Dual CPU's vs. Single?

A

Artreid

Just curious?

Which would run faster:

A Machine with (2) separate Dual Core processors
or
A machine with (1) Quad Core processor?

My thought is the double Dual core machine would be faster but I'm not sure
about Bus involvement?
 
C

charlie

Just curious?

Which would run faster:

A Machine with (2) separate Dual Core processors
or
A machine with (1) Quad Core processor?

My thought is the double Dual core machine would be faster but I'm not
sure about Bus involvement?
If it were "only" two separate cores, with the proper support on the
motherboard, the speed might be approximately the same as quad core
single processor. However, the "Devil" is in the details.
Multicore processors located together with memory management can have
significant speed advantages.

My son has a "server" motherboard that takes two Intel processors.
I have a "Sabertooth" 990 MBD using an AMD FX8350 "8-core"
The two processors for the server MBD have a $600 cost, and the
FX-processor under $200. The Intel two processor server MBD can be up to
10% faster under certain test conditions.
The FX MBD shows a win test number of 8.1 for all but the mid/lower
range video card. The Server MBD does not currently use/run windows.
 
P

Paul

Artreid said:
Just curious?

Which would run faster:

A Machine with (2) separate Dual Core processors
or
A machine with (1) Quad Core processor?

My thought is the double Dual core machine would be faster but I'm not
sure about Bus involvement?
To some degree, it could depend on the vintage of the quad.

The Q6600, for example, a quad, consists of two silicon dies inside.
Each die is a dual core processor. Cache coherency traffic travels over the
FSB that joins the two of them, to the external FSB pins.

I couldn't find a proper block diagram, so we have to go with
the marketing diagram instead.

http://hardwarelogic.com/articles/reviews/processors/INTEL_QX9650/penryn_quadcore.jpg

So in some ways, a Q6600, would be similar to a pair of duals from
the same era.

A quad of a different design, might perform slightly better than a dual dual.

Let's take this quad as an example. The bus bottleneck described in
the Q6600, is resolved to a great extent by having a shared L3.
All four cores are on the same piece of silicon.
And since the memory controller is also built right into the thing,
this has better behavior than the Q6600. You get closer to perfect
speedup, with the way this is built.

http://www.bit-tech.net/hardware/2012/04/23/intel-core-i7-3770k-review/1

But four cores, is kinda the limit for that approach. When you
do six cores that way, instead of it running 1.5x the four core one,
it only runs 1.35x. So it's already starting to tail off. The six
cores still run faster, but not by as much as you'd hope.

So if you have to connect them together with busses external to
an individual silicon die, it generally slows them down. You don't
get linear speedup after a certain point.

And when you have a ton of cores, they stop trying to do it that
way entirely. I think Larrabee uses a ring bus inside (perhaps
similar to what a GPU uses). And Larrabee performed so well, it
didn't make it into regular production (except for special customers).

http://www.legitreviews.com/article/271/3/

http://www.tomshardware.com/reviews/intel-larrabee-graphics,2253-8.html

http://en.wikipedia.org/wiki/Larrabee_(microarchitecture)

Paul
 
J

James Silverton

If it were "only" two separate cores, with the proper support on the
motherboard, the speed might be approximately the same as quad core
single processor. However, the "Devil" is in the details.
Multicore processors located together with memory management can have
significant speed advantages.

My son has a "server" motherboard that takes two Intel processors.
I have a "Sabertooth" 990 MBD using an AMD FX8350 "8-core"
The two processors for the server MBD have a $600 cost, and the
FX-processor under $200. The Intel two processor server MBD can be up to
10% faster under certain test conditions.
The FX MBD shows a win test number of 8.1 for all but the mid/lower
range video card. The Server MBD does not currently use/run windows.
Is 10% faster worth $400. I doubt it.
 
J

John Williamson

Is 10% faster worth $400. I doubt it.
In *some* applications, yes. For instance, it could give me all the
video manipulation in real time without having to render some layers in
advance.

Or it could give me more tracks with effects as 192KHz sample rate in a mix.
 
J

James Silverton

In *some* applications, yes. For instance, it could give me all the
video manipulation in real time without having to render some layers in
advance.

Or it could give me more tracks with effects as 192KHz sample rate in a
mix.
Sounds more like memory capacity than speed!
 
J

John Williamson

Sounds more like memory capacity than speed!
They're both processor intensive tasks which need to be done in real
time. Lots of memory is needed, as is lost of processor power.
 
J

James Silverton

They're both processor intensive tasks which need to be done in real
time. Lots of memory is needed, as is lost of processor power.
That reminds me some of when parallel processing was introduced on big
IBM machines. I spent a great deal of time incorporating parallel
FORTRAN in a large matrix program but was disappointed to achieve an
overall speed up of only 10-15% even if matrices were multiplied or
inverted several times as fast.
 
Y

Yousuf Khan

Just curious?

Which would run faster:

A Machine with (2) separate Dual Core processors
or
A machine with (1) Quad Core processor?

My thought is the double Dual core machine would be faster but I'm not
sure about Bus involvement?
Multi-socket motherboards are usually server boards, not necessarily
desktop boards. There two issues here. Given identical generations/makes
of processors, then the dual-socket machine will usually include
separate channels for each socket. This might help in keeping RAM
overuse under control. On the other hand, a single socket processor
would be sharing data through a common L2 and/or L3 cache, so this in
its own might help make things faster if the cores need to share stuff
with each other.

So it depends on how much data sharing your applications need to do.

Yousuf Khan
 
Y

Yousuf Khan

My son has a "server" motherboard that takes two Intel processors.
I have a "Sabertooth" 990 MBD using an AMD FX8350 "8-core"
The two processors for the server MBD have a $600 cost, and the
FX-processor under $200. The Intel two processor server MBD can be up to
10% faster under certain test conditions.
The FX MBD shows a win test number of 8.1 for all but the mid/lower
range video card. The Server MBD does not currently use/run windows.
This probably has more to do with the difference between AMD and Intel
processors than anything else.

Yousuf Khan
 
P

philo 

Just curious?

Which would run faster:

A Machine with (2) separate Dual Core processors
or
A machine with (1) Quad Core processor?

My thought is the double Dual core machine would be faster but I'm not
sure about Bus involvement?


The propagation time within a single (quad) cpu
would be less than that between two (dual core) cpu's...so if all else
the same, the quad core cpu would be the better choice.

I am sure someone could find an interesting exception however.
 
C

charlie

This probably has more to do with the difference between AMD and Intel
processors than anything else.

Yousuf Khan
The cost difference was the most significant to me.

I try to get a favorable compromise between cost and performance,
with a machine that has very good all-around characteristics.
I.E
Inexpensive case
~ 750W PS
5 year warranty hard drive(s)
Video card(s) are the major performance and cost variable ~$150-$900
4GHz 4 or 8 "core" "unlocked" processor
High reliability/upper end MBD
8Gg 1600 or preferably better DDR3 with decent specs.
 
J

Jeff Barnett

Artreid wrote, On 3/24/2013 9:24 PM:
Just curious?

Which would run faster:

A Machine with (2) separate Dual Core processors
or
A machine with (1) Quad Core processor?

My thought is the double Dual core machine would be faster but I'm not
sure about Bus involvement?
For the information of all of you who swing a bat at this post: check
the ACPI spec includes a method to specify affinity/closeness of
processors. In other words, such things as hyper-threading, multi-core,
multi chips on a motherboard, multi-motherboards, etc., can be
distinguished! The OS has the option to use this information and process
intercom information in making processing dispatch decisions that
determine where processes will execute. Whether any of our popular home
hobby-kit systems play the game I do not know. The reason I point this
out is that having the information available and using it properly can
make an enormous difference.
 
