular line, but somewhere above. I spent an hour looking at that small area of the script. I didn't see anything obvious.

The very top line, appearing as a comment, references Linux folders. I didn't change it. If it is just a comment, then it should have no affect. Another item, appearing in (Windows) Strawberry scripts, is $|=1; What this does, I haven't a clue. It appears below the use statements.

[KEP]

Originally Posted by chalsall

Appreciate the knowledge. And yup, already modeled. Also, the credits need to be modeled to reward wave-front work, while at the same time not being inflationary.

Primegrid uses these "bonus" steps to push the interest in the conjectures with very long workunits:

This project has a 10% long job credit bonus and a 10% conjecture credit bonus. (Extended Sierpinski Problem)

This project has a 35% long job credit bonus and a 10% conjecture credit bonus. (Prime Sierpinski Problem)

This project has a 50% long job credit bonus and a 10% conjecture credit bonus. (Seventeen Or Bust)

This project has a 10% conjecture credit bonus. (Sierpinski Riesel base 5)

This project has a 10% long job credit bonus and a 10% conjecture credit bonus. (The Riesel Problem)

Maybe these direct copy paste from PG, can offer you a hint to non inflated wavefront credit. The reason behind the "inflated" credit, was to stimulate users to offer ressources they might otherwise spend on non conjecture work or a lot of small n testing. It sure did seem to help the conjectures with these bonuses and I sure do expect the much needed 77-78 bit work would benefit from a 50-60% long duration bonus If you have any questions or need any suggestions or experienced advice, feel free to reach out.

[chalsall]

Originally Posted by KEP

If you have any questions or need any suggestions or experienced advice, feel free to reach out.

Copy. Thanks.

[bayanne]

Can't seem to grab any exponents from the 332000000 range and up at present, although there seem to be plenty available.

Can you advise?

[bayanne]

Originally Posted by bayanne

Can't seem to grab any exponents from the 332000000 range and up at present, although there seem to be plenty available.

Can you advise?

Now been able to grab 77 bit exponents

[chalsall]

Originally Posted by bayanne

Now been able to grab 77 bit exponents

Yup. The work to 76 is exhausted, so 77 is the next "cheapest" available WU.

The web-based UI was updated shortly before the 76 work was fully assigned. The Colab select statement is FactTo ascending, Exponent ascending, so it should have given out the last score of work to 76, and then automatically moved to 77.

BTW... Since very few people were actually going to 81, I've set GPU72 to release 332M candidates at 79 bits. Those who are serious about this kind of work and are using Colab might consider choosing "LMH Depth First", to get them off our books.