ECM users - version 30.9/30.10 (see post#168)

[Prime95]

Originally Posted by lycorn

Summary of results:
The number of curves to run was given by GMP-ECM.

Exponent: 4567

B1 B2 runtime curves to run
110 M 1000 * B1 950 + 17.9 25849
110 M 500 * B1 950 + 11.9 29306
110 M 200 * B1 950 + 7.1 35419
110 M 105 * B1 950 + 5.3 40485
110 M 100 * B1 950 + 110.7 14396 (actual B2 = 28217 *B1, computed by Prime95)

For larger values of B2, stage 2 runtime would grow accordingly:

110 M 1.5e13 950 + 293.4 11285
110 M 3.0e13 950 + 500.7 10211
110 M 6.0e13 950 + 793 9307

Best from this table is: 1.5e13
11285 × ( 950 + 293.4 ) = 14,031,769

This is not best configuration for this computer though. The fact that only one core is used during stage 1 is a big issue.

[lycorn]

Originally Posted by Prime95

The fact that only one core is used during stage 1 is a big issue.

True, but that´s because the FFT is very small, isn´t it? What I am doing now is running 2 workers, 2 threads each. The global throughput is much better.

[Prime95]

Best setup for you is likely 4 workers an MaxHighMemWorkers set to 2

[ATH]

Here are some tests with the old Core i7-5960X with 24000MB RAM for Prime95, only using 1 core for both Stage 1 + 2, since I was not sure what number of cores was best to test with in Stage 2, and if multiple cores did much difference at these ultra low FFTs.
The ones with "*" is the auto chosen B2 values.

Is there any value in doing more tests with this old CPU ? Like T50, T45, etc. or tests with more cores in Stage2 or different amounts of RAM?

				Actual				GMP-ECM
		B1	B2	B2		B2/B1	RAM(MB)	curves	Curve time	Total Txx time
------------------------------------------------------------------------------------------------------
M2719-T65	800M	8e11	8.2978e11	1037	3124	116870	5563s + 61s	182577 h
M2719-T65	800M	8e12	8.045e12	10057	11772	81086	5563s + 221s	130278 h
*M2719-T65	800M		5.04656e13	63082	23542	61656	5563s + 684s	109465 h
M2719-T65	800M	8e13	8.411e13	105138	23542	57270	5563s + 1035s	104963 h
M2719-T65	800M	2.4e14	2.51416e14	314270	21644	48983	5563s + 1694s	 98742 h
M2719-T65	800M	8e14	8.411e14	1051375	22023	40963	5563s + 3487s	102976 h

M6073-T60	260M	2.6e11	2.6796e11	1031	4686	62444	3598s + 74s	63693 h
M6073-T60	260M	2.6e12	3.1039e12	11938	14570	42901	3598s + 306s	46524 h
*M6073-T60	260M		1.50228e13	57780	22957	32933	3598s + 774s	39995 h
M6073-T60	260M	2.6e13	2.7935e13	107443	23036	30158	3598s + 1141s	39700 h
M6073-T60	260M	7.8e13	8.411e13	323500	22682	25965	3598s + 2478s	43823 h
M6073-T60	260M	2.6e14	2.73764e14	1052939	19984	21807	3598s + 4927s	51640 h

M10303-T55	110M	1.1e11	1.2416e11	1129	4362	25075	2228s + 78s	16062 h
M10303-T55	110M	1.1e12	1.1174e12	10158	14451	16758	2228s + 294s	11740 h
*M10303-T55	110M		4.34546e12	39504	22945	13511	2228s + 689s	10947 h
M10303-T55	110M	1.1e13	1.2018e13	109257	22913	11712	2228s + 1150s	10990 h
M10303-T55	110M	3.3e13	3.3644e13	305855	21090	10000	2228s + 2229s	12381 h
M10303-T55	110M	1.1e14	1.2456e14	1128691	23596	8373	2228s + 5225s	17334 h

[Prime95]

Originally Posted by ATH

Is there any value in doing more tests with this old CPU ? Like T50, T45, etc. or tests with more cores in Stage2 or different amounts of RAM?

Thanks for the great data! From that I see that the "prime95 guessed optimal B2/B1" is not terribly bad. We need to wait for a prime95 that is fully optimized in stage 2. Then gather similar data for a variety of machines and exponents before tuning prime95's B2/B1 selection.

BTW, interesting to see the optimal B2/B1 is dropping as exponents get larger and t# gets lower. My earlier statement of B2/B1 > 1000 not being valuable may have been correct in the M900000 area I tried but is clearly wrong for these small exponents.

[Stef42]

I like to do ECM, so gave 30.9b1 a go, got a factor on the very first exponent!

Factor: 352867573711979454617014177 / (ECM curve 242, B1=50000, B2=38914785, Sigma=861408202433943)

M3981161

[axn]

Another success:

ECM found a factor in curve #1, stage #2
Sigma=3154775547629108, B1=250000, B2=235419030.
UID: axn/c1, M903407 has a factor: 1066347780167947196737336563937 (ECM curve 1, B1=250000, B2=235419030)

Group order is not particularly impressive: 2^5 · 3^3 · 5^4 · 11 · 919 · 5039 · 139199 · 149759 · 1859617. Could've been found by regular bounds

[Prime95]

Originally Posted by Stef42

got a factor on the very first exponent!

Well done. 242 curves showed great patience!

[ixfd64]

Am I correct to assume this is based on the new stage 2 implementation that doesn't support the Brent–Suyama extension?

[Prime95]

Originally Posted by ixfd64

Am I correct to assume this is based on the new stage 2 implementation that doesn't support the Brent–Suyama extension?

Yes.

[axn]

Couple of hits in quick succession:

ECM found a factor in curve #17, stage #2
Sigma=3692827876263630, B1=250000, B2=235419030.
UID: axn/c2, M905621 has a factor: 11139426256343765204839823 (ECM curve 17, B1=250000, B2=235419030)

Group Order: 2^3 · 3 · 89 · 4349 · 73819 · 75437 · 215337467


ECM found a factor in curve #14, stage #2
Sigma=6883129626396355, B1=250000, B2=235419030.
UID: axn/c3, M905629 has a factor: 875812073686928536845781855983559 (ECM curve 14, B1=250000, B2=235419030)

Group Order: 2^4 · 3^2 · 13 · 673 · 37337 · 39397 · 43961 · 157733 · 68154937