2005 Sieving and factoring challenge

[MikeH]

The 2005 sieving and factoring challenge is now well under way.

For anyone that has sieved or P-1 factored in the past this is a quiet call to come back and give it a try again. The playing field has been levelled, the scores (http://www.aooq73.dsl.pipex.com/2005/scores.htm) are based only on finds made during 2005.

If you've never tried sieving or factoring before. Come and give it a try, right now with a couple of lucky factors you'll find yourself in the top three.

The early leader comes primarily (no pun intended) from the P-1 camp, but the number two spot is currently occupied by a serious siever with a lot of very good factors.

My personal efforts are currently not looking too representative of my all time (http://www.aooq73.dsl.pipex.com/scores.htm) performance. I guess that's a polite way of saying I'm looking pretty rubbish right now.

Best of luck to everyone. Hope to see some new and returning faces. :cheers:

[royanee]

I was #1, but I'm not looking so hot now... :(

[vjs]

Not looking so impressive for me either...

8th!!! out of 21 ???

:help:

[maniacken]

Count me in. Ill see what numbers i can knock out in 05.

[hc_grove]

Originally posted by MikeH
The early leader comes primarily (no pun intended) from the P-1 camp,


I do find P-1 factoring more interesting than sieving (the math behind the basic idea is more fun), and the reason I'm currently leading the 2005 challenge is som factors found doing P-1 factoring.

But if you check my all time sieving and factoring scores (http://www.aooq73.dsl.pipex.com/ui/2437.htm) you'll see that I've only scored a little less than 5M points doing P-1 but more than 7.1M doing sieving.

But let's hope this brings along someone with more computing power than me, who can take the #1 spot.

[maddog1]

I know for sure I'll never get even close to the #1 spot, but I find the sieving aspect of the project much more fun than the regular PRP tests.
Since the "normal" reasons for doing this, like helping the project as a whole etc etc have been repeatedly mentioned by MikeH, Vjs and other guys, I'll list my out-of-the-ordinary reasons here, feel free to use them in any advertising of this subproject (points 1-3 might be able to lure some other users in this too) :D

1) You actually FIND something frequently! I was getting pretty bored of crunching for 1-2-3 weeks (depending on PC uptime...) just to see a Result: 3 at the end ;) Here, I can always come back to the PC the next morning and find at least one new fresh factor waiting for me...
2) I hate to see a good fast Athlon being put to shame by some crappy Celerons in the main project stats, SoB main client being the only thing these not-so-fast-at-all Cels are good at doing...
3) Staying in the stats dept, sieving is a good place to be if you want to always make it to the top 100 in the user stats :D
4) This is my tribute to Eratosthenes (http://en.wikipedia.org/wiki/Sieve_of_Eratosthenes) , the great grand father of sieving. I *might* even be the 2nd Greek after him to use this method!
5) You can call me MAD(dog) for this one, but I actually like the total lack of any automated procedures in the operation of the sieving client-I still have fun fiddling with my PC all the time... :crazy:

May all of you guys keep finding some nice high-scoring factors for quite some time in the future! :thumbs:

[vjs]

:cheers: :thumbs:

[MikeH]

...and now with Team score breakdowns all time (http://www.aooq73.dsl.pipex.com/scores_t.htm) and 2005 (http://www.aooq73.dsl.pipex.com/2005/scores_t.htm)

[vjs]

Hey Mike how about another round of archives

Status soon after k=28433 removed (first day of sieving 10 k)

Status prior to prime found for k=28433 (last day of sieving 11 k)

[MikeH]

last day of 11 k (http://www.aooq73.dsl.pipex.com/history/2005_01_03/scores.htm)
first day of 10 k (http://www.aooq73.dsl.pipex.com/history/2005_01_04/scores.htm)

[vjs]

Very nice Mike Thanks, Let hope you have to do this again fairly soon, last 10k, first 9k. :)

[MikeH]

So by pure luck I have crawled my way back to 3rd.


1 hc_grove 910210.75+(32.83) 1+( 17+) 0 ( 10 ) 2 0 0 3.9 27.3/30.3
2 engracio 597570.34+(21.55) 5 ( 210+) 0 ( 109+) 18 0 0 16.9/18.9 42.3/75.0
3+ MikeH 505476.70+(18.23) 0 ( 64+) 0 ( 38+) 5 0 0 35.0 90.0/97.0
4- frodo42 389916.85+(14.06) 0 ( 6+) 0 ( 0 ) 0 0 0 0.1 1.3/1.8
5 OrkunTuncay (Nuri) 139886.64+( 5.05) 0 ( 56+) 0 ( 31+) 3 0 0 7.5 53.0/65.0 With six factors waiting in the wings ready to score in about 9 days time, all I need is for hc_grove and engracio to have an unlucky time and I might have a short visit to the top of the table. :rolleyes:

Great job everyone. Good luck all :cheers:

[engracio]

MikeH,

You just might, you just might MikeH.:) Have not been getting any good hits lately. If nothing else, at least next to hc_grove.

Halfway down from my current reserved ranges and aint made a move upwards in almost a week. :mad:


e:)

[Death]

I see this too late. Late start... damn.

/me rush to reservation thread....

[MikeH]

1+ MikeH 1512044.91+(35.09) 1 ( 85 ) 1 ( 48 ) 7 0 0 35.0 90.0/98.8 :)

[Frodo42]

1+ frodo42 2023117.33+(30.06) 3 ( 18+) 0 ( 2 ) 0 0 0 0.1 1.3/1.8
2- MikeH 1680195.70+(24.97) 4+( 105-) 2+( 53-) 9 0 0 35.0 90.0/98.8

:|party|:

I guess I'd better enjoy this while it lasts ... had a really good factoring streak :)

Btw. wouldn't it be possible to link from http://www.aooq73.dsl.pipex.com/ to the 2005 scores?
Not only (partly maybe) because I want as many as possible to see how extremely good I am :p

[MikeH]

Btw. wouldn't it be possible to link from http://www.aooq73.dsl.pipex.com/ to the 2005 scores? That page needs a lot of updating. I'll try to do it at the weekend. With my current factor drought I think you'll be there for a little while now. My prediction is that the number one spot will be taken by engracio before the end of this month, and when he gets there he's sure going to be difficult to displace.:viking:

[engracio]

MikeH,


I don't know about that MikeH, I'm thinking I moved to the Gobi Desert at the rate I have been going lately. Got lots of factor but most of them were duds. Oh well, maybe next time around. :)

Must be a nice view up there. One day, some day. :)



e:)

[Nuri]

I guess I'll never get a place better than the fifth.... which is nice (for the project).

[Nuri]

Ooops!

8th already.. :(

[cjohnsto]

While not having much power donated, I have suffered a huge gap in usefull factors, my last gap was 226.6G :(
Perhaps that is normal but it is the worst I have ever had.

[MikeH]

1+ MikeH 2715840.96+(21.18) 14 ( 137+) 4 ( 84 ) 32+ 1 0 35.0 98.8/105.8
2- frodo42 2601868.63+(20.29) 3 ( 22 ) 0 ( 2 ) 0 0 0 0.1 1.8/2.8
3 engracio 2423211.67+(18.90) 12+( 560+) 7 ( 313+) 58+ 0 0 16.9/18.9 75.0/105.5 :)
I'm not expecting it to last long, so making the most of it!:cheers:

[Nuri]

Mike, is it possible to conduct a Computational effort (estimated from ranges declared [complete]) estimate for 2005 stats?

[MikeH]

Mike, is it possible to conduct a Computational effort (estimated from ranges declared [complete]) estimate for 2005 stats? Right now Ican't do it because I haven't been recording the dates when ranges are reserved and completed.

I'll have a bit of a think, should be relatively easy to do for new stuff, but don't know about historical data. But then saying that, I do have data that tells me exactly when all the factors in the ranges were submitted, so I should be able to do this.

Need to think though....:idea:

[vjs]

Hey Mike...

Rather than looking at specific ranges times for users etc, etc, etc...

Perhaps there is a much simplier way of calculating simply G/day.

You could use the gaps page to determine how many G we have eliminated between 0T and 2000T or even simply use the 90% T level...

Simply calculate the change per day and a weekly average etc...


Then you could get something like 1,500 G/day sieve effort today and 1,500 G/day (2 week average) or your choice of course.


Then state a conversion... how about ?~25? G/day would be a 1Ghz athlon chip... report the value in 1G athlon years or days ???

