Cunningham ECM efforts

[VBCurtis]

Originally Posted by frmky

2,2194L and M are using lims of 300M (pushing the memory limit to the edge of 1GB/core but also crossing 2^28 for a small runtime hit).[...]

I think the best way to fit a larger lim into the required memory footprint is to stick to 250 or 260 on the 3LP side, and boost the lim on the 2LP side as high as memory-limits allow.

I'd go 260/340 or 250/350 and expect both yield and speed to be better than 300/300. However, I think you're reaching a job size where 3LP on both sides may be nearly as fast as the typical 2/3 settings, at least given that lims used are so small.

If you'd like to post the poly for your next job, I'm happy to test a variety of mfb choices for your 35/36 LP setting; perhaps I can find a '3LP on both sides' setting that is faster than what you're otherwise planning to use.

[frmky]

Originally Posted by VBCurtis

I'm happy to test a variety of mfb choices for your 35/36 LP setting; perhaps I can find a '3LP on both sides' setting that is faster than what you're otherwise planning to use.

2,2206L would use

n: 385391677907023068816875399702625816359161842943133261369673832051936437247070622270782933158404692445296119872430400383813398898811140980240524745812750729966969876527198830673858429779606449065686186811518577347751947370710959080728712731829
skew: 1.54407
type: snfs
c6: 1
c5: 0
c4: 0
c3: -2
c2: 0
c1: 0
c0: 2
Y1: 1
Y0: -24519928653854221733733552434404946937899825954937634816
rlim: 300000000
alim: 300000000
lpbr: 36
lpba: 35
mfbr: 105
mfba: 71
rlambda: 3.9
alambda: 2.8

I'm planning to add a couple easy ones before queueing this one so there's some time to play with it.

[pinhodecarlos]

Easy ones... you wouldn't say that a few years ago...lol

I suppose you are talking about 2,1497+ SNFS 300 and 2,1497- SNFS 300. Glad you are doing these ones, at least I get bored of running the more difficult ones, just because I have a very slow computer and I do like to see that status page update more often. Anyway, with the overall CPU momentum available at the moment it is wise to do these quicker ones.

[frmky]

Originally Posted by pinhodecarlos

Easy ones... you wouldn't say that a few years ago...lol

... at least I get bored of running the more difficult ones

Good point. The labels "easy" and "difficult" are relative to the computing power available.

[VBCurtis]

Originally Posted by frmky

2,2206L would use

n: 385391677907023068816875399702625816359161842943133261369673832051936437247070622270782933158404692445296119872430400383813398898811140980240524745812750729966969876527198830673858429779606449065686186811518577347751947370710959080728712731829
skew: 1.54407
type: snfs
c6: 1
c5: 0
c4: 0
c3: -2
c2: 0
c1: 0
c0: 2
Y1: 1
Y0: -24519928653854221733733552434404946937899825954937634816
rlim: 300000000
alim: 300000000
lpbr: 36
lpba: 35
mfbr: 105
mfba: 71
rlambda: 3.9
alambda: 2.8

I'm planning to add a couple easy ones before queueing this one so there's some time to play with it.

I've tested a bunch of alternative params, all at Q=500M so far.
3LP on both sides is good for 40% better yield, but at the expense of 60% longer sec/rel. Since you're not running out of Q to sieve, this seems unhelpful- but I'll test sec/rel at the projected "last" Q, as it's possible that 3LP on both sides for small Q (like 200-400M) may be a net benefit still.

Here are the fastest params I've found so far:

rlim: 250000000
alim: 350000000
lpbr: 36
lpba: 35
mfbr: 105
mfba: 70
rlambda: 3.85
alambda: 2.55

These yield about 2% better than the ones quoted above, but at a speed nearly 20% faster: 0.532 sec/rel vs 0.645 sec/rel.
I've tested only at Q=500M so far, so next time I have some free time I'll test a more full range of Q to make sure the new params are faster everywhere. I'll also test a couple other small variations on these params with 2LP on A side.

[frmky]

Originally Posted by VBCurtis

These yield about 2% better than the ones quoted above, but at a speed nearly 20% faster: 0.532 sec/rel vs 0.645 sec/rel.
I've tested only at Q=500M so far, so next time I have some free time I'll test a more full range of Q to make sure the new params are faster everywhere. I'll also test a couple other small variations on these params with 2LP on A side.

Thanks! Amazingly, we're about done with the three smaller ones already and it's time to queue this one. It's a relatively small change in parameters, so I'm happy to just go with it and see what happens.

[VBCurtis]

I'm giving exams this week, so I can't try other tweaks to get more speed. I hope this works (better)!

[swellman]

2,2194M was factored today by NFS@Home. Very nice!

2,2206L is currently sieving, personally I’m hoping for it to be factored by mid March but it’s not a race.

Yoyo@Home’s throughput is proving to a bit higher than NFS@Home so I’m just going to throw the rest of the Gang of 31 into Yoyo’s queue. Latest estimate for Yoyo to complete all base-2 (1987) work is NLT Halloween 2024. YMMV.

[pinhodecarlos]

Originally Posted by VBCurtis

I'm giving exams this week, so I can't try other tweaks to get more speed. I hope this works (better)!

I've been to post this a couple of weeks ago but your tweaked parameters are indeed fast. I've never seen my old laptop running thiese Wu's in less than 3200 seconds. Haven't normalised past Wu's vs new Wu's (for same conditions, range etc) but boinc points output is higher than before, definitely sure.
Thank you and we'll done!

Regarding Yoyo, was wondering what his plans are for future since one of his apps will be completed soon. Will the server feed or force the idle clients with sieving and ECM apps?! If all clients have ticked to run all apps for sure there will be a boost on ECM side. Only Yoyo to confirm.

[pinhodecarlos]

I believe we shall see an increase of ECM activity on Yoyo. One of his projects is about to end, tiny amount of work to be done. Not sure what would be for our case the impact of more CPU deployed but overall all ECM projects might receive a little nice boost.

[Tyler Busby]

Sorry if this is the wrong place to ask, but I saw this thread pop up on the new posts and figured I'd ask here, is there a current record of ECM done for various Cunningham numbers? Or a general rules for what can realistically be expected to have been performed?

I've attempted to fill out this page's ECM status of Cunningham numbers, but for higher bases (or large n, for that matter) I'm not certain enough to make up data, especially if it seems no factors have been found via ECM (>30 digit seemingly non-algebraic factors). FWIW, I've checked all the places listed in the "Sources" section.

If there are any ECM figures that seem inaccurate on that page, let me know and I can update them. I mostly just want to make sure I'm not wasting cpu time if I attempt to factor Cunningham numbers starting at the work listed there.