Prime finding rate, Sierp vs. Riesel?

CGKIII · 2012-09-01, 18:19

Has anyone done analysis of whether the prime finding rate is higher for Sierp or Riesel?

The most appropriate comparison, I think, would be to look at bases with the same conjecture, and then look at the distribution of k's remaining at a given n level. Though same bases with different conjectures might be fine as well (especially if you only consider k's < the min conjecture). Different bases, different conjectures might be worthwhile, but that seems like the least appropriate of comparisons.

If there's data someone can send me, I'd be happy to investigate myself.

The question I'm trying to answer is: "Does it appear to be easier to find primes for S or R conjectures?" If it does look like there's a (reasonably) clear winner, it will affect the type of work I choose going forward.

MyDogBuster · 2012-09-01, 20:06

Quote:

Has anyone done analysis of whether the prime finding rate is higher for Sierp or Riesel?

I have no stats. but my observations over the years is that they are about the same, with maybe a slight edge to Riesel but not much. The difference you might see in the stats is attributed to Riesel being tested more than Sierp.

Testing times are similar based on n. The test times do increase for the same n for higher bases. I'm assuming you are running on Windows. Do you know if your machines are Sandy Bridge architecture?

henryzz · 2012-09-01, 21:10

Which of Riesel or Sierp has more algebraic factors? I imagine algebraic factors would scue the stats a little.

rogue · 2012-09-01, 23:20

Quote:

Originally Posted by henryzz

Which of Riesel or Sierp has more algebraic factors? I imagine algebraic factors would scue the stats a little.

Riesel because even powers have algebraic factors.

CGKIII · 2012-09-02, 17:03

Quote:

Originally Posted by MyDogBuster

I'm assuming you are running on Windows. Do you know if your machines are Sandy Bridge architecture?

Yes, I'm on Windows. I've got an AMD Phenom II quadcore desktop, an i5 quadcore laptop, and then an AMD Athlon II dualcore laptop. Unsure how those correspond to Sandy Bridge.

MyDogBuster · 2012-09-02, 17:19

Quote:

Yes, I'm on Windows. I've got an AMD Phenom II quadcore desktop, an i5 quadcore laptop, and then an AMD Athlon II dualcore laptop. Unsure how those correspond to Sandy Bridge.

They are not Sandy Bridge because they are AMD. Sandy Bridge is a chipset for Intel machines with an instruction set (AVX) that utilizes advanced vectoring. A few people ported CLLR to use this new instruction set. It's about 30-40% faster than pfgw or cllr. Just checking.

henryzz · 2012-09-02, 19:35

Quote:

Originally Posted by MyDogBuster

They are not Sandy Bridge because they are AMD. Sandy Bridge is a chipset for Intel machines with an instruction set (AVX) that utilizes advanced vectoring. A few people ported CLLR to use this new instruction set. It's about 30-40% faster than pfgw or cllr. Just checking.

The i5 could be sandy bridge. If you use the latest official version of llr(3.8.9) then it will use AVX if detected(and not if not detected). The latest pfgw is the same. AMD Bulldozer supports AVX but runs slower when using the AVX code so it is disabled for them by default.

@CGKIII Do you mean an i7 quadcore laptop or i5 quadcore desktop? All i5 laptops are dualcore not quad. :) Not that this matters.

CGKIII · 2012-09-02, 23:21

Henryzz, the laptop says "Intel(R) Core(TM) i5 CPU M 560 @ 2.67 GHz," 2.66 GHz, 3.42 GB of RAM. When I pop open the Task Manager, go to the Performance tab, it shows four slots four CPU usage. I get ~25% usage per active client. Not sure if there's a better place for me to be looking. It's a Dell Latitude E6410.

I have llrexe =cllr.exe, using cllr.exe from the crus_pack in one of the intro threads.

Dubslow · 2012-09-03, 00:54

Quote:

Originally Posted by CGKIII

Henryzz, the laptop says "Intel(R) Core(TM) i5 CPU M 560 @ 2.67 GHz," 2.66 GHz, 3.42 GB of RAM. When I pop open the Task Manager, go to the Performance tab, it shows four slots four CPU usage. I get ~25% usage per active client. Not sure if there's a better place for me to be looking. It's a Dell Latitude E6410.