[MikeH]

2+ MikeH 14605169.39+(18.38) 58-( 300 ) 23 ( 187+) 465 30 0 35.0 114.3/117.3
3- engracio 14319119.70+(18.02) 75+( 977-) 42+( 574 ) 159+ 10 0 16.9/18.9 120.1/122.1
Won't last long, so need to make the most of it....:)

[maddog1]

I'm perfectly happy with #7, as this was accomplished using just one machine :)
*might* even catch a glimpse of #6 as I am expecting two nice 200K+ factors to come into active window within the next couple of days ;)

[hhh]

Originally posted by maddog1
*might* even catch a glimpse of #6 as I am expecting two nice 200K+ factors to come into active window within the next couple of days ;)

You won't. I have some factoring factors ready, I just need to submit them.:Pokes:
H.

[vjs]

Personally I'm very happy with my position generally do well in the daily and overall, not fighting Mike for #1 anymore but...

I'm just very suprised since I have not done any first pass ranges in quite some time. BUt I have done alot of rechecking with the 991<n<50M dat and found a large number of factors p<100T, they don't score very high at all.

Regardless looks like E is taking the lead with his large amount perviously sieved , He commented to me he is going back to main ranges with the 991<n<50M dat.

I'll probably stick it out in the low-p high-n for a while longer.

How are the factoring people fairing Nuri, HC-Grove etc... ? looks like they are still pulling in the high scores from time to time. How is the factor output compared to before?

[Frodo42]

Originally posted by vjs

How are the factoring people fairing Nuri, HC-Grove etc... ? looks like they are still pulling in the high scores from time to time. How is the factor output compared to before?
I guess I am part of that crowd.

It looks like I am about to loose my #1 spot to either MikeH or engracio ... but I had it for quite some months though ;)

I have my P4 3 GHz still doing factoring and a 2.4 GHz also doing a bit of factoring and some double checking at the same time.

I have been trying to run both some P-1 with low value (1.2-1.4) , some with medium (1.5-2) and some with very high bounds (up to value of 10) ... my feeling is that it produces most factors to run a lot of the low-value tests, but I just can't help myself to do some checks with higher value from time to time (I am starting to discriminate a bit between the diffrent k's ... just for the fun of it).

I do have a feeling that with sieving being as far as it is, the number of factors found have decreased somewhat lately ... but I may also just be on unlucky strike.

[vjs]

My gosh Frodo,

Sorry for not including your name your name it just didn't come to me in a quick response.

I too was curious about the higher bounds, I tried a couple a while back with very high B1 and B2 slightly greater than B1 but not efficiently so.

Last time I factored I used something like B1=20K B2=100K and found a few in that time, this last time it was B1=100K and B2=? Didn't find anything...

So I guess it looks like factory default from prime95 is still the best for factors found. :D


(I am starting to discriminate a bit between the diffrent k's ... just for the fun of it).

deeper for smaller k?

[Frodo42]

Originally posted by vjs

deeper for smaller k?
Actually deeper for 4847, 55459 and 67607.
4847 and 67607 for some reason has always been favorites of mine (don't know why, maybe it's the last digit of seven I like) and 55459 to get a certain amount of tests with high bounds.
A little spin of from this may that I am able to to see how many more factors I find by choosing higher value. But it will take a lot more factors before I am able to say anything from my results.

[vjs]

Well before I had a thing where I use to check what B1,B2 bounds would have been required for every factor that I found through sieving. Generally only those within the active range or P-1'ing range, or those I wound interesting etc.

Example

41911846476041 | 21181*2^21806060

41911846476041-1 = 41911846476040 = 2 ^ 3 x 5 x 10691 x 98 007311

so
B1= 11K
B2= 100M

Couple other p's of factors just for fun (these are all p=~42T) which find factors for n>20M. All p's are p-1 already

41 912584 229100 = 2 ^ 2 x 3 x 5 ^ 2 x 13 x 31 x 346 671499
41 913165 268200 = 2 ^ 3 x 3 x 5 ^ 2 x 23 x 79 x 4877 x 7883
41 917499 090086 = 2 x 20 958749 545043
41 918911 187298 = 2 x 3 x 19 x 157 x 2342 100301
41 920159 687590 = 2 x 3 x 5 x 401 x 2719 x 1 281587
41 921554 700148 = 2 ^ 2 x 3 x 1 449983 x 2 409313

Before I use to think that more factors could be found with a small B1 and then one very large B2. But of course this rational doesn't work b/c the factors are actually small factors (sieve) not large ones (which should be p-1)... hope this makes sence.

