mersenneforum.org  

Go Back   mersenneforum.org > Prime Search Projects > Riesel Prime Search

Reply
 
Thread Tools
Old 2013-08-25, 17:26   #1
Trilo
 
Trilo's Avatar
 
"W. Byerly"
Aug 2013
81*2^2797443-1

1638 Posts
Default Question about multi-core prime searching.

I am using a 7 core 2.2ghz computer, which I have 6 cores running. 4 of them are running a newpgen file on pfgw that has been split into four sections. When I run 1 core ittakes pfgw 10s to test a candidate. However, using 6 cores takes 15s to test a candidate of similar size. Am I doing something wrong, or is this normal? Also, do is processing an n of size 116000 in 10-15s a normal speed, or do I have a slow computer? Is what I am doing the fastest way to test candidates? Because it takes me a day to seach through n= 100000- 110000 on this computer and themi hear about people finding primes with n's in the 700000's and I am wondering how they do it, do they use really fast computers, or do they wait weeks to find them.

Last fiddled with by Trilo on 2013-08-25 at 17:32
Trilo is offline   Reply With Quote
Old 2013-08-25, 17:33   #2
rogue
 
rogue's Avatar
 
"Mark"
Apr 2003
Between here and the

194616 Posts
Default

Quote:
Originally Posted by Trilo View Post
I am using a 7 core 2.2ghz computer, which I have 6 cores running. 4 of them are running a newpgen file on pfgw that has been split into four sections. When I run 1 core ittakes pfgw 10s to test a candidate. However, using 6 cores takes 15s to test a candidate of similar size. Am I doing something wrong, or is this normal? Also, do is processing an n of size 116000 in 10-15s a normal speed, or do I have a slow computer? Is what I am doing the fastest way to test candidates?
Is that logical or physical cores?

BTW, llr is faster than pfgw for base 2.
rogue is offline   Reply With Quote
Old 2013-08-25, 17:35   #3
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

11·461 Posts
Default

Without more information about your specific setup (I have never heard of a 7-core machine, unless you're including your GPU), we can only tell you that Prime95/LLR/PFGW tax the memory subsystem quite thoroughly. As you add copies of those programs to your task list, you may find a number of copies such that the next run you fire up slows down all your tests.

This number of simultaneous tests does depend on the size of the exponent. There are other factoring/sieving/primenumber-related programs that tax memory less, which can be run on the rest of the cores.
VBCurtis is online now   Reply With Quote
Old 2013-08-25, 19:51   #4
Trilo
 
Trilo's Avatar
 
"W. Byerly"
Aug 2013
81*2^2797443-1

5×23 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
Without more information about your specific setup (I have never heard of a 7-core machine, unless you're including your GPU), we can only tell you that Prime95/LLR/PFGW tax the memory subsystem quite thoroughly. As you add copies of those programs to your task list, you may find a number of copies such that the next run you fire up slows down all your tests.

This number of simultaneous tests does depend on the size of the exponent. There are other factoring/sieving/primenumber-related programs that tax memory less, which can be run on the rest of the cores.
yes, I considered the 7th core my gpu, and I was using base 2.
Trilo is offline   Reply With Quote
Old 2013-08-25, 19:52   #5
TheMawn
 
TheMawn's Avatar
 
May 2013
East. Always East.

11×157 Posts
Default

7 cores sound interesting to me, too. The previous comment is entirely correct. Certain programs (I don't know much about the Riesel Prime stuff) do have to access the memory an awful lot. The extra cores have to wait their turn to access the memory which just can't keep up with the demand.

If you have 6 cores doing one candidate in 15 ms, you're doing roughly 400 candidates per second. Because you are likely bandwidth limited, 400 candidates is going to be your max throughput no matter what you do, short of overclocking the memory. Try running it on four cores to see if you get 10 ms per candidate. Or 12.5 ms on five cores.

If this is what is happening, you'll have hit a memory bottleneck at four cores, and you might want to look into running three cores and putting the others to work on something less memory bandwidth limited.
TheMawn is offline   Reply With Quote
Old 2013-08-25, 19:54   #6
Trilo
 
Trilo's Avatar
 
"W. Byerly"
Aug 2013
81*2^2797443-1

5×23 Posts
Default

my question is, how do people find 2^700000 prime numbers when it takes me all night long to find them in the 2^110000 range?
Trilo is offline   Reply With Quote
Old 2013-08-25, 19:57   #7
Trilo
 
Trilo's Avatar
 
"W. Byerly"
Aug 2013
81*2^2797443-1

11100112 Posts
Default

Quote:
Originally Posted by TheMawn View Post
7 cores sound interesting to me, too. The previous comment is entirely correct. Certain programs (I don't know much about the Riesel Prime stuff) do have to access the memory an awful lot. The extra cores have to wait their turn to access the memory which just can't keep up with the demand.

If you have 6 cores doing one candidate in 15 ms, you're doing roughly 400 candidates per second. Because you are likely bandwidth limited, 400 candidates is going to be your max throughput no matter what you do, short of overclocking the memory. Try running it on four cores to see if you get 10 ms per candidate. Or 12.5 ms on five cores.