C

charlie

Artreid wrote, On 3/24/2013 9:24 PM:
For the information of all of you who swing a bat at this post: check
the ACPI spec includes a method to specify affinity/closeness of
processors. In other words, such things as hyper-threading, multi-core,
multi chips on a motherboard, multi-motherboards, etc., can be
distinguished! The OS has the option to use this information and process
intercom information in making processing dispatch decisions that
determine where processes will execute. Whether any of our popular home
hobby-kit systems play the game I do not know. The reason I point this
out is that having the information available and using it properly can
make an enormous difference.
Specs and implementation are two different things!
It's a challenge to find implementation details complete enough to
compare such things.
 
J

Jeff Barnett

charlie wrote, On 3/26/2013 9:46 PM:
Specs and implementation are two different things!
It's a challenge to find implementation details complete enough to
compare such things.
I know. However, I think most motherboards deposit the information in OS
mapped memory at cold (and maybe hot) boot time. The question is whether
anyone uses it. Other information, including bus structure, devices,
energy states, etc., is also deposited and used.
 
C

charlie

charlie wrote, On 3/26/2013 9:46 PM:

I know. However, I think most motherboards deposit the information in OS
mapped memory at cold (and maybe hot) boot time. The question is whether
anyone uses it. Other information, including bus structure, devices,
energy states, etc., is also deposited and used.
It's been so long since I bothered to "muck around" with the innards of
windows that I don't have the foggiest idea. My experience has been that
"reference" drivers from the hardware OEMs do not implement all
the features of the chipsets that they sell. BIOS implementations
also seem to share this quirk.
 
D

DanS

They're both processor intensive tasks which need to be done in real
time. Lots of memory is needed, as is lost of processor power.
Luckily however, video NLE s/w typically caches unchanged scenes/sections
for realtime full-speed/scale playback previews until it's changed again.

(Not to minimize the PITA it is having to wait while the PC does a slower
re-render for preview when edits are made.)

Many NLE packages support proxy editing as well, which let's you work
with a smaller video size/less CPU intensive codec while editing, but
will render with the original HQ/HD video as the source.

Again, not to minimize the fact that of course, more hardware equals
better performance, I just wanted to mention how the s/w deals with that.

Many people think you need an 8-core / 12 Gig RAM machine with expensive
NLE s/w to do any video editing effectively, but that is not the case.

(Obviously if you're editing for broadcast TV, or some high-level end-
product like that, you'd also have matching hardware, so I'm more
speaking of home users, or even videographers of whatever type.)

You'd be surprised what can be done with a $200 full HD camcorder and a
$50 s/w package, on a $300 commodity PC........

.......and of course, you need some talent and 'an eye', and have some
vision, etc, and need to put in some work learning at least the basics of
using the tools, the s/w "ain't gonna do it all for ya".

My only point is that if you think video production is something you'd
(and not 'you' to who I am replying) like to explore, don't be scared off
by the "seemingly" costly hardware requirements. Yes, a lesser PC will be
slower than some esoteric specially built rig, but just getting started
to even see if you like doing this, just for fun or in some professional
manner, doesn't cost large amounts of money.

(There's even one or two quite capable Linux-based NLE s/w packages that
are no-cost I use at times.)
 
J

J. P. Gilliver (John)

In message <[email protected]>, DanS
Many people think you need an 8-core / 12 Gig RAM machine with expensive
NLE s/w to do any video editing effectively, but that is not the case. []
My only point is that if you think video production is something you'd
(and not 'you' to who I am replying) like to explore, don't be scared off
by the "seemingly" costly hardware requirements. Yes, a lesser PC will be
slower than some esoteric specially built rig, but just getting started
to even see if you like doing this, just for fun or in some professional
manner, doesn't cost large amounts of money.

(There's even one or two quite capable Linux-based NLE s/w packages that
are no-cost I use at times.)
Yes; I do (very) basic video editing in VirtualDub on this XP machine,
with its single-core 1.6G processor and 2G RAM; OK, it renders at 1/2 or
1/3 real time, but does what I want.
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Top