Quote:
Originally Posted by VBCurtis
Thanks for confirming, Ryan!
T70 is often represented by curves at B1=2.9e9 or so. If that level is complete, then T75 curves are what is useful now that's B1=7e9 or bigger. That's one curve per core per day range.... yeesh.
I'll give a curve at B1=1e10 a try, and see what ecm v reports for B2, memory use, and # of curves to a T70 or T75.

I will run some tests using AVXECM. Scaled tests, at B1=7e6, show that firststage curve throughput is about 2.4x larger than GMPECM.
Code:
echo "2^12771"  ../../ecm704linux/ecm v 7000000
GMPECM 7.0.4 [configured with GMP 6.2.0, enableasmredc] [ECM]
Tuned for x86_64/k8/params.h
Input number is 2^12771 (385 digits)
Using special division for factor of 2^12771
Using B1=7000000, B2=17125579390, polynomial Dickson(12), sigma=0:12224056895737954441
dF=16384, k=6, d=158340, d2=11, i0=34
Expected number of curves to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
167 1024 7351 60402 558826 5744532 6.5e+07 7.8e+08 1.1e+10 2.8e+11
Step 1 took 67016ms
vs.
Code:
./yafu "ecm(2^12771,8)" v v B1ecm 7000000
10/22/20 15:17:47 v1.35beta @ cpu, System/Build Info:
Using GMPECM 7.0, Powered by GMP 6.2.0
detected Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz
detected L1 = 32768 bytes, L2 = 28835840 bytes, CL = 64 bytes
measured cpu frequency ~= 2500.116400
using 1 random witnesses for RabinMiller PRP checks
===============================================================
======= Welcome to YAFU (Yet Another Factoring Utility) =======
======= bbuhrow@gmail.com =======
======= Type help at any time, or quit to quit =======
===============================================================
cached 664579 primes. pmax = 9999991
>> process id is 141455
commencing parallel ecm on 2601983048666099770481310081841021384653815561816676201329778087600902014918340074503059860433081046210605403488570251947845891562080866227034976651419330190731032377347305086443295837415395887618239855136922452802923419286887119716740625346109565072933087221327790207134604146257063901166556207972729700461767055550785130256674608872183239507219512717434046725178680177638925792182271
ECM has been configured with DIGITBITS = 52, VECLEN = 8, GMP_LIMB_BITS = 64
Choosing MAXBITS = 1456, NWORDS = 28, NBLOCKS = 7 based on input size 1277
linesieve took 0.022753 seconds
cached 5761455 primes < 99999989
Input has 1277 bits, using 1 threads (8 curves/thread)
Processing in batches of 100000000 primes
Initialization took 0.1013 seconds.
Building curves took 0.0007 seconds., B2=100*B1
commencing Stage 1 @ prime 2
Stage 1 took 219.2920 seconds
So, 219.3/8 = 27.4 sec/curve versus 67 sec/curve.
AVXECM stage 2 took 118 sec, so 14.75 sec/curve, but these are just the standard continuation up to 100*B1.
It should scale linearly up to 7e9. But I've never run AVXECM with B1 anywhere close to that large. My bet is it crashes... but I'll test and see what happens.