Performance considerations

2018-10-14, 08:06

I have a dual-core hyperthreaded that is running trial factoring with Debian 9.5 and mprime.
It had the GNOME graphical interface, and the timing was 37 ms.
I have removed the graphical interface and the timing is now 35 ms.

2018-10-14, 08:53

Quote:

Originally Posted by SELROC

I have a dual-core hyperthreaded that is running trial factoring with Debian 9.5 and mprime.
It had the GNOME graphical interface, and the timing was 37 ms.
I have removed the graphical interface and the timing is now 35 ms.

I should change my glasses.
In rereading it is sec. not ms.

not surprising though, GNOME has at least 15 processes running in background.

ET_ · 2018-10-14, 09:16

Quote:

Originally Posted by SELROC

I should change my glasses.
In rereading it is sec. not ms.

not surprising though, GNOME has at least 15 processes running in background.

Whatever has to do with threading, coordinates' and (video) memory access calculation consumes CPU... In the old times, when Prime95 ran as a screen-saver, people used to setup a blank screen to avoid unnecessary CPU usage for printing beautiful and not useful pretty graphics.

I had a big performance boost on my Raspberry PI systems just deselecting GUI...

2018-10-14, 10:31

Quote:

Originally Posted by ET_

Whatever has to do with threading, coordinates' and (video) memory access calculation consumes CPU... In the old times, when Prime95 ran as a screen-saver, people used to setup a blank screen to avoid unnecessary CPU usage for printing beautiful and not useful pretty graphics.

I had a big performance boost on my Raspberry PI systems just deselecting GUI...

There is a fundamental difference between a Desktop and a Compute node.
A compute node is optimized for computing and in some way is also specialized for the task. Desktops are optimized for graphics.

ET_ · 2018-10-14, 15:37

Quote:

Originally Posted by SELROC

There is a fundamental difference between a Desktop and a Compute node.
A compute node is optimized for computing and in some way is also specialized for the task. Desktops are optimized for graphics.

So, do you consider your Debian environment

1 - a Desktop, or
2 - a computing node?

If 1, then you should not deselect the GUI just to achieve a 6.5% benefit on just one program.
If 2, then you should already be aware that deselecting the GUI you achieve a performance boost.

2018-10-14, 15:45

Quote:

Originally Posted by ET_

So, do you consider your Debian environment

1 - a Desktop, or
2 - a computing node?

If 1, then you should not deselect the GUI just to achieve a 6.5% benefit on just one program.
If 2, then you should already be aware that deselecting the GUI you achieve a performance boost.

I have both separately, my "Debian environment" is probably my Desktop where I only do short tests, compute node is the production system.

paulunderwood · 2018-10-15, 10:44

Desktop-less Gentoo is the fastest when I have run tests on Linux OS's. Having said this, George's assembly code will dominate on any OS.

Mark Rose · 2018-10-15, 16:33

I can also recommend Clear Linux. It was noticeably faster than Ubuntu. I switched back to Ubuntu for NVidia/CUDA support though.

irowiki · 2018-10-16, 19:41

I have been using CentOS 7 (been running a CentOS webserver for over 10 years) with some Debian machines because I've found Debian makes for a quicker setup overall.

I'll give Clear linux a try!

2018-10-17, 12:07

Quote:

Originally Posted by paulunderwood

Desktop-less Gentoo is the fastest when I have run tests on Linux OS's. Having said this, George's assembly code will dominate on any OS.

Quote:

Originally Posted by Mark Rose

I can also recommend Clear Linux. It was noticeably faster than Ubuntu. I switched back to Ubuntu for NVidia/CUDA support though.

Quote:

Originally Posted by irowiki

I have been using CentOS 7 (been running a CentOS webserver for over 10 years) with some Debian machines because I've found Debian makes for a quicker setup overall.

I'll give Clear linux a try!

I don't know if there is a mprime port for BSD, but this is worth trying out

https://www.phoronix.com/scan.php?pa...3-1280v5&num=1

kriesel · 2018-10-19, 18:45

Quote:

Originally Posted by SELROC

I don't know if there is a mprime port for BSD, but this is worth trying out

https://www.phoronix.com/scan.php?pa...3-1280v5&num=1

FreeBSD various versions yes. Plus source. https://www.mersenne.org/download/

2018-10-14, 08:06	#1
SELROC AAA₁₆ Posts	Performance considerations I have a dual-core hyperthreaded that is running trial factoring with Debian 9.5 and mprime. It had the GNOME graphical interface, and the timing was 37 ms. I have removed the graphical interface and the timing is now 35 ms.

2018-10-15, 10:44	#7
paulunderwood Sep 2002 Database er0rr 3,739 Posts	Desktop-less Gentoo is the fastest when I have run tests on Linux OS's. Having said this, George's assembly code will dominate on any OS. Last fiddled with by paulunderwood on 2018-10-15 at 10:44

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
CADO-NFS vs GGNFS and other NFS considerations [split from New Factoring Algorithm]	Dubslow	Factoring	6	2018-04-27 19:56
RAID 1 set-up, USB vs. NAS, and other considerations	Uncwilly	Hardware	15	2016-09-25 13:39
64 bit performance?	zacariaz	Hardware	1	2007-05-10 13:08
LLR performance on k and n	robert44444uk	15k Search	1	2006-02-09 01:43
Performance	battlemaxx	Prime Sierpinski Project	4	2005-06-29 20:32

2018-10-15, 16:33	#8
Mark Rose "/X\(‘-‘)/X\" Jan 2013 2²×733 Posts	I can also recommend Clear Linux. It was noticeably faster than Ubuntu. I switched back to Ubuntu for NVidia/CUDA support though.

2018-10-16, 19:41	#9
irowiki Sep 2018 3·23 Posts	I have been using CentOS 7 (been running a CentOS webserver for over 10 years) with some Debian machines because I've found Debian makes for a quicker setup overall. I'll give Clear linux a try!