If this is what is happening, you'll have hit a memory bottleneck at four cores, and you might want to look into running three cores and putting the others to work on something less memory bandwidth limited.
no, it takes 10 SECONDS per candidate in the 2^110000 range with my 7 core( 6cores + gpu) 2.2ghz labtop. using 6 cores takes 15 seconds per candidate. Do I have a slow computer or am I doing something wrong. These times are really slow:
maybe the big k value??

3*5*11*13*98584369*2^110116-1 is composite: RES64: [3EB0BD2471F424A1] (8.9393s+0.0002s)
3*5*11*13*98584369*2^110118-1 is composite: RES64: [D05EBB62E6B549D1] (8.9463s+0.0002s)
3*5*11*13*98584369*2^110120-1 is composite: RES64: [7419127763C23D3B] (8.9553s+0.0002s)
3*5*11*13*98584369*2^110124-1 is composite: RES64: [95403DBCC36CB577] (8.9407s+0.0002s)
3*5*11*13*98584369*2^110128-1 is composite: RES64: [7B81D14EDE46FF75] (8.9376s+0.0002s)
3*5*11*13*98584369*2^110130-1 is composite: RES64: [4FBEAB3B24AED91D] (8.9947s+0.0002s)
3*5*11*13*98584369*2^110135-1 is composite: RES64: [5B6635E716E3FB59] (8.9533s+0.0002s)
3*5*11*13*98584369*2^110140-1 is composite: RES64: [E736E76FE726A95E] (8.9488s+0.0002s)
3*5*11*13*98584369*2^110144-1 is composite: RES64: [6E4677AC6C6C3E44] (8.9127s+0.0002s)
3*5*11*13*98584369*2^110145-1 is composite: RES64: [4B449D255FCBD109] (8.9385s+0.0002s)
3*5*11*13*98584369*2^110147-1 is composite: RES64: [B9E02B34D05B280C] (8.9361s+0.0002s)

Note: This was when I wasn't using my computer, if i'm using it, it takes 10 seconds

Last fiddled with by Trilo on 2013-08-25 at 20:01 Reason: added sample pfgw output
Trilo is offline   Reply With Quote
Old 2013-08-25, 20:29   #8
pepi37
 
pepi37's Avatar
 
Dec 2011
After milion nines:)

27178 Posts
Default

Can you write exactly processor type? How much memory you have? What is OS? 32 or 64 bit?
Considering your setup as 7 core is very unusual. You dont have 7 cores, you have 6+1, and if you ask us to help you then you first need to tell as as more info as you can get, not hiding it .
On my Intel I7

D:\PGEN>llr64 -d -q"211463471505*2^110116-1"
Starting Lucas Lehmer Riesel prime test of 211463471505*2^110116-1
Using zero-padded AVX FFT length 12K, Pass1=256, Pass2=48
V1 = 5 ; Computing U0...done.
211463471505*2^110116-1 is not prime. LLR Res64: 5887F92232E8A5A9 Time : 5.918 sec.

running on 4 cores does not have any influence on speed.

Last fiddled with by pepi37 on 2013-08-25 at 20:33
pepi37 is offline   Reply With Quote
Old 2013-08-25, 21:20   #9
Trilo
 
Trilo's Avatar
 
"W. Byerly"
Aug 2013
81*2^2797443-1

5·23 Posts
Default

1 terabyte memory 8 gb RAM 64 bit,
intel core i7-2670QM @2.2ghz
CPU speed: 2487.21 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2
L1 cache size: 32kb
L2 cache size: 256kb

Last fiddled with by Trilo on 2013-08-25 at 21:26
Trilo is offline   Reply With Quote
Old 2013-08-25, 21:34   #10
Thomas11
 
Thomas11's Avatar
 
Feb 2003

5·383 Posts
Default

The i7-2670QM is a quad-core cpu, not a hexa-core!

Although you could run up to 8 threads due to hyperthreading, I wouldn't suggest to do so for LLR or PFGW. Only for sieving you may benefit from hyperthreading.

BTW: This explains why you get 15 secs per test when running 6 instances instead of the 10 secs for just one instance (or up to 4 instances).

Last fiddled with by Thomas11 on 2013-08-25 at 21:44
Thomas11 is offline   Reply With Quote
Old 2013-08-25, 22:00   #11
Trilo
 
Trilo's Avatar
 
"W. Byerly"
Aug 2013
81*2^2797443-1

5·23 Posts
Default

oh, I got confused because it said core i7
Trilo is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
multi-core TF dbaugh Information & Answers 3 2011-09-06 01:32
GMP-Fermat and multi-core processors ixfd64 Factoring 1 2011-03-16 16:07
Multi-Core / Multi-CPU Assignments (missing) worknplay Software 3 2008-11-05 17:26
Automatic multi-core detect and use? remaker Software 11 2007-04-02 20:18
My prediction about multi-core chips jasong Hardware 6 2006-02-14 16:07

All times are UTC. The time now is 20:49.


Sat Dec 4 20:49:09 UTC 2021 up 134 days, 15:18, 1 user, load averages: 1.04, 1.20, 1.23

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

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.