Distributed sieving discussions - Page 7

michaf · 2006-09-01, 06:49

I most definitely see an improvement on the Athlon 64

updated data file 307 k's to 305 k's as well..
(That should be say 1% improvement...)

But went from 30.5kp/s to about 36.5kp/s!

Wow

All thanks to yout wonderfull programming skills :)

axn · 2006-09-01, 19:14

Quote:

Originally Posted by geoff

sr5sieve 1.1.3 is about 10% faster than 1.1.2. This resulted from using 32 bits instead of 64 bits for some local variables in the modular inverse function (so probably only 32-bit machines will see the benefit).

Wow! 10% is really conservative. I am seeing 14-15% improvement for my P-M and Core Duo Laptops. WTG

UPGRADE! UPGRADE! UPGRADE!

Quote:

Originally Posted by michaf

updated data file 307 k's to 305 k's as well..
(That should be say 1% improvement...)

You do realise that after _your_ latest prime find, we now have 304 k's left, don't you?

michaf · 2006-09-01, 20:29

Quote:

Originally Posted by axn1

Wow! 10% is really conservative. I am seeing 14-15% improvement for my P-M and Core Duo Laptops. WTG

UPGRADE! UPGRADE! UPGRADE!

You do realise that after _your_ latest prime find, we now have 304 k's left, don't you?

Erm... in fact, I didn't :)
.3% faster still!!

geoff · 2006-09-02, 02:41

Quote:

Originally Posted by michaf

I most definitely see an improvement on the Athlon 64

Are you are running the 32 bit (mingw32/linux32) sr5sieve binary?

michaf · 2006-09-02, 07:05

Quote:

Originally Posted by geoff

Are you are running the 32 bit (mingw32/linux32) sr5sieve binary?

Yep, I'm running sr5sieve-1.1.3-mingw32-i686.zip now.
I have not the foggiest idea how to compile a program myself, otherwise I would try to do it to get me some more improvement :>

geoff · 2006-09-03, 02:05

sr5sive 1.1.4 should be about 5% faster than 1.1.3. This comes from using the fact that all of our candidate sequences are k*b^n+c with c=+\-1, which simplifies the calculation of -c/k (mod p).

Quote:

Originally Posted by michaf

Yep, I'm running sr5sieve-1.1.3-mingw32-i686.zip now.
I have not the foggiest idea how to compile a program myself, otherwise I would try to do it to get me some more improvement :>

The program can probably only be compiled with GCC at the moment and the mingw GCC compiler doesn't support 64 bit Windows yet, so there is probably not much you can do anyway.

I still have no idea how fast the 64 bit Linux version is compared to the 32 bit version. I have looked at the assembler that GCC produces for the k8 and it is full of SSE2 instructions with many variables kept in registers, very different to the code produced for the P4 which hardly uses SSE2 at all, or the 32 bit versions where most variables have to be kept in memory.

Maybe I will have to buy an Athlon64 :-)

michaf · 2006-09-03, 06:59

Quote:

Originally Posted by geoff

sr5sive 1.1.4 should be about 5% faster than 1.1.3. This comes from using the fact that all of our candidate sequences are k*b^n+c with c=+\-1, which simplifies the calculation of -c/k (mod p).

wow again :> hmm... is you program getting smaller each and every time? :)

Quote:

Originally Posted by geoff

The program can probably only be compiled with GCC at the moment and the mingw GCC compiler doesn't support 64 bit Windows yet, so there is probably not much you can do anyway.

Pity, but I'd still would have had a lot to learn to make me a first compile anyway :>

Quote:

Originally Posted by geoff

I still have no idea how fast the 64 bit Linux version is compared to the 32 bit version. I have looked at the assembler that GCC produces for the k8 and it is full of SSE2 instructions with many variables kept in registers, very different to the code produced for the P4 which hardly uses SSE2 at all, or the 32 bit versions where most variables have to be kept in memory.

Maybe I will have to buy an Athlon64 :-)

I can recommend them... never bailed out on me yet (Though it had not that much time online in one run...)

geoff · 2006-09-06, 04:47

sr5sieve 1.1.5 is about 10% faster than 1.1.4. due to use of the modular inverse function from jasonp's Msieve source. Thanks Jason!

michaf · 2006-09-12, 05:56

Either I'm an idiot, or I just can't find how:

how do I start a new sieve on a number of sequences

lets say

5114*5^n+1
6436*5^n+1
7528*5^n+1
8644*5^n+1

How do I get the sieve to start on this on a new sieve?

axn · 2006-09-12, 17:46

Quote:

Originally Posted by michaf

how do I start a new sieve on a number of sequences

sr5sieve can't do it -- You need to use srsieve from http://geocities.com/g_w_reynolds/srsieve/

michaf · 2006-09-12, 18:37

Quote:

Originally Posted by axn1

sr5sieve can't do it -- You need to use srsieve from http://geocities.com/g_w_reynolds/srsieve/

Oh darn... I knew that :) thanks

2006-09-01, 06:49	#67
michaf Jan 2005 111011111₂ Posts	I most definitely see an improvement on the Athlon 64 updated data file 307 k's to 305 k's as well.. (That should be say 1% improvement...) But went from 30.5kp/s to about 36.5kp/s! Wow All thanks to yout wonderfull programming skills :) Last fiddled with by michaf on 2006-09-01 at 07:01

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Soapbox Discussions	only_human	Soap Box	41	2019-11-16 15:46
Our discussions here: how can we improve things?	Brian-E	Soap Box	105	2013-11-10 12:26
n=390000 discussions and old reservations	Mini-Geek	Twin Prime Search	31	2010-05-22 23:13
Primegrid discussions	pacionet	Twin Prime Search	17	2007-01-20 11:22
Automated PRP discussions	ltd	Sierpinski/Riesel Base 5	20	2006-09-02 22:19

2006-09-06, 04:47	#74
geoff Mar 2003 New Zealand 13×89 Posts	sr5sieve 1.1.5 is about 10% faster than 1.1.4. due to use of the modular inverse function from jasonp's Msieve source. Thanks Jason!

2006-09-12, 05:56	#75
michaf Jan 2005 479₁₀ Posts	Either I'm an idiot, or I just can't find how: how do I start a new sieve on a number of sequences lets say 51145^n+1 64365^n+1 75285^n+1 86445^n+1 How do I get the sieve to start on this on a new sieve?