Which is better for sieving C153, lpba 29 or 30?

[tgrdy]

I am using the GGNFS with Brian Gladman's factmsieve.py script.
http://mersenneforum.org/showthread.php?t=12981

To factor a C153 semiprime number, by default it uses lpba 29.
It gives out for a the estimated minimum relations 53 MB (53914715).
I tested, the lpba 30 will speed up 1.7x of the lpba 29. But the script still gives out the same estimated minimum relations. Is this the error estimation by the script?
Generally, The bigger lpba needs more relations,
For lpba 30 , It should need 2x relations as of the lpba 29,
be 2X53 MB = 106 MB ?

I still think the lpba 29 is better than lpba30, for a C153.

done some sieve test with gnfs-lasieve4I14e ,the speed I tested:
lpbr: 29
lpba: 29
mfbr: 58
mfba: 58
(0.13192 sec/rel)

lpbr: 30
lpba: 30
mfbr: 60
mfba: 60
(0.07729 sec/rel)


The poly file I used:

name: r508
n: 701942719692285972017668322577259510159702012577691761277297697988435148980783132399346322593063755616723648504118576010809485370794566876962551921464611
skew: 464934.73 
# norm 1.72e+21
c5: 16830240
c4: -20538938733520
c3: -23051975433694226846
c2: 2894552516408845349131547
c1: 1284600519908070681512922426696
c0: -164330075544484612166847223627265397
# alpha -6.31
Y1: 29238293171566853
Y0: -133058399830223448481474832174
# Murphy_E 3.41e-12
# M 546710741549512951898622587351174022163915197937425276678519394975048846546962014251322873483566005197375340795121090644515856049855969059861273577273869
type: gnfs
rlim: 24400000
alim: 24400000
lpbr: 30
lpba: 30
mfbr: 60
mfba: 60
rlambda: 2.6
alambda: 2.6
qintsize: 100000
q0:12000000

[FactorEyes]

Your conclusions are correct: 29-bit large primes.

Don't sweat the minimum threshold for filtering in the script. It has little connection to the actual number of (unique) relations you'll need, which is around 45 million for a factorization of that size.

[debrouxl]

Sieve using 29-bit large primes, until you have 55M raw relations at the very least.