fbncsieve - a new fixed n sieve - Page 3

axn · 2017-11-28, 16:49

On my skylake, the x86 version gives about 8.74M p/sec, while x87 gives 8.27M p/sec. So x86 is about 5-6% faster.

pepi37 · 2017-11-28, 18:23

Quote:

Originally Posted by axn

On my skylake, the x86 version gives about 8.74M p/sec, while x87 gives 8.27M p/sec. So x86 is about 5-6% faster.

Can you compare it with newpgen? Is it really 2-2.5x faster?

rogue · 2017-11-29, 22:22

I applied those changes and I found a typo in the code. 1.0.1 is now posted.

pepi37 · 2017-11-30, 08:59

Where is posted? archive on first page is still 1.0.0 ( at least that is what 7z file is named)

rogue · 2017-11-30, 14:11

Quote:

Originally Posted by pepi37

Where is posted? archive on first page is still 1.0.0 ( at least that is what 7z file is named)

Sorry. I thought I fixed that link. I'll do that after I get home from work.

axn · 2017-11-30, 15:47

Quote:

Originally Posted by pepi37

Where is posted? archive on first page is still 1.0.0 ( at least that is what 7z file is named)

Manually changing the OP link to http://www.mersenneforum.org/rogue/fbncsieve_1.0.1.7z works.

axn · 2017-11-30, 16:20

Quote:

Originally Posted by pepi37

Can you compare it with newpgen? Is it really 2-2.5x faster?

Sieving k*3^1024+1 (k = 2..1e9).
fb 1.0.1 x86 took 8m9s to reach 97G
snewpgen took 20m31s to reach 97G

So 20.5/8.15 ~= 2.5x faster

I'll now do the testing using base 2 where newpgen is much faster.

EDIT:-
Sieving k*2^333333+1 (k=2..1e9)
fb 1.0.1 x86 took 12m15s to reach 98G
snewpgen took 7m8s to reach 98G
So it went from 2.5x slower in base 3 to 1.7x faster in base 2.

rogue · 2017-11-30, 16:37

Quote:

Originally Posted by axn

Sieving k*3^1024+1 (k = 2..1e9).
fb 1.0.1 x86 took 8m9s to reach 97G
snewpgen took 20m31s to reach 97G

So 20.5/8.15 ~= 2.5x faster

I'll now do the testing using base 2 where newpgen is much faster.

Did you compare the outputs?

There are zero optimizations for base 2 in fbncsieve.

axn · 2017-11-30, 17:00

And lastly base 10

k*10^999999+1 (k=2..10^8)

fb - 7m10s to get to 56G
NPG 12m48s to 28G, so 25m36s to get to 56G. That is almost 3.5x slower compared to fbnc.

So the conlusion is: base-2 - stick with NewpGen, Everything else go fbnc

axn · 2017-11-30, 17:01

Quote:

Originally Posted by rogue

Did you compare the outputs?

I spot checked initial few lines, and they match between both.

masser · 2017-11-30, 17:52

Quote:

Originally Posted by rogue

Although this will help Conjectures 'R Us more than any other project, I intend to use this code as the base to write a Mac/Linux version of fermfact.

Please excuse my ignorance, but how does this help CRUS?

2017-11-30, 08:59	#26
pepi37 Dec 2011 After milion nines:) 1,451 Posts	Where is posted? archive on first page is still 1.0.0 ( at least that is what 7z file is named) Last fiddled with by pepi37 on 2017-11-30 at 08:59

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Sieving k * 2^n +- c with Nvidia GPU's for fixed k	diep	GPU Computing	5	2016-09-23 19:19
A siever for K (b, n, c fixed)?	pepi37	Software	7	2015-07-10 04:42
Sieving k*2^n-1 With Fixed n	c10ck3r	Riesel Prime Search	14	2013-02-03 00:19
User interface bug fixed on LLR V3.8.4	Jean Penné	Software	0	2011-01-22 16:47
KEP is reporting computer fixed	KEP	Twin Prime Search	3	2007-02-13 18:29

2017-11-28, 16:49	#23
axn Jun 2003 2·3·7·11² Posts	On my skylake, the x86 version gives about 8.74M p/sec, while x87 gives 8.27M p/sec. So x86 is about 5-6% faster.

2017-11-29, 22:22	#25
rogue "Mark" Apr 2003 Between here and the 2·3²·353 Posts	I applied those changes and I found a typo in the code. 1.0.1 is now posted.

2017-11-30, 17:00	#31
axn Jun 2003 1001111011010₂ Posts	And lastly base 10 k*10^999999+1 (k=2..10^8) fb - 7m10s to get to 56G NPG 12m48s to 28G, so 25m36s to get to 56G. That is almost 3.5x slower compared to fbnc. So the conlusion is: base-2 - stick with NewpGen, Everything else go fbnc