[QUOTE=jrk;263193]I am trial-sieving it now with 14e to get the expected specialq range needed.[/QUOTE]

Here, with params:

[code][color=green]# sieve with ggnfs siever 14e on alg side from Q=20M to 102M[/color]
n: 6803747787022489499410807910155630431837414452992393014928087339260361117757308737623381424238689884938205106426429459841345969872513300070476005698207576897418939
# norm 7.291977e-16 alpha -8.871924 e 9.121e-13 rroots 3
skew: 23956310.52
c0: -1553900333513181717543755527226977429566825
c1: 37448916881759468280096265435732185
c2: 8650436181652021797521533543
c3: -282264696045886134641
c4: -6593552669934
c5: 446760
Y0: -27323406066622153187163767265832
Y1: 1101413592433303669
rlim: 55000000
alim: 55000000
lpbr: 30
lpba: 30
mfbr: 60
mfba: 60
rlambda: 2.5
alambda: 2.5[/code]

[QUOTE=jrk;263203]Here, with params:

# sieve with ggnfs siever 14e on alg side from Q=20M to 102
[/QUOTE]

c163 with 14e? Was it a close call (14e/15e) or rather unambiguously 14e?

[QUOTE=Andi47;263206]c163 with 14e? Was it a close call (14e/15e) or rather unambiguously 14e?[/QUOTE]

14e was slightly faster

...but enough to cover the redundancy penalty?

[QUOTE=Batalov;263232]...but enough to cover the redundancy penalty?[/QUOTE]

aq4788:2509 had a harder poly, used 14e and had a redundancy of 23% out of 104M raw relations. The range of specialq needed for this aq4788:2672 poly is smaller and should find near 115M raw relations (overshot a bit more for a smaller matrix) and shouldn't even be as bad as that.

I'd expect that 15e (with appropriate changes to params) would drop the duplication rate to 15%-20%, but would be a little slower. I've seen it done both ways with polys around this size and the difference is slim, you just gotta go with one...

[color=gray](If the range of specialq needed for a comfortable matrix runs past 120M (starting at 20M) using 14e, then I would suggest starting with 15e instead. Now, I put that point at around [u]e = 8.6e-13[/u] in my experience.)[/color]

Is anyone still running ECM on aq4788:2672? From the posts here it looks like we have about 2/3 of t55 done which is probably enough for sieving to start.

I've started sieving and it should be sieved in about 120 hours and completed in about 48 after that. I'm using alim=48 million and 15e, sieving 14.4M - 48M, because I can run matrices fast enough that oversieving isn't so important.

[QUOTE=jrk;263235]Is anyone still running ECM on aq4788:2672? From the posts here it looks like we have about 2/3 of t55 done which is probably enough for sieving to start.[/QUOTE]

I'm up to about 3500 @ 11e7. Since fivemack can do the sieving so fast (what monster machine did you get?!) I'll stop now and devote my wattage to something else.

fivemack recently got a 4 x Opteron 6168 computer, which can pull some pretty interesting results: [url]http://www.mersenneforum.org/showpost.php?p=263000&postcount=56[/url]

With the ever increasing computation power, chances are that in RSALS, we'll have to add the 15e siever, alongside the old trusty 14e, in late 2011 or early 2012.

I reckon I can run a 10M^2 matrix in 72 hours on the big machine, and the sieving for a C170 took about twelve days. My estimates for this job are a little dubious, it may take as much as 150 hours to sieve.

The big machine runs about twice as fast as the rest of my farm put together, uses about half as much electricity as the rest of my farm put together, and cost about 40% more than the rest of my farm put together.

4788.2672 progress report
 

Total sieve time 23917390 CPU-seconds in 161 wall-clock hours

[code]
Wed Jun 15 02:55:13 2011 found 18496407 hash collisions in 109285743 relations
Wed Jun 15 03:11:21 2011 found 16929936 duplicates and 92477844 unique relations
Wed Jun 15 04:48:40 2011 matrix is 5424269 x 5424453 (1654.1 MB) with weight 520693703 (95.99/col)
[/code]

At 08:36: linear algebra completed 1496582 of 5421322 dimensions (27.6%, ETA 9h48m)

Should have factors tonight