PDA

View Full Version : Octos compared for BOINC Performance



Dirk Broer
01-09-2013, 06:42 PM
You often read that people do not really need more cores because the existing software hardly makes use of those extra cores. While this is increasingly not true in itself, it is absolutely untrue in BOINC, where every core counts. I will show you some scores for octo-core Bulldozers and Piledrivers, compared against hexa- and quad-core Phenom II's and a quad-core Athlon that most of you will still be using.
As with the quads I will break it up into project categories, starting with the Astrophysical ones (A. Because the 'A' comes first in the alphabet and B. Because WuProp has these values complete, unlike some other categories).
Two first value is the maximum per core per day, the next is the value per CPU per day, multiplying the scores with 8 for octos, 6 for hexas and 4 for quads.

1: Astrophysics


Project Name
Athlon II X4 640
Phenom II X4 965
Phenom II X6 1055
Phenom II X6 1090
Athlon FX-8120
Athlon FX-8150
Athlon FX-8320
Athlon FX-8350


Asteroids

1,040/4,160

1,286/5,144

1,078/6,468

1,241/7,446

878/7,024

987/7,896

1,527/12,216

1,108/8,864


Constellation

592/2,368

709/2,836

575/3,450

650/3,900

466/3,728

492/3,936

814/6,512

747/5,976


Cosmology

904/3,616

658/2,632

706/4,236

996/5,976

549/4,392

553/4,424

715/5,720

853/6,824


Einstein-(SSE/SSE2)

823/3,292

843/3,372

787/4,719

917/5,502

505/4,037

843/6,740

na

ub


Seti

654/2,616

1,235/4,938

1,006/6,033

1,389/8,334

514/4,109

921/7,371

680/5,436

867/6,932


na= Not Available
ub=unbelievable (as in too high to be true)

Ramjet
01-09-2013, 11:15 PM
This pretty much agrees with what I was saying, the 8150 even having two more cores, is only slightly better in some of the projects and worse in others, than a Phenom 1090T. Pretty frustrating after spending all that time an money upgrading, only to find I had to OC the heck out of it to match the 1090 I replaced. I kept looking around for a parachute or anchor attached to the 8150 that I could cut loose, but I still haven't found one. :( I'll keep looking, maybe it has electric brakes or hyperthreading I can utilize.

Dirk Broer
01-10-2013, 12:07 AM
I think you may have to wait for optimized applications that make full use of the extra instruction sets that Bulldozer (http://www.cpu-world.com/CPUs/Bulldozer/AMD-FX-Series%20FX-8150.html)http://www.perumodding.com/wp-content/uploads/2011/02/Bulldozer_AMD.jpg supports as compared to Phenom (http://www.cpu-world.com/CPUs/K10/AMD-Phenom%20II%20X6%201090T%20Black%20Edition%20-%20HDT90ZFBK6DGR%20(HDT90ZFBGRBOX).html)http://ic.tweakimg.net/ext/i/1270719532.jpeg

Ramjet
01-12-2013, 12:19 AM
I think you may have to wait for optimized applications that make full use of the extra instruction sets that Bulldozer (http://www.cpu-world.com/CPUs/Bulldozer/AMD-FX-Series%20FX-8150.html)http://www.perumodding.com/wp-content/uploads/2011/02/Bulldozer_AMD.jpg supports as compared to Phenom (http://www.cpu-world.com/CPUs/K10/AMD-Phenom%20II%20X6%201090T%20Black%20Edition%20-%20HDT90ZFBK6DGR%20(HDT90ZFBGRBOX).html)http://ic.tweakimg.net/ext/i/1270719532.jpeg

After viewing your graphic Dirk, I now understand why the bulldozer is so slow, it takes a lot of power to turn those huge steel tracks to make it go. :icon_mrgreen:

Dirk Broer
01-13-2013, 10:36 AM
It might pay to obtain the sources for the various BOINC applications and re-compile them using an AMD compiler, read this (http://www.agner.org/optimize/blog/read.php?i=49)

AMDave
01-13-2013, 11:01 AM
Supplementary resource:
http://developer.amd.com/tools/cpu-development/x86-open64-compiler-suite/

Dirk Broer
01-13-2013, 03:54 PM
With these resources we might be able to launch our own 'lunatics' builds (at least for Linux/RHES/SLES...)