I have llrexe =cllr.exe, using cllr.exe from the crus_pack in one of the intro threads.

The Sandy Bridge CPUs are all designated 2xxx, while yours is 560, so it's not Sandy Bridge. (Ivy Bridge are 3xxx.)

henryzz · 2012-09-03, 08:55

Quote:

Originally Posted by CGKIII

Henryzz, the laptop says "Intel(R) Core(TM) i5 CPU M 560 @ 2.67 GHz," 2.66 GHz, 3.42 GB of RAM. When I pop open the Task Manager, go to the Performance tab, it shows four slots four CPU usage. I get ~25% usage per active client. Not sure if there's a better place for me to be looking. It's a Dell Latitude E6410.

I have llrexe =cllr.exe, using cllr.exe from the crus_pack in one of the intro threads.

The thing that is probably confusing you is that your CPU is dual core with hyperthreading. This makes it appear like you have 4 cores. With some workloads hyperthreading can provide a speedup. This is so optimised it doesn't provide any worth mentioning. If you find a way of turning it off(hard if not impossible on laptops) then it should use a little less power and produce less heat.

LaurV · 2012-09-03, 14:25

With P95 you don't need to turn off nothing, just launch only 2 workers. The program is enough clever to use the right affinity. To make sure, you can use the option/benchmark function and see if any improvement 2 workers against 4 workers, most probably not, and 4 workers may be even slower for higher expos, due to memory bandwidth limit.

2012-09-01, 18:19	#1
CGKIII Aug 2012 19₁₆ Posts	Prime finding rate, Sierp vs. Riesel? Has anyone done analysis of whether the prime finding rate is higher for Sierp or Riesel? The most appropriate comparison, I think, would be to look at bases with the same conjecture, and then look at the distribution of k's remaining at a given n level. Though same bases with different conjectures might be fine as well (especially if you only consider k's < the min conjecture). Different bases, different conjectures might be worthwhile, but that seems like the least appropriate of comparisons. If there's data someone can send me, I'd be happy to investigate myself. The question I'm trying to answer is: "Does it appear to be easier to find primes for S or R conjectures?" If it does look like there's a (reasonably) clear winner, it will affect the type of work I choose going forward.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Riesel/Sierp base 2/4 reservations/statuses/primes	Jean Penné	Conjectures 'R Us	461	2021-06-20 17:17
Mersenne prime discovery rate	davieddy	PrimeNet	0	2012-11-26 03:26
Riesel and Sierp numbers bases <= 1024	R. Gerbicz	Conjectures 'R Us	22	2009-12-29 20:21
Sieving Riesel & Sierp base 16	gd_barnes	Conjectures 'R Us	13	2009-12-14 09:23
Riesel/Sierp #'s for bases 3, 7, and 15	Siemelink	Conjectures 'R Us	105	2009-09-04 06:40

2012-09-01, 21:10	#3
henryzz Just call me Henry "David" Sep 2007 Cambridge (GMT/BST) 2³·3·5·7² Posts	Which of Riesel or Sierp has more algebraic factors? I imagine algebraic factors would scue the stats a little.

2012-09-02, 23:21	#8
CGKIII Aug 2012 25₁₀ Posts	Henryzz, the laptop says "Intel(R) Core(TM) i5 CPU M 560 @ 2.67 GHz," 2.66 GHz, 3.42 GB of RAM. When I pop open the Task Manager, go to the Performance tab, it shows four slots four CPU usage. I get ~25% usage per active client. Not sure if there's a better place for me to be looking. It's a Dell Latitude E6410. I have llrexe =cllr.exe, using cllr.exe from the crus_pack in one of the intro threads.

2012-09-03, 14:25	#11
LaurV Romulan Interpreter Jun 2011 Thailand 9612₁₀ Posts	With P95 you don't need to turn off nothing, just launch only 2 workers. The program is enough clever to use the right affinity. To make sure, you can use the option/benchmark function and see if any improvement 2 workers against 4 workers, most probably not, and 4 workers may be even slower for higher expos, due to memory bandwidth limit.