[QUOTE=alpertron;404995]How do you know that David Campeau was working with the same exponent seven years earlier?[/QUOTE]

that's not what I'm saying. I'm saying it could be all the same person/computer you are claiming outright it's not. your rebuttal was that the result from 2005 was not anonymous. However that has no relevance if the person from 2005 part of the listing was not known at the time of the other anonymous results. only when this is ruled out can we say it's more than 1 computer.

[QUOTE=science_man_88;404996]that's not what I'm saying. I'm saying it could be all the same person/computer you are claiming outright it's not. your rebuttal was that the result from 2005 was not anonymous. However that has no relevance if the person from 2005 part of the listing was not known at the time of the other anonymous results. only when this is ruled out can we say it's more than 1 computer.[/QUOTE]

The old v4 logs show that:
[CODE]1998-06-01 ANONYMOUS NF no factor to 2^62
1998-05-16 ANONYMOUS NF no factor to 2^61[/CODE]
are from different users with different versions of Prime95 (v16 and v15 respectively).

As George mentioned, old P95 versions didn't show the from-to bit levels... the earlier one from May 1998 should have found it no matter what though since even if it was 60-61 and not 0-61 it should have found that factor TJAOI found.

All I know about that result is it was from Prime95 v15 from a user named "TomFakes".

The second one was Prime95 v16 and from user "nuutti"

They only get listed as "anonymous" in the website display because there's no v5 mapping for those users and the only place to pull the actual UID in that case is from the raw log message.

[QUOTE=Madpoo;404997]All I know about that result is it was from Prime95 v15 from a user named "Tom[b][color=red]Fakes[/color][/b]".[/QUOTE]There's our first clue.

If I recall correctly the number of clients working on trial factoring the low bit levels in the range 79.3M - 1000M was fairly limited. Given the significant finds, is it possible to analyse this range using the v4 data and see how many of the original clients turn up in the new data user TJAOI provided?

[QUOTE=retina;405009]There's our first clue.[/QUOTE]

Yeah, I thought about that too. :smile:

Let's break some stuff down on this user:[LIST][*]1940 total "no factor found" entries for TomFakes[*]349 factors have been found for exponents where TomFakes didn't find anything[*]19 of those are definite "misses" by that user (he checked "to 2^61) and the factor found later was between 60-61 bits)[*]1 is a definite miss where the factor was 61-62 bits and TomFakes checked "to 2^62"[*]147 of those had factors < 60 bits and TomFakes only checked "to 2^61" but it's unknown where it started TF at... could have been 2^60, could have been less, we just don't know.[*]182 had a factor found that was > 62 bits, and TomFakes didn't do any testing above that bit level for any of them.[*]The definite "miss" rate is 20 / 1940 = 1.03% ... should have found a factor but didn't.[/LIST]
Does that help? I don't know if anyone would care to repeat the 2^60 - 2^61 (or 2^61 - 2^62 range) for any of those that haven't already been done in that same range by anyone else? Might take some work to put that together since figuring out the exact ranges that an exponent has been factored to is a bit tricky the way things are logged.

Note to George: Maybe the DB needs a bit-mapped column that keeps track of the exact bit levels done for each exponent... I had a weird feeling that some people "skip" bit levels, like if an exponent was done up to 2^62, someone might do more starting at 2^65 or something, leaving the 2^62 - 2^65 range out. It could just be that someone did that and the logs reflecting that don't exist, but there's no way to know. To keep it simple I'd start the bit-map at like 2^50 or something... just assume that's been done... avoid having the new bitmap field by a 100 bit digit. Then it only needs to be 50 bits or whatever.

We could go back through and parse the existing logs to fill that data in, and spot any gaps a lot easier. Anyway, just an idea. :smile:

[QUOTE=tha;405017]If I recall correctly the number of clients working on trial factoring the low bit levels in the range 79.3M - 1000M was fairly limited. Given the significant finds, is it possible to analyse this range using the v4 data and see how many of the original clients turn up in the new data user TJAOI provided?[/QUOTE]

I'm sure that's correct, I was one of the very early Lone Mersenne hunters. Working even lower than that.

However the exponent we are talking about is in the 7M range not 79M+

How many "factor found" results has TomFakes returned? Does his success rate look like bad hardware (lower than usual but finding some) or broken software (finding nothing)?

If TJAOI is going to check the whole range for missing factors is there any point double checking suspect TF results?

Chris

[QUOTE=chris2be8;405042]How many "factor found" results has TomFakes returned? Does his success rate look like bad hardware (lower than usual but finding some) or broken software (finding nothing)?

If TJAOI is going to check the whole range for missing factors is there any point double checking suspect TF results?
[/QUOTE]

Well... that's a very good question and one I should have thought of on my own. Whoops.

The answer is kind of funny... an initial query showed me that, hey, TomFakes has found 130 factors. Well, that's not bad right? Out of nearly 2000 "no factor found".

Then I looked at some of the actual factors being found. Would it surprise anyone else to see an entry like this?
"UID: tomfakes/<xxx>, M7054741 has a factor: 1"

Because I know that surprised me. 103 out of the 130 "has a factor" are where it managed to figure out that all #'s are divisible by 1. Um... good job? How many CPU cycles did it take to figure that out? (they got zero GHz_days credit for those)

So anyway it only *actually* found 27 factors out of all that. That user also did 71 LL tests.

All of those "has a factor: 1" results were from just one machine (this user used several different machines). I wonder if it'd be worth re-checking the ranges for just the "no factor found" results that machine reported?

There are 1770 "no factor found" from that machine, but 315 of them had a factor found later (either a higher bit range or whatever). That leaves 1455 exponents with no known factor, and the work this user did was 2^60 or 2^61 or 2^62 (no known starting bit level).

2^60 = 10 exponents
2^61 = 1010 exponents
2^62 = 435 exponents

These are all in the ~ 3.6M - 7M range ... a good GPU could blaze through those in short order, I imagine?

EDIT: Oh, and I have to remember that someone like TJAOI might have already re-done those... I'd have to check, but that would take more digging.

[QUOTE=Madpoo;405050]
These are all in the ~ 3.6M - 7M range ... a good GPU could blaze through those in short order, I imagine?
[/QUOTE]

I'm always willing to take some worktodo file, if there is one that makes sense taking.

[QUOTE=Madpoo;405020]Does that help? I don't know if anyone would care to repeat the 2^60 - 2^61 (or 2^61 - 2^62 range) for any of those that haven't already been done in that same range by anyone else? Might take some work to put that together since figuring out the exact ranges that an exponent has been factored to is a bit tricky the way things are logged.[/QUOTE]

PM me the list, and I'll run them from 1 to 61 (or 62, if you think it's appropriate).

As an experiment, I did 7024807 on one Titan (of two) on a cg1.4xlarge -- took 16.351s to find the factor.

[QUOTE=chalsall;405056]PM me the list, and I'll run them from 1 to 61 (or 62, if you think it's appropriate).

As an experiment, I did 7024807 on one Titan (of two) on a cg1.4xlarge -- took 16.351s to find the factor.[/QUOTE]

I'm probably a doofus... I couldn't figure out how to attach a file to a PM, so it's here. Everyone but chalsall just ignore. LOL

I generated it to have the same end bit level as what that user reported in.

It's 5 less (all in the 2^60 range) since I noticed that, for some reason, he'd done each of those twice. Weird.

I don't know where a good starting bit level is... I'm sure all of these have been tested to some trivial bit level already but I have no idea so I just used "1" as the starting bit level. I'm pretty sure Prime95 will start them out around 2^52 or something around there.

Feel free to tweak and toy with it to whatever you know works best. I'm not a factoring person so I have no idea. :smile: