mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Information & Answers (https://www.mersenneforum.org/forumdisplay.php?f=38)
-   -   What Bounds to choose, and what are Bounds (https://www.mersenneforum.org/showthread.php?t=22105)

144 2017-03-04 12:19

What Bounds to choose, and what are Bounds
 
2 Attachment(s)
I have read about Bounds, but it's too advanced for me, and I'm not sure how to scale Bounds to the exponent.

The Optimal Bounds I got auto assigned when I requested a P-1 test for M332255197 are B1=4660000 and B2=4660000 (same).

I have i5-6600k cpu and 16GB ram if it's important.
And what is the number right for "Number to " in P-1 settings.

[ATTACH]15710[/ATTACH]

[ATTACH]15711[/ATTACH]

If you could, explain in general terms, I don't even have an intermediate understanding of these numbers and terms.

144 2017-03-04 13:29

I changed the max memory in settings to 8192 day and 10240 night and get B1=3070000, B2=85192500 bounds now (posting this here since I can't edit my first post).

VictordeHolland 2017-03-04 18:13

The second stage (B2) requires more memory than stage 1 (B1) to run efficient. If you dont allocate enough memory it will only run stage 1 (B1=B2).

Prime95 calculates bounds before starting the assignment. Taking in consideration the amount of Trial Factoring done and the chance of finding a factor (thus saving 2 LL test). Bounds can change depending on the processor and/or architecture.

Usually giving Prime95 1GB+ should allow it to run a decent stage 2. You can also use the maxhighmemoryworkers setting to allow only 1 or 2 stage 2 runs at the same time, so you can still use the machine for other things (or run them at night when you're not using the machine).

Edit:
You do realise 100milion digit numbers (332M) are not for the faint hearted? Its your choice what to do with your computing power/electricity, but people sometimes dont realise how big/long those assignments can be. Nothing wrong with P-1 at the current wavefront (80M)..

144 2017-03-05 23:03

Thanks. I guess I'll use suggested Bounds for now. I do have about 10GB available at any time, so I hope it will get utilized.
I have 2 tests running at a time, so setting "MaxHighMemWorkers" to 1 could be useful.

Yeah, I started realizing that 100m digit LL tests are not something I will undertake (estimated ~550 days on my CPU), only maybe if I can get it to work on my RX 480 AMD GPU (not sure it's supported (only mfakto supports AMD atm?)).

LaurV 2017-03-06 05:40

[QUOTE=144;454312] only maybe if I can get it to work on my RX 480 AMD GPU (not sure it's supported (only mfakto supports AMD atm?)).[/QUOTE]
There is a clLucas that supports LL test on AMD cards, you can give it a try. It is not as fast as cudaLucas, however, except for exponents where a FFT power of two is optimal to use, where the two packages have about the same performance. Your mileage may vary with the card, of course. For LL, RX480 has about the performance of a GTX580 on average, and it is faster where the FFT is power of two. See [URL="http://www.mersenne.ca/cudalucas.php"]here[/URL] a comparison.

Lorenzo 2017-03-15 13:36

[QUOTE=144;454312]Yeah, I started realizing that 100m digit LL tests are not something I will undertake (estimated ~550 days on my CPU)[/QUOTE]

hmmm, weird. Regarding to your CPU LL test must be take ~120 days.


All times are UTC. The time now is 11:26.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.