Someone said before that the best bet is to spend equal time in stage1 as stage2. If B2 is highly optimized use large bounds if B2 is slow use a large B1.

It's sort of an interesting mental game and the fact people are doing factoring is a very good thing. Garo made a very good write up on factoring a while back.

I just wish a newer faster sieve/factoring program would come out for 64-bit.

[hhh]

I went back into factoring after a few months, and my feeling also is that it becomes harder to find factors; not because there is less probability of finding them, but the tests take longer. The PRP tests take longer, too, though, so I don't know whether it is worth or not. (If not for the scores).

I have 3* P4 2GHz and found 4 factors in a week or two, but one Celeron 1GHz found by sieving 3 factors near 1000000G. Ok, the saved tests are far away, but they are worth a lot of PRPwork, too.

I will take a low p high n sieve range when I finished my factoring work and let it run on the P4 (unless you scream). They don't have much RAM, so it's not a big loss, I think.

What I don't like about factoring, in fact, is that we are looking for factors that we could already have found in a more effective manner, in order to save time for the main effort which has already too much power, compared to sieving. If somebody has good arguments to encourage me, I am more than willing to hear them.

See you, H.

[vjs]

H,

Of couse you can always do what you will with your machine.
What is the sieve speed with sse2 and can you HT with that machine?

I'd be interested in seeing what the sieve speed is under the following conditions.

1. Just sieve (of course use sse2 and 991<n<50M)

2. Sieve with 2 clients running on an HT box

3. Sieve with 1 main client secondpass running.

You may just wish to run two instances of secondpass I certainly think that is a vaild effort for P4's at the moment.

You could also join me in finishing up holepatch but you don't get personal credit and the chances there is a prime there are slim to none according to louie.

[hhh]

The problem is that the possibilities are very restricted. On none of these machines, I can run the normal client (no registry access), and on the slow celeron, both prime95 and sse2 sieve client are quit after a few seconds of crunching. Only cmov works.
On the other ones, I tried only prime95, but only with 100MB in daytime and 170 in the night.
Sieving should work, too.
I have no HT. (The idea in mersenneforum came just like this).

That's why. The crunching speed of the slow Celeron is 200 (kp/s? Not sure about the unit anymore).

What is your conlusion now? I will probably see what I find until the end of the range, calculate a little bit and decide then.
Yours H.

[vjs]

Actually that celeron 1000 is pretty good in sieve my C-1000 gets about 220 kp/s compared to the athlon thunderbird slot a at around 240 kp/s it holds its own.

The reason why sse2 quits is b/c it doesn't have the sse2 extension in the celeron. It basically sits there thinking for a minute then exits right...???? Cmov is the only way to go with that processor.

As for the P4's it's a tough call considering you can't run the normal client your only option really is sieve or factor. 100mb of memory is not bad for P-1 but like Nuri (I think) said a while back factoring is getting slim.

It's not that factoring isn't useful it's just that we have now sieved deep enough that factoring is basically where it should be.

hhh, your pretty informed as far as I see so I think you'll have to draw you own conclusions. I'd personally go with the 1 factor is worth a little less than 2 tests rule. I'd have to guess that a prp test takes about 1 week on those p4 machines, so can you find at least 1 factor every two weeks?

Second what is your sieve speed? I think sieve may actually be optimal with that p4 considering the circumstances.

If you consider...
- We get at least one factor every 100G and on average
(based on a 10 factor per 1000G... it's probably closer to 15 per 1000G)
- You can probably sieve at least 20G per day with that machine
(This translates into 5 days per factor, again you'll probably sieve more)
- That factor on average would be a 11M value
(Based on a 2M<n<20M dat)
- It would take you more than 2.5 days to test a 11M k/n pair once.


Of course I'd suggest you use the 991<n<50M dat with first pass ranges. What do you think??? Does anyone disagree with the above.

I'd also be interested in the actual sieve speed that you get with the P4.

BTW can you reboot that machine and determine if HT is disabled in the bios. I thought all P4's >1.7G had HT. Perhaps that's just xeon's. Also I saw the same thread... interesting running P4 cmov with main client on an HT Box.

[hhh]

If you google for Celeron Hyperthreading, your first hit is that they don't have it, even if it seems that you can enable it in the bios. I am not going to mess up these computers by fuddeling around:p .

As for your calculations, that's what I would have done; It seems to me that factoring is on the threshold to not being effective, at least with not much memory; That's why I wanted to check out sieve.
But anyway, don't take too much care of me, all this can be discussed in a week when I start to play around.
Of course I would use the large dat file.

If we forget a little while about comparision of the efforts, the logic way would be to push sieve to 2^50, then make doublecheck until there is a new software, while catching up with the large dat. Then we can restart considerations of efficiency. If I look at it like this, it makes more sense to me to sieve.

Well, I think we should cook everything on a little lower fire...

H.

[MikeH]

1+ MikeH 15726348.55+(18.90) 60+( 301+) 23 ( 187 ) 465 30 0 35.0 114.3/117.3
2- frodo42 15607198.45+(18.76) 78 ( 28 ) 0 ( 10 ) 0 0 0 0.1 2.8
3 engracio 15123553.91+(18.17) 79+( 977-) 48+( 572 ) 161 10 0 16.9/18.9 120.1/122.1
No-body's more surprised than me, but I'll enjoy it for the short while it exists. ;)

[hhh]

As for the sieving speed of the
P4 Celerons 2GHz, it's crappy: 200kp/s,
just as the
x86 Family6 Model8 Stepping10 , 1GHz, 200kp/s,
with the new dat.
With factoring I have a factor every week, that's fine, I'll stay. I found the argument for factoring I missed before: It's work that cannot wait.
See you, H.

[MikeH]

...and so 2005 (http://www.aooq73.dsl.pipex.com/2005/scores.htm) ends. Well done to engracio for showing us that the sievers were still on top in 2005. My heart goes out to Hadesau who would have been on top had we not paused for reflection and done all that enforced DC work. And a special mention to Joe_O, who whilst finishing a lowly 18th managed to single handedly save 590 DC tests through his hole patching and 50M efforts.

...and so on to 2006 (http://www.aooq73.dsl.pipex.com/2006/scores.htm). I'm sure this year will be the year of the factorers.

[jasong]

Actually deeper for 4847, 55459 and 67607.
4847 and 67607 for some reason has always been favorites of mine (don't know why, maybe it's the last digit of seven I like) and 55459 to get a certain amount of tests with high bounds.
A little spin of from this may that I am able to to see how many more factors I find by choosing higher value. But it will take a lot more factors before I am able to say anything from my results.
Off-topic:

I bought a book on selling once. It had a chapter about prices. If you have an item that's worth $20.00 and you sell it for $19.95, $19.97 and right at $20, the $19.97 price has the statistical advantage for most sales.

Most likely the seven again.

[Joe O]

And a special mention to Joe_O, who whilst finishing a lowly 18th managed to save 590 DC tests through his hole patching and 50M efforts.

With a lot of help from VJS and others. You know who you are, so pat yourselves on the back, and accept my apologies for not listing all of you indivdually.

[vjs]

I think pat's on the back are well deserved all around.

Mike for keeping up the sieve stats page, without Mike we probably wouldn't have a sieve and factor sub-projects. Or at least very limited efforts without stats. Keep up the good work.

Joe and Chuck for their efforts and forsite with the 2^50 limit quickly approaching.

Joe for maintaining the 50M dat and showing us all the light with high n-sieve.

E for managing to sieve himself to the top even with a job change and a move.

Hades_Au my special thanks for getting involved, you have stolen first place in sieve for TPR.

I'd also like to point out Matt's hard work in the auto-sieve reservation system. It's so well done that we almost forget it exists.

-----------------------------------------

As for 2006, I can't really agree that factoring will take top scores this year. At least not yet, see Aliens post, secondpass tests factors will score in the 80k's in addition factoring must find smooth factors >1P now. 06 will be interesting.

Keep up all of the good work guys lets finish off this sieve <2^50 that should be one short term goal.