Factoring to 87 bits - Page 4

JuanTutors · 2021-10-17, 16:04

Perhaps the right way to prp a 1 billion digit mersenne is to have two people with similar speed computers perform the test simultaneously and then compare results along the way. Then upload results for future workers to do in case one tester quits along the way.

kriesel · 2021-10-17, 22:48

Quote:

Originally Posted by JuanTutors

Perhaps the right way to prp a 1 billion digit mersenne is to have two people with similar speed computers perform the test simultaneously and then compare results along the way. Then upload results for future workers to do in case one tester quits along the way.

No. That would be very wasteful. The correct way is to wait patiently for, or help work toward:

1) Ernst to complete coding PRP proof generation in Mlucas;

2) At least 2-4 months of very active bug finding and reporting effort by multiple testers, and Ernst fixing confirmed issues in the new feature;

3) George or someone to develop code that can accept proof files, and generate cert input files, and perform Cert assignments, beyond the current 64M fft limit of prime95 / mprime, up to ~192M fft length required for 1Gdigit Mersennes;
(Note that there is no indication I'm aware of, that George is on board with creating such an mprime / prime95 or server extension. The mersenne.org server does not currently support 597M proofs, much less 1G+ or 3.32G+)

4) More months of bug finding and fixing for that hypothetical mprime / prime95 extension;

5) Development of hardware, fast enough to be worth trying a gigadigit Mersenne PRP/GEC/proof-generation, that will complete within a year or two from start (About 3-6 times the speed of a Radeon VII GPU). Existing supercomputers probably meet this requirement running Mlucas. But there is no budget for sufficient supercomputer time.

There is nowhere to upload interim PRP residue files (or interim LL residue files) onto project servers.
There is nowhere to upload P-1 factoring results for p>1G onto project servers.
There is nowhere to upload PRP test results for p>1G onto project servers.
Forum attachment size is limited. PRP interim files will be far larger than the limit.
We're regularly finding and fixing bugs in the Mlucas P-1 factoring feature, months after its initial release.

Approx. GHD per primality test:
1Gdigit 703,000.
1G 47,400.
100Mdigit 4,940.
100M 381.
60M 133.
so 1 gigadigit PRP DC takes as long as ~5300. 60M DC. That giant use of resources for an avoidable DC is worth avoiding. We have the number theory. At some point we'll also have the implementation.

kriesel · 2021-10-21, 23:49

Alternatively, and hypothetically:

Ernst develops Cert processing capability in some future version of Mlucas.
Or someone steps up and creates a standalone Cert processing software capable of handling gigadigit Mersennes.

Addressing speed of hardware, someone creates multi-GPU-capable PRP/GEC software. (Imagine n RadeonVIIs working together on one large exponent's PRP, where n = 2 to 5.)

Note, no one has yet credibly proposed doing any of these.

kriesel · 2021-10-27, 18:47

Moderator please rename this thread to "TF on OBD candidates".

clowns789 · 2021-11-01, 05:27

Quote:

Originally Posted by clowns789

I have a 3060 Ti, but going to 92 bits would take about 10 weeks.

Quote: