Opinions: Should I set P-1 "tests saved" to 9?

JuanTutors · 2019-08-02, 17:17

This is partially an opinion request post and a timing question.

When I was assigned a first time PRP of a 100M mersenne, the "tests saved" setting was set to 2 as usual. Just to see how long it would take, I set it to 9, and my percent completed went from 18.75% to 3.97%, and my (B1,B2) went from (2 560 000,58 880 000) to (12 575 000,421 262 500). Does anyone have an estimate of how long this second one will take, both in overall time and compared to the first one?

For reference, my daytime memory is set to 16000 and nighttime memory is set to 25000, and this laptop did a (successful!) DC on M52271179 at 2.5 ms/iter. I tried to do a proper benchmark/timing to print out so I can copy/paste it here, but couldn't figure it out.

I know there is a "P-1 for life" group out there, and I'm not about that life, but finding a factor is quite definitive and saves the need of both a first time and DC test. It's hard to not use up my 32 GB ram on this laptop

But OTOH, throughput is a pretty solid measure of properly used time.

petrw1 · 2019-08-02, 17:33

What did the expected success rate change from/to?

https://www.mersenne.ca/prob.php
will tell you expected success rate and also total GHZ Days.
If you know your daily rate you can calculate the days to complete.

You should have that from the DC you just finished.
Though in my experience P1 is less efficient than DC.

JuanTutors · 2019-08-02, 17:41

Quote:

Originally Posted by petrw1

What did the expected success rate change from/to?

https://www.mersenne.ca/prob.php
will tell you expected success rate and also total GHZ Days.
If you know your daily rate you can calculate the days to complete.

You should have that from the DC you just finished.
Though in my experience P1 is less efficient than DC.

Expected success rate went from to 3.53% to 5.8%. According to that page it's 121 GHz-days vs 751 GHz-days. //EDIT: It looks like I am doing 103.9 GHz-days in 1.9 days, so 54.68 GHz-days/day if I'm reading it right. So I'm talking 2.1 days vs 13.7 days to complete P1 factoring.

Besides saving tests, is there any other benefit to actually having those factors, if in fact they are found?

kriesel · 2019-08-02, 19:41

For production use, more than 2 tests saved is generally regarded as wasted time. Except on some gpus, that might fall short of the gpu72 bounds, a nudge upward by using 3 may be merited. I'm in the process of benchmarking prime95 P-1 on test_saved=2 sampling across the mersenne.org exponent spectrum, and there are indications it also defaults to less than the gpu72 goals with 8GB allocated, so perhaps would benefit from an upward nudge or more RAM allocated. See https://www.mersenneforum.org/showpost.php?p=503892&postcount=3
for scattered data points up to 801M currently.

JuanTutors · 2019-08-03, 02:04

Quote:

Originally Posted by kriesel

For production use, more than 2 tests saved is generally regarded as wasted time. Except on some gpus, that might fall short of the gpu72 bounds, a nudge upward by using 3 may be merited. I'm in the process of benchmarking prime95 P-1 on test_saved=2 sampling across the mersenne.org exponent spectrum, and there are indications it also defaults to less than the gpu72 goals with 8GB allocated, so perhaps would benefit from an upward nudge or more RAM allocated. See https://www.mersenneforum.org/showpost.php?p=503892&postcount=3
for scattered data points up to 801M currently.

Do you recommend nudging it up to 3, based on your current intuition?

VBCurtis · 2019-08-03, 09:07

We like having factors a bit more than having two matching residues, so if you don't mind the slightly inefficient use of computer time to have a better chance at a factor, boost it to 3.
I would do so myself, because I like finding factors (but not because it will save time in the long run).

The B1 stage should scale linearly; that is, if B1 rises by a factor of 5, stage 1 should take 5x longer. Dunno about stage 2.

2019-08-02, 17:17	#1
JuanTutors Mar 2004 540₁₀ Posts	Opinions: Should I set P-1 "tests saved" to 9? This is partially an opinion request post and a timing question. When I was assigned a first time PRP of a 100M mersenne, the "tests saved" setting was set to 2 as usual. Just to see how long it would take, I set it to 9, and my percent completed went from 18.75% to 3.97%, and my (B1,B2) went from (2 560 000,58 880 000) to (12 575 000,421 262 500). Does anyone have an estimate of how long this second one will take, both in overall time and compared to the first one? For reference, my daytime memory is set to 16000 and nighttime memory is set to 25000, and this laptop did a (successful!) DC on M52271179 at 2.5 ms/iter. I tried to do a proper benchmark/timing to print out so I can copy/paste it here, but couldn't figure it out. I know there is a "P-1 for life" group out there, and I'm not about that life, but finding a factor is quite definitive and saves the need of both a first time and DC test. It's hard to not use up my 32 GB ram on this laptop But OTOH, throughput is a pretty solid measure of properly used time.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Stockfish game: "Move 8 poll", not "move 3.14159 discussion"	MooMoo2	Other Chess Games	5	2016-10-22 01:55
Aouessare-El Haddouchi-Essaaidi "test": "if Mp has no factor, it is prime!"	wildrabbitt	Miscellaneous Math	11	2015-03-06 08:17
Your opinions on "x times less" and other errors	Jayder	Miscellaneous Math	12	2014-10-26 14:41
Would Minimizing "iterations between results file" may reveal "is not prime" earlier?	nitai1999	Software	7	2004-08-26 18:12

2019-08-02, 17:33	#2
petrw1 1976 Toyota Corona years forever! "Wayne" Nov 2006 Saskatchewan, Canada 2×5×7×67 Posts	What did the expected success rate change from/to? https://www.mersenne.ca/prob.php will tell you expected success rate and also total GHZ Days. If you know your daily rate you can calculate the days to complete. You should have that from the DC you just finished. Though in my experience P1 is less efficient than DC.

2019-08-02, 19:41	#4
kriesel "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 2×3²×7×43 Posts	For production use, more than 2 tests saved is generally regarded as wasted time. Except on some gpus, that might fall short of the gpu72 bounds, a nudge upward by using 3 may be merited. I'm in the process of benchmarking prime95 P-1 on test_saved=2 sampling across the mersenne.org exponent spectrum, and there are indications it also defaults to less than the gpu72 goals with 8GB allocated, so perhaps would benefit from an upward nudge or more RAM allocated. See https://www.mersenneforum.org/showpost.php?p=503892&postcount=3 for scattered data points up to 801M currently.

2019-08-03, 09:07	#6
VBCurtis "Curtis" Feb 2005 Riverside, CA 2²·1,217 Posts	We like having factors a bit more than having two matching residues, so if you don't mind the slightly inefficient use of computer time to have a better chance at a factor, boost it to 3. I would do so myself, because I like finding factors (but not because it will save time in the long run). The B1 stage should scale linearly; that is, if B1 rises by a factor of 5, stage 1 should take 5x longer. Dunno about stage 2.