Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
1)
Message boards :
Number crunching :
PPS (LLR) Milestones, Goals, and Progress
(Message 62333)
Posted 3769 days ago by bdodson*
2012 Goals - Progress update
...
Proth Prime Search Update
...
100<k<500 to 2.2M - (currently at n=2M)
500<k<1200 to 2.2M - (currently at n~1.6M)
1200<k<10000 to 1.03M - ongoing leading edge n=1.03M) DONE
---------------------------------------------------------
Shown here is the status right now.
We will not reach all goals but K=3 - K<32 is done.
32<k<100 & 100<k<500 will not reach the goals :( ....
I suggest
...
100<k<500 to 2.2M
500<k<1200 to 2.0M - (currently at n~1.65M)[/color][/b]
1200<k<10000 to 1.1M If all other goals are reached we continue here.
Any thoughts ???
Lennart
Updated 2012-11-28 02:30 UTC
And Updated again 25 January. Not sure whether anyone else
is wondering where the PPS (LLR) is running during the Tour (= Feb),
but WU's are being sent from two ranges, the extended search,
1200 < k < 10000 (current c. n = (1.047...)M) and from the original search
k < 1200 in the range 100 < k < 500 (current c. n = 1.65M).
Current status is from the PRPNet Stats on the Tour, where the count
is four ("4") primes with 100 < k < 500 and forty-six ("46") primes from the extended
range k > 1200. There's a nontrivial difference in size (and compute-time/WU)
between n = 1.047M at 315K-digits and n = 1.65M at 471K-digits, and also
lots more k's in the extended range. It isn't clear to me whether there's
a difference in the number of WU's being sent (extended -vs- non-extended);
perhaps somewhat more in the extended range, but not in proportion to the
number of k's. A plausible hypothesis for the 4 -vs- 46 primes found (by PPS/LLR
so far in Feb) is that 315K-digit numbers have a better chance of being prime
than 471K-digit integers (relative "density of primes"); and, if anything, the
four larger primes would be fewer yet if WU's went out in proportion to the
number of k's [that's 400 = 500-100, -vs- 8800 = 10000 - 1200]. Uhm,
that's a "definitely" there are way more 471K-digit candidates being checked
than 400-to-8800 or 1-to-22. By just the naive count; someone else probably
has a more accurate count, but even 1-to-10 undercounts the number of 471K-digit
LLR-tests.
Still waiting for my first, so too much time for analysis! -Bruce*
|
2)
Message boards :
Number crunching :
PPS (LLR) Milestones, Goals, and Progress
(Message 62023)
Posted 3780 days ago by bdodson*
2012 Goals - Progress update
Good progress is being made in advancing PPS (LLR) towards its 2012 goals.
Only about one month left. Let's keep up the good work!!!
Work will continue for the 1200-10000 ranges up to n=1.1M. SGS is online and we want to make sure PPS is high enough to remain on the Top 5000.
Thank you again to everyone helping out with this effort. :)
Proth Prime Search Update
...
100<k<500 to 2.2M - (currently at n=2M)
500<k<1200 to 2.2M - (currently at n~1.6M)
1200<k<10000 to 1.03M - ongoing leading edge n=1.03M) DONE
---------------------------------------------------------
Shown here is the status right now.
We will not reach all goals but K=3 - K<32 is done.
32<k<100 & 100<k<500 will not reach the goals :(
We will not start anything above 6M We need to sieve more.
I suggest
...
100<k<500 to 2.2M
500<k<1200 to 2.0M - (currently at n~1.65M)[/color][/b]
1200<k<10000 to 1.1M If all other goals are reached we continue here.
Any thoughts ???
Lennart
Updated 2012-11-28 02:30 UTC
If I'm reading correctly, 100 < k < 500 and 500 < k < 1200 are well
above the SGS range; and it is only
1200<k<10000 to 1.1M
that's under 389K-digits. We seem to be working on WU's in this
range, with my new one
9175*2^1038858+1 312,732 (decimal)
If I'm reading my logarithms correctly,
k = 9200, n = 1.1M has c. 331K-digits
k = 9200, n = 1.2M has c. 361K-digits
k = 9200, n = 1.3M has c. 391K-digits
and k = 1200, n = 1.29K is just barely under SGS range. Looks like
n = 1.2905K gets us above SGS range. Meanwhile, with n = 1.0388M
up around 2200th of the Top 5K these primes won't be getting bumped
off any time soon. Guess I'd regard getting up to the goal of n = 1.1M
as a worthwhile step in a sequence 1.1M, 1.2M, 1.291M.
Regards, Bruce*
|
3)
Message boards :
Number crunching :
PPS (LLR) Milestones, Goals, and Progress
(Message 60843)
Posted 3807 days ago by bdodson*
[size=18]
...
500<k<1200 to 2.2M - (currently at n~1.6M)
...
---------------------------------------------------------
Shown here is the status right now.
We will not reach all goals but K=3 - K<32 is done.
32<k<100 & 100<k<500 will not reach the goals :(
Been a while since the most recent update. Looks like I'm
the Double Checker on
935 ยท 2^1707129 + 1, digits 513901 -bdodson*
...
We will not start anything above 6M We need to sieve more.
I suggest
k=3 to 10M
k5-k7 suspended to we have a good sievefile.
k=9 to 6M
10<k<32 to 3M
32<k<100 to 2.5M
100<k<500 to 2.2M
500<k<1200 to 2.0M - (currently at n~1.65M)[/color][/b]
1200<k<10000 to 1.1M If all other goals are reached we continue here.
Any thoughts ???
Lennart
Updated 2012-11-28 02:30 UTC
|
4)
Message boards :
Generalized Fermat Prime Search :
GFN primes now shown in Prime lists
(Message 58404)
Posted 3881 days ago by bdodson*
So, you're saying GFN numbers shouldn't be on the list? If so, what makes them worthy of a unique exclusion?
Or that all primes that aren't Proth numbers shouldn't be on the list? Then we should exclude our Riesel, Woodall, Sierpinski Base 5, Riesel Base 5, Factorial, Primorial, and Generalized Cullen primes?
Or just primes that don't have a scientific or mathematical purpose? That pretty much excludes all the tens of thousands of PPS primes we've found, since their only real purpose is to put people on the top 5000 list.
I'm not saying any of these things. Although, for the record, I did
find a Fermat divisor; which pretty much made my year, prime-findingwise.
Congratulations on the GFN with base 75898 and exponent 524288.
Probably your success gave some needed encouragement for the
subsequent finds with the same base and larger exponents. I'd count
that as a plausible purpose for the finding of your GFN; proof-of-concept.
Was this one of the ones just before the Turing Challenge? I'm not able
to find an official statement for these two primes; is the other one still
pending?
I'm afraid I still don't get your point, unless that point is "Please don't ignore the other projects." If that's it, rest assured, they're not being ignored. Over the last week, I've added close to ten thousand primes onto the list, and at least 90% are PPS. Hopefully, I'll be able to add in many more (tens of thousands), and almost all of them are from our PPS searches.
Uhm, most of these don't make the top5000, presumably --- or are they
mostly re-replacing the bottom 100? No that doesn't fit the 72-hour
reports on the Prime Pages. At most 100s there.
Perhaps I'm just lamenting having the top5 placed up in near-inaccessible
range. Scott looks likely to bump you from 4th to 5th, so I'd need to
bump my score up from 12k past 21k --- by which time there'll most likely
be a bunch more pushing all 3 of us down. As in, the next two will move
the top5 scores up past 31k. So that's doubling then tripling 12k, and
inaccessible; outright. So rather than a top20 reflecting successes of
several of Primegrid's projects, it's in line to become a top20 all_GFN list.
That's something unique-new to the GFN project, the scoring formula
pre-GFN no longer does what the formula used to do --- giving several
projects a chance to be represented in the top20. I haven't thought
about whether that's worth addressing, but I was wondering whether
the Primegrid Admin had. As I've already mentioned, I'll be interested
to see what happens when we break the World Record, replacing
GIMPS. Maybe a bunch of new contributors, a massive outflow from
mersenne-hunters to GFN-hunters? Server issues here? Interesting.
Thanks for the replies. -Bruce*
|
5)
Message boards :
Generalized Fermat Prime Search :
GFN primes now shown in Prime lists
(Message 58383)
Posted 3881 days ago by bdodson*
...but having checked, virtually all of your score is from the one
GFN, and it is the smallest of your four primes!
No, it's not.
You missed the part above (in the very first message in this thread) where I mentioned that the server script that displays the primes doesn't work properly for GFN (or, indeed, any prime with B>2), and the length displayed on the web page is inaccurate.
...
I understand why you would be upset about a "change of focus" here, but everything you're saying is based upon the erroneous lengths displayed on the webpage. There is, in fact, no bias, nor change of focus. The GFN numbers are simply HUGE numbers.
Shortly after posting I recalled having read something about the length
on the webpage not being accurate. I ought to have checked the
announcements, in any case. Sorry.
That accounts for the first paragraph of my post; my mistake. I've
read and re-read the reasons why the current "Top Prime Finders"
rankings are consistent with the formulas established before the
recent GFN successes. Perhaps those of us looking for Proth primes
will get used to discounting top20 scores from people with "primes
found" counts in single digits (easy to implement, too).
I'll be interested to see how a World Record set with a GFN will be
viewed. It is of course true that Mersenne primes have been largest
among top10 primes largely due to hardware considerations: that
they're easier to compute. I do recall when
391581*2^216193-1 65,087 1989 Amdahl 1200
briefly replaced the Mersenne's as largest; and the interest more
recently when one of the SoB's broke into the top10 --- ah, looks
like SoB10 is still holding on at 10th. Part of the interest is that each
of these have a history. So far as I can tell, GFN's will be the first record
set by a prime that has no interest other than being largest; and
specifically associated with GPU specific hardware (pending subsequent
cpu GFN World record success, if there is to be one).
Meanwhile, I hope that at least portions of the Primegrid elite (you
know who you are!) make some effort to retain the interest of those
that aren't yet interested in GFN primes. (As in "why did you want
one of those primes?" Yes, they're huge alright.)
-Bruce*
|
6)
Message boards :
Generalized Fermat Prime Search :
GFN primes now shown in Prime lists
(Message 58352)
Posted 3882 days ago by bdodson*
...the GFN thread, rather
than crunching or PPS! -Bruce* (with 4 idle tesla C2030/2050's)
It's the PRPNet (mostly) and GFN (both BOINC and PRPNet) primes that are being added, so this was posted in the GFN and PSA topics. Those are the two groups who have been asking for this to occur, in some cases for years.
I was just getting ready to congratulate _you_ on your four large
primes; but having checked, virtually all of your score is from the one
GFN, and it is the smallest of your four primes! Could you remind those
of us still toiling away on Proth Primes how the scoring works?
You mention that some of the GFN finders (and/or doublecheckers) used
cpu rather than gpu; so how does a cpu GFN test compare with a
llr for Proth or Sophie/Twin? Hard to imagine that the differential in
scoring is matched by a comparison of computational effort. We are
discussing computation, rather than hardware? Most of my own
Primegrid score comes from when our C2050s were running (not 2030's,
and it's the 2070's that have yet to score a single point, sigh, and Thanks
to Scott for the PM re PSA). I do understand when a GPU score is based
on a computation that runs (way) better than the corresponding cpu
computation; and have compared PPS sieving on the two platforms.
Also, how do the GFN scores compare with the single prime scores
that were in the top10 before the scoring revision? As I was working
my way up the score list from 10th to 2nd (err, I mean, 3rd to 10th to
9th, moving ahead of Lostboy's score for the 2M-digit 321-prime) there
were representatives from several of Primegrid projects. Spinner@ had
a 2M-digit Cullen, Gesker a 1.9M Cullen; and the double-checkers weren't
in the top10. There was also Mumps[MM] 1.8M-digit 321-prime.
I'm not usually in favor of Primegrid visitors to the Contests whining
about scoring, before returning to whatever project they were computing
on aside from contest credits. There is, and has been, sustained interest
by people from computational number theory and crypto in a range of
Primegrid projects. Mersennes hold a special place in the history of
primality testing and in the related factoring problems. Likewise Fermat
numbers. I'm not sure that Primegrid's Admin ought to turn the project
into one focused of finding GFN primes, to the point of questionable
scoring decisions. I'd be reasonably happy to have you remove a few
of the questions; but two years in PSA isn't a very long time frame by
RSA key-braking standards, much less primality and factoring standards.
-Bruce*
|
7)
Message boards :
Generalized Fermat Prime Search :
GFN primes now shown in Prime lists
(Message 58316)
Posted 3883 days ago by bdodson*
However, regardless of what type of processor is used to do the calculations, I *do* think that the top of the list will probably be dominated by GFN primes for the foreseeable future. Like Mersenne primes, they're (relatively) easy to test.
Guess that's why this explanation appears in the GFN thread, rather
than crunching or PPS! -Bruce* (with 4 idle tesla C2030/2050's)
|
8)
Message boards :
Generalized Fermat Prime Search :
GFN primes now shown in Prime lists
(Message 58309)
Posted 3883 days ago by bdodson*
All of PrimeGrid's GFN primes at N=32768, 65536, 262144, and 524288 will now show up in your personal prime list. This is the list you see when you click on any of the "primes found" numbers on your main account page.
Please note that the decimal length shown is incorrect and the actual length of the prime is substantially longer. This is a known bug and will be fixed. Eventually. :)
The primes will also be reflected in the standings on the "Top Prime Finders" page, although it may take up to an hour until some of the primes show up there.
This includes all GFN primes found at PrimeGrid whether via BOINC or PRPNet.
This would explain why I was about to be 3 points away from 2nd,
but have overnight dropped to 10th! With the exception of Scott, it's
now a GFN only list; and no longer a CPU-effort measure, instead open
to GPU-only. Just wondering whether that was the intended effect.
-bdodson*
|
9)
Message boards :
Number crunching :
PPS (LLR) Milestones, Goals, and Progress
(Message 58139)
Posted 3891 days ago by bdodson*
...
I'm seeing
7555*2^1010048+1 is not prime.
4869*2^1010097+1 is not prime.
for tasks that have just been sent, as in
4 Oct 2012 | 13:30:46 UTC 4 Oct 2012 | 13:47:48 UTC
With c. 27 hours to the start of the challenge, looks like
we'll be competing in n=1.01. -Bruce*
Looks like the leading edge at/near the end of the challenge
was in 1.017M
Is prime? 9125*2^1017239+1 is not prime.
...
413603656 290601 6 Oct 2012 | 15:17:36 UTC 6 Oct 2012 | 17:53:22 UTC Aborted by user 0.00 0.00 --- PPS (LLR) v1.01
413848949 187389 6 Oct 2012 | 17:56:38 UTC 7 Oct 2012 | 2:53:40 UTC Completed and validated 8,374.76 8,323.37 11.41 PPS (LLR) v6.13
shows 1.017239M sent at 15:17, then resent at 17:56, just short of
the 18:00 UTC close. And for one just past the close
Is prime? 6699*2^1017827+1 is not prime.
...
413686351 291102 6 Oct 2012 | 18:27:50 UTC 6 Oct 2012 | 22:23:40 UTC Aborted by user 0.00 0.00 --- PPS (LLR) v6.13
413853911 187389 6 Oct 2012 | 18:28:12 UTC 7 Oct 2012 | 5:11:13 UTC Completed and validated 8,568.37 8,507.76 9.89 PPS (LLR) v6.13
413872377 239220 6 Oct 2012 | 22:23:56 UTC 7 Oct 2012 | 0:37:04 UTC Completed and validated 551.37 456.55 9.89 PPS (LLR) v1.01
with quorum 2, sent at 18:27 UTC, for 1.0178M
Server seems to have gotten past the load from the contest; I'm
seeing 180 tasks to be validated now, under 100 a few minutes ago;
and "primes found" back up (rather than "down for contest").
-Bruce* (@peppert*: nice cores!)
|
10)
Message boards :
Number crunching :
PPS (LLR) Milestones, Goals, and Progress
(Message 57995)
Posted 3894 days ago by bdodson*
We should have changed to a smaller k but I will keep the low n in the challenge.
It is only 24hr.
Lennart
Uhm; that's "low n" meaning large k, and continuing with
1200<k<10000 to 1.03M - ongoing (trailing edge n=998493; leading edge n=1.005M)
if I'm reading correctly. The smaller k's have n at/above 2M.
I'm seeing
7555*2^1010048+1 is not prime.
4869*2^1010097+1 is not prime.
for tasks that have just been sent, as in
4 Oct 2012 | 13:30:46 UTC 4 Oct 2012 | 13:47:48 UTC
With c. 27 hours to the start of the challenge, looks like
we'll be competing in n=1.01. -Bruce*
|
Next 10 posts
|
