Prime95 v30.3

[ATH]

I got the new message on "Manual Results" page:

Your software is outdated.
Your L-L result has been accepted for now, but please upgrade your software to at least v30.3 as soon as possible.

But I was handing in LL DC results. I understand the message for new LL results but for LL DC?
Do we want everyone to switch to 30.3 even for LL DC, so we are doing PRP + CERT negating the original LL results?

[kriesel]

Originally Posted by ATH

I understand the message for new LL results but for LL DC?
Do we want everyone to switch to 30.3 even for LL DC, so we are doing PRP + CERT negating the original LL results?

Maybe. On average it will take ~1.04 LL to confirm a first LL, due to error rate, but only ~1.01 PRP equivalent with the effort of a CERT.

I wonder what it does with CUDALucas results or Gpuowl LLDC. PRP via Gpuowl is not an option for some gpu models, whose drivers don't support a high enough OpenCL version. It would be a shame to push the older Quadros, Teslas, etc with decent SP/DP ratio down to TF or recyclers. (The models most likely to have the OpenCL incompatibility are also the most likely to have limited ram for P-1 in CUDAPm1. And CUDAPm1 is still alpha software with issues.)

[ZFR]

Originally Posted by Prime95

I'd say the behavior is pretty close to a bug. It certainly is unexpected.

The hours-per-day setting affects very little. Some estimated completion dates, work assignments if you use "What makes sense" work preference.

I'll look into what I need to fix in the next release.

Thanks. For the time being I'll just keep hours-per day greater than 12 then.

[James Heinrich]

Originally Posted by ATH

Do we want everyone to switch to 30.3 even for LL DC, so we are doing PRP + CERT negating the original LL results?

I think that would be good. A PRP+Cert will clear the exponent just as well as a successful LL-DC would, and be quicker if the DC doesn't match.

There will no doubt be lots of old installations that won't get updated quickly, if ever. I would suggest that pre-30 installations soon (if not immediately) be relegated to DC-only work, even if they asked for first-time LL assignments. I know George has a much more lenient and tolerant approach to users who refuse to upgrade than I would so it might not happen, but I think we need to encourage as many users as possible to upgrade as soon as possible to avoid unneccesary effort being wasted on two checks when one will suffice.

[Uncwilly]

Originally Posted by ATH

Do we want everyone to switch to 30.3 even for LL DC, so we are doing PRP + CERT negating the original LL results?

A broader install base of v30 will also spread the load of running the certs around. This will help. Let slower machines do the certs while faster machines do more of the PRP work.

[Prime95]

Originally Posted by ATH

I got the new message on "Manual Results" page:



But I was handing in LL DC results. I understand the message for new LL results but for LL DC?
Do we want everyone to switch to 30.3 even for LL DC, so we are doing PRP + CERT negating the original LL results?

I think encouraging the upgrade is a good idea regardless.

I'm ambivalent about DCes being done as LL or PRP+CERT. Cost is about the same. LL-DC provides feedback on the original computer which is of some minor value.

[Uncwilly]

Originally Posted by Prime95

I think encouraging the upgrade is a good idea regardless.

I'm ambivalent about DCes being done as LL or PRP+CERT. Cost is about the same. LL-DC provides feedback on the original computer which is of some minor value.

How about LL-DC's for assignments less than 2 years old. Anything much longer and the data of a bad LL is not actionable.

Also, start giving any machine that is returning first time LL's, 25% DC's then in 2 months move it to 35%, then 45%, etc. If they want to avoid the DC's they need to upgrade.

[richs]

I searched through this thread and was not able to find how to stop getting cert work with the new version. One of my older laptops that I upgraded to the latest version yesterday has been receiving cert work but keeps getting the message "Error getting CERT starting value." The program tries twice then aborts the assignment. I would prefer to stop getting cert work on this laptop.

[James Heinrich]

Originally Posted by richs

I searched through this thread and was not able to find how to stop getting cert work with the new version.

Prime95 GUI has the checkbox under Test > Worker Windows... > [ ] Get occasional proof certification work.
Alternately, in local.txt put CertWork=0 to disable Cert work.

[jwnutter]

Originally Posted by James Heinrich

I think that would be good. A PRP+Cert will clear the exponent just as well as a successful LL-DC would, and be quicker if the DC doesn't match.

There will no doubt be lots of old installations that won't get updated quickly, if ever. I would suggest that pre-30 installations soon (if not immediately) be relegated to DC-only work, even if they asked for first-time LL assignments. I know George has a much more lenient and tolerant approach to users who refuse to upgrade than I would so it might not happen, but I think we need to encourage as many users as possible to upgrade as soon as possible to avoid unneccesary effort being wasted on two checks when one will suffice.

Your suggestion wouldn't impact me in any way and I can't disagree, but if the name of the game is maximizing total available clock cycles to the GIMPS project a forced upgrade may do more to harm the cause than help. But I would agree, there is a trade off between maximizing the total number of available clock cycles and the efficiency of each cycle. Regardless, encouraging users to upgrade as soon as possible would be of great benefit.

[James Heinrich]

Originally Posted by jwnutter

.. a forced upgrade may do more to harm the cause than help.

I doubt there would ever be any "forced" upgrade (as far as I know there's still 15+ year old v24 installations chugging along contributing their little bit to the project). But an encouraged upgrade would make the cycles of those who do upgrade more useful.