Do # of cores or GHz/per cores perform better?

The most demanding Projects are only available to a small percentage of very high-end servers.

Moderators: Site Moderators, PandeGroup

Re: Do # of cores or GHz/per cores perform better?

Postby k1wi » Wed Feb 15, 2012 6:03 am

I did some searching for price and found on ebay that one seller is offering both cpus new at:

p650 199.99
p960 279.99 (40% more)

Obviously there are CPUs out there that are a lot cheaper (as your purchase proved), but I thought that was a good yardstick to use.

I imagine that the voltage is lower on the p960 because each core is clocked lower. I imagine, however, that the lower voltage is offset by a larger current used to supply the additional cores.
k1wi
 
Posts: 1152
Joined: Tue Sep 22, 2009 10:48 pm

Re: Do # of cores or GHz/per cores perform better?

Postby orion » Wed Feb 15, 2012 12:07 pm

This was the difference I saw back on 2/28/11 & 3/2/11 between an x6 @ 3.9GHZ (6 core, 23.4) vs. a 2p with quad cpu's @ 3.1GHz (8 cores, 24.8).
source x6: viewtopic.php?f=55&t=17730&start=30#p176708
source: 2p viewtopic.php?f=55&t=17730&start=30#p176914

Project: 6901
Average time/frame: 00:28:56
CPU: 1090T @ 3.9 GHz
# of CPU sockets: 1
# of Physical cores: 6

# of FAH CPU processes: 6
# of FAH GPU Clients: 0

RAM GB installed: 8
RAM Type: DDR3
RAM Speed:1333 8-8-8-24-2T
NUMA enabled/disabled/not used: na

OS name/kernel version: Ubuntu 8.04 2.6.24-28 generic
Client: LINUX 6.34
Running in VM: No
Dedicated 24/7 folder: YES

----------------------------------------------------------------------

Project: 6901
Average time/frame: 00:28:50
CPU: 2393 @ 3.1 GHz
# of CPU sockets: 2
# of Physical cores: 8

# of FAH CPU processes: 8
# of FAH GPU Clients: 0

RAM GB installed: 8
RAM Type: DDR2
RAM Speed: 667 5-5-5-15
NUMA enabled/disabled/not used: na

OS name/kernel version: Ubuntu 8.04 2.6.24-28 generic
Client: LINUX 6.34
Running in VM: No

Would the 2p have been faster with faster memory, probably, but not by much.
iustus quia...
User avatar
orion
 
Posts: 285
Joined: Sun Dec 02, 2007 12:45 pm
Location: neither here nor there

Re: Do # of cores or GHz/per cores perform better?

Postby PinHead » Wed Feb 15, 2012 11:37 pm

Orion,

That newer Phenom holds up pretty well against the older Opterons. Total GHz are pretty close, wouldn't have expected the time to be so close. But at lower clock speeds, I guess the NOP's hurt a little more.
PinHead
 
Posts: 351
Joined: Tue Jan 24, 2012 3:43 am

Re: Do # of cores or GHz/per cores perform better?

Postby orion » Thu Feb 16, 2012 11:25 am

Yep.

I do wish I would have run the x6 memory @ 667 just to make the two that much closer in specs. I must have been like most review sites when compare product X & Y so as to make product X look better then product Y ;)
User avatar
orion
 
Posts: 285
Joined: Sun Dec 02, 2007 12:45 pm
Location: neither here nor there

Re: Do # of cores or GHz/per cores perform better?

Postby PinHead » Thu Feb 23, 2012 2:30 am

So I added a 2nd 16 core x 2.1 GHz to a 1P doing 85.xx min TPF on a 6904 WU. As expected, it wasn't a 50% reduction in TPF. TPF decreased to around 45.xx on the same RCG WU. So I am seeing a 3.xx% overhead for the additional cpu and ( OS or client ) bouncing the thread between CPU's and CPU cache. It's still a heafty gain, but I think ( if it were made ) a 16 x 4.2 would show closer to a TPF being cut in half. Now what happens to the power and heat is a whole different story. Think I'll stick with the 3% overhead. :D
PinHead
 
Posts: 351
Joined: Tue Jan 24, 2012 3:43 am

Re: Do # of cores or GHz/per cores perform better?

Postby MtM » Thu Feb 23, 2012 2:11 pm

PinHead are you using tear's affinity script/the kraken?
MtM
 
Posts: 3054
Joined: Fri Jun 27, 2008 2:20 pm
Location: The Netherlands

Re: Do # of cores or GHz/per cores perform better?

Postby PinHead » Fri Feb 24, 2012 1:42 am

MtM wrote:PinHead are you using tear's affinity script/the kraken?


Not yet, it's downloaded; but I am playing with other settings right now. I did see where someone with the same rig shaved 3-4 minutes off their tpf using it.

So it's next on the list. :P
PinHead
 
Posts: 351
Joined: Tue Jan 24, 2012 3:43 am

Re: Do # of cores or GHz/per cores perform better?

Postby War » Sun Mar 04, 2012 12:28 pm

I think I would go for the 6164-6180SE 12 core Magny cours cpu as it has 12 FPU (floting point unit) as this is what folding takes advantage of.
The 16 core will be more cores and sound better, but each pair of cores is actually two integer processing units with a shared floating point unit so the total will be 8 FPU.
War
 
Posts: 11
Joined: Thu Feb 05, 2009 4:37 am

Re: Do # of cores or GHz/per cores perform better?

Postby dreiter » Tue Jan 29, 2013 7:45 pm

bruce wrote:....Since FAH uses mostly floating point processing, a "core" that shares a FPU with another "core" is worth very slightly more than half of what a "core" that has it's own FPU is worth.


Sorry to necro the thread, but is this still true with the Abu Dhabi / Piledriver chips? I know they still share 1 FPU across 2 cores, but have architecture improvements given some increased performance? AKA is a 16 'core' Abu Dhabi chip still only as good as an 8 core Interlagos chip? I can't find any benchmarks. :(
dreiter
 
Posts: 26
Joined: Mon Jul 18, 2011 4:43 am
Location: CA, USA

Re: Do # of cores or GHz/per cores perform better?

Postby Nathan_P » Tue Jan 29, 2013 9:00 pm

To my knowledge no-one has yet tested 63xx opterons with F@h and published the results, for now if you are thinking of AMD 2p or 4p you are better off with the 61xx Magny cours chips. I don't expect any great improvement in this until steamroller gets released
Image
Nathan_P
 
Posts: 1422
Joined: Wed Apr 01, 2009 9:22 pm
Location: Jersey, Channel islands

Re: Do # of cores or GHz/per cores perform better?

Postby mmonnin » Tue Jan 29, 2013 11:18 pm

If someone really wanted a little better comparison than the laptop one given they could disable 2 of the cores in a quad at say 3 GHz and then run the quad at 1.5 GHz with all 4 cores enabled.
mmonnin
 
Posts: 335
Joined: Wed Dec 05, 2007 1:27 am

Re: Do # of cores or GHz/per cores perform better?

Postby ProDigit » Sun Dec 16, 2018 4:23 pm

With my limited knowledge of FAH, but larger knowledge of CPUs,
I would say that it depends on what architecture.

In order execution CPUs, like Intel Atom processors, Intel Xeon Phi, certain Via chipsets, and ARM; as well as most power efficiency optimized chips (eg: those without SSE, VT, etc..), will benefit a lot more from more cores, less Ghz.
Out of order architectures, as well as hyperthreading benefit more from higher Ghz over more cores; but that would depend on how efficient the algorithm of FAH is in using all the possible instruction sets and added extensions of the CPU.
ProDigit
 
Posts: 242
Joined: Sun Dec 09, 2018 10:23 pm

Re: Do # of cores or GHz/per cores perform better?

Postby bruce » Tue Dec 18, 2018 2:46 am

FAHCore_a4 uses SSE. FAHCore_a7 will invoke AVX if you have it. Both are pretty efficient, being written in hand-optimized ALC. (When the original SSE code was released, FAH did a pretty good job of pushing well designed overclocked CPUs past their thermal limits. I suspect the same may be true for AVX.)
bruce
 
Posts: 22026
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Previous

Return to SMP with bigadv

Who is online

Users browsing this forum: Magpie [Crawler] and 2 guests

cron