[QUOTE=chalsall;374192]
Not meaning to blow inappropriate sunshine. But what you and Oliver (et al) do is (IMO) quite impressive.[/QUOTE]

+1 I agree.

[QUOTE=James Heinrich;374203]I don't know about on PrimeNet since I'm not all that comfortable with database interactions there, but I'd be happy to make such data available in raw and aggregated form on mersenne.ca[/QUOTE]

Maybe you could get a list of exponents with factors "close to the beginning" so to say and make a list? Then you could run the list which would take a little of time, outputting the console to text? Just a (possibly stupid) idea.

[QUOTE=chalsall;374204]LOL... If you could "make it so", is would be appreciated and useful.[/QUOTE]I can and will, once there is data to play with. Well, I guess I can start working on it but it would be easiest if I had some idea what eventual form the proposed new benchmark mode in mfakt_ would take.

[QUOTE=kracker;374210]Maybe you could get a list of exponents with factors "close to the beginning" so to say and make a list? Then you could run the list which would take a little of time, outputting the console to text? Just a (possibly stupid) idea.[/QUOTE]I'm not sure I follow? :unsure:
[i](thinks about it some more)[/i]
Oh, you mean a list of exponents with factors that would be found in the first (few) class(es) in mfakt_ and use that as my own benchmark profile. Hmm. I should be able to do that. Actually about a year ago I dredged up a list of a few thousand (or maybe it was million) exponents with easy-to-find factors for Bdot. I'll have to dig up that list, or better-yet, the code used to generate that list. For the purpose of analyzing throughput it would probably be best to [i]not[/i] have a factor in the first class, but maybe 2-3 classes in. In the "real" benchmark the presence of a factor wouldn't be required, the exponent could simply be aborted after running for a few classes (whatever is sufficient to get a stable throughput estimate).

[QUOTE=James Heinrich;374214]
I'm not sure I follow? :unsure:
[i](thinks about it some more)[/i]
Oh, you mean a list of exponents with factors that would be found in the first (few) class(es) in mfakt_ and use that as my own benchmark profile. Hmm. I should be able to do that. Actually about a year ago I dredged up a list of a few thousand (or maybe it was million) exponents with easy-to-find factors for Bdot. I'll have to dig up that list, or better-yet, the code used to generate that list. For the purpose of analyzing throughput it would probably be best to [i]not[/i] have a factor in the first class, but maybe 2-3 classes in. In the "real" benchmark the presence of a factor wouldn't be required, the exponent could simply be aborted after running for a few classes (whatever is sufficient to get a stable throughput estimate).[/QUOTE]

Yes... that was exactly what I was thinking. :smile:

Bdot did add a lot of self-tests for -st2 in 0.12, see [URL="https://github.com/Bdot42/mfakto/commit/60f984af31dd1430a6cd640c50887d520f11ef87"]this[/URL] and [URL="https://raw.githubusercontent.com/Bdot42/mfakto/master/src/selftest-data.h"]this[/URL] :razz:

[QUOTE=kracker;374218]Bdot did add a lot of self-tests for -st2 in 0.12[/QUOTE]Ah, well, there you go then. He even blames me for them. :smile:
I guess it was closer to 2 years ago that I did that.

I have returned. Grabbing some factoring jobs and back to crunching.

[QUOTE=James Heinrich;374228]Ah, well, there you go then. He even blames me for them. :smile:
I guess it was closer to 2 years ago that I did that.[/QUOTE]

Well, mfaktc apparently added it as well. (0.18)
[code]
- new commandline option: "-st2" runs a even longer selftest with *new*
testcases
[/code]

[QUOTE=James Heinrich;374214]a list of exponents with factors that would be found in the first (few) class(es) in mfakt_[/QUOTE]I spent the last few days going through 100-million factors and tagging them with the mfaktc class (out of 4620) where the factor would be found. Actually calculating the class is trivial, took not more than 10 minutes. The other dozen hours was wrangling the database to make a new column and stuff the calculated value into 100+ million records. :smile:
The upside is that I can now easily pull together a list of factors that will be found in any particular class. Not entirely sure how this is useful, but I'll try making a list as [i]kracker[/i] suggested for benchmarking.

LLTF Category 4...
 

Just so everyone knows...

We are going to start having difficulty maintaining feeding the Category 4 "churners" with all candidates TF'ed to 74 again -- at least for a while.

Thus, I have instructed "Spidy" to stop reclaiming candidates already TFed to 73.

Additionally, MISFIT users who use the "Let GPU72 Decide" option are now being given assignments to 73 instead of 74. These will be kept by GPU72 as part of its "rip-cord", and only released if anything below 73 is about to be released by Primenet for LL'ing or P-1'ing (or, of course, if someone then takes them up to 74).

I am hoping this will only be needed for a month or so. And, as I said above, while ~1,000 candidates are requested per day in the Cat 4 range, only ~72 per day are actually completed. Or, in other words, we should be able to reclaim most of them again in the future, and TF appropriately (possibly once they enter the Cat 3 range).

As always, thanks to everyone contributing fire-power. And for any who can bring additional fire-power to bear, it would be welcome, needed and appreciated. :smile:

(BTW, we're looking good in the DCTF Cat 4 range. Pretty much steady-state with the fire-power we have working there currently.)

I will load up with LLTF, then. I've done a few days of DCTF, but those will clear by this evening some time.

[QUOTE=chalsall;374603]We are going to start having difficulty maintaining feeding the Category 4 "churners" with all candidates TF'ed to 74 again -- at least for a while.[/QUOTE]

What happened? Did we just fall behind?

[QUOTE=TheMawn;374617]What happened? Did we just fall behind?[/QUOTE]

probably Oliver, we don't have his 20,000 GHz/days anymore :smile: