Intel's Quad Core not considered a native quad core?

[Eurasianman]

Can someone explain this more in detail.

I think I read somewhere that the reason Intel's Kensington (sp?) isn't a native quad core is because it's just two Conroe's on one chip. But still...

If someone could explain it to me, that'd be great

It just sort of surprises me that I've googled and searched the forums and couldn't really find anything that explains the differences, even though quad core CPUs have been out for quite some time now. *shudders*

 

[dasickninja]

You mean Kentsfield? Technically and in all actuality, its not a monolithic, or for simplicity's sake, native quad core since there are two dies (chips) on one package. The technical name for this would be MCM or multi-chip module. AMD uses monolithic multi-core designs, although currently they have only a dual core option. Each design brings its own advantages and problems.

Generally, when you hear someone talking about native vs "glued", ignore it, since usually fanboyism follows.

 

[NMDante]

Kentsfield?

Yes, it's not a true quad core, in the sense that it's not a monolithic core, meaning, 4 cores on one die. It's essentially, 2 dual cores using the same communication "bridge" to act as 4 cores at one time.

That's my easy to understand definition. I'm sure I'll get corrected or you'll get more detailed information on this.

 

[NightlySputnik]

Can someone explain this more in detail.

I think I read somewhere that the reason Intel's Kensington (sp?) isn't a native quad core is because it's just two Conroe's on one chip. But still...

If someone could explain it to me, that'd be great

Kentsfield is effectivly a dual-dual-core if I can put it this way. There is 2 dual-core on the same package. While performances might be a tad lower then a true monolitic quad-core, this is only partially true.

For true multitasking, as in running 4 applications at the same time, there is no difference in performances for every application running if you compare true and "fake" (if we can say that) quad-core. The ressources are not shared. When it comes to multithreaded applications things might change. Since different parts of the said application is being worked on by different core, informations gets exchanged. In an Intel C2D, when that happen, informations get stock in L2 and can be reused by any core when needed. In a "fake" quad-core, informations has to go out to the front side bus and back in the other dual-core. If the FSB gets overload, there is a lost of performances from this. Usually tough, that doesn't occur, so quad-core is still better, at same frequency, then dual-core, especially when multitasking a lot. Considering that explaination tough, even an A64X2 is not a true dual-core, since information has to go on the HT links to gets exchanged.

That "fake" quad-core thing should get fixed in upcoming AMD's AGENA/BARCELONA (Q3-2007)and Intel's NEHALEM (Q2/3-2008). These will be real quad-core, (altough not sure yet on Intel side, but the opposite would be really stupid on there part) maybe even octo-core.

 

[cwj717]

That "fake" quad-core thing

I really don't understand why people keep calling it a "fake" quad... there really are 4 cores... just on two separate die. As for the FSB, I haven't seen any benchmarks that show the FSB is saturated under normal workloads on a QX processor... wasn't one of the forum members going to run some benchmarks on a QX6700 with different FSB clocks to see if there was much of a performance difference (JJ?).

 

[mpjesse]

As some have said, Kentsfield isn't "native" quad core for 2 reasons:

The CPU package has two seperate dies
It uses the FSB to communicate between dies

An example of a native quad core would be 1 single die with 4 cores and with no external communication method for dies to communicate between. Using a single die design eliminates slow and cumbersome bus speeds and allows all the processors to manage cache in a more efficient way (at least in AMD's design). Intel still isn't using a L3 cache for Conroe based CPU's; each core has its own cache. But I think that's going to change...

 

[cwj717]

Intel still isn't using a L3 cache for Conroe based CPU's; each core has its own cache. But I think that's going to change...

That isn't quite true... they don't have an L3 but there is a shared L2 cache for every two cores.

 

[mpjesse]

Sorry, I meant each DIE. Thx. ;-)

 

[cwj717]

Generally, when you hear someone talking about native vs "glued", ignore it, since usually fanboyism follows.

Wouldn't the QX be an example of 2-way glueless SMP anyway?

 

[Andrius]

Why does it matter?

The only real world disadvantage could be from a performance point of view since it was not designed to work that way (more of a rabbit out of the hat thing).
A native quad core design would take the larger memory bandwidth usage of four cores into consideration. Intel's (early stroke of genious) solution for this is the extra large cache for 2 cores.

Since memory bandwidth is not really a handicap for the quads, except in some very specific mostly benchmark and server applications everyone who says it's less of an achievment is either a fanboy or just ignorant.

Other posible disadvantages for a two die sollution would include voltage, heat output, (over)clockability differences from using two chips in one package (binning of chips matches those up so you don't notice a thing).

The bottom line is Intel has a quadcore package on the market. AMD has no such thing. Native or not it works as a quad. Software for it is another story.

 

[djgandy]

Question...

2 x Dual core K10's

VS

1 X Quad core K10

Same frequency w/e

Which wins?

 

[accord99]

Question...

2 x Dual core K10's

VS

1 X Quad core K10

Same frequency w/e

Which wins?

Most likely the quad-core K10 because of the memory latency impact of the second socket on AMD's platform; similar to how a dual-core Opteron beats the same frequency dual socket single-core Opteron in almost every application.

 

[TechnologyCoordinator]

It's not monolithic, but it does the trick with very few trade-offs. Most workloads will not be bottlenecks by the FSB. If you want to see FSB bottlenecks you have to use FAKE, sythetic benchmarks.

Currently AMD has no quad-core solution. They do have a dual-socket pro-sumer product, but it hasn't really gained much adoption.

 

[NightlySputnik]

I totally agree with you. That's why I put fake like that... "fake".

Where I don't agree with you is about the performances gains. Cinebench goes from 1.88X gain for dual-core C2D (out of 2x possible) to 3.3x gains about (out of 4 possible) for quad-core C2Q. Not exactly bad, but it shows that the solution might not be ideal.

Anyway, I'll personally get a Q6600 this summer as soon as the price goes below 300$, so put me in as a pro-quad anyway.

 

[Ycon]

There is just no difference.
You just have to pay less for a non-native multi-core CPU, since production is way cheaper.
Native multi-core in fact does yield minor reductions in power draw because in MCM designs there always is additional silicon that is not needed but still has to be powered.

 

[Eurasianman]

Thank you all for your time and knowledge. I tried googling this information, but not much of it made sense. Wikipedia didn't really help either.

Well, unfortunately for my motherboard, I don't think it will ever be able to support Intel's Nehalem due to the microdes and chipset. But oh well, by then, my computer will have hit the 2 year mark (I have a bad tendency to build a new computer every 2 years).

Here is my next question in pertaining to the topic somewhat :?

Will Intel's 975X chipset as well as the 965 chipset be able to support a true native quad core? Or perhaps a more clear question would be, is Intel going to make a native quad core that will work with these chipsets? (including nVidia's 6x0 chipset)?

 

[bfellow]

No one mentioned the QuadFX which is 2 dual-core chips needed on 2 separate sockets

 

[ches111]

Nightly,

The problem here does NOT lie with the die/process/MCM/Monolithic/...

What you are describing is the problem of scalability..

Now I am NOT saying that the processor arch does NOT have anything to do with this. As it does. However it is far from the only reason. What you will find is that the system as a whole has much to do with this issue.

What do I mean by system?

OS (probably the largest culprit for issues of scalability),
Memory (not just speed but controller as well),
Storage subsystem (how are your drives set up what bandwidth is needed and how is it provided),

The software being run also play a large part in this. Is it optimized for the inclusion of two additional processors (does it even care?)?

All of these thing affect how a system scales.

I totally agree with you. That's why I put fake like that... "fake".

Where I don't agree with you is about the performances gains. Cinebench goes from 1.88X gain for dual-core C2D (out of 2x possible) to 3.3x gains about (out of 4 possible) for quad-core C2Q. Not exactly bad, but it shows that the solution might not be ideal.

Anyway, I'll personally get a Q6600 this summer as soon as the price goes below 300$, so put me in as a pro-quad anyway.

 

[InteliotInside]

Thank you all for your time and knowledge. I tried googling this information, but not much of it made sense. Wikipedia didn't really help either.

Well, unfortunately for my motherboard, I don't think it will ever be able to support Intel's Nehalem due to the microdes and chipset. But oh well, by then, my computer will have hit the 2 year mark (I have a bad tendency to build a new computer every 2 years).

Here is my next question in pertaining to the topic somewhat :?

Will Intel's 975X chipset as well as the 965 chipset be able to support a true native quad core? Or perhaps a more clear question would be, is Intel going to make a native quad core that will work with these chipsets? (including nVidia's 6x0 chipset)?

No, the 975X, P965, P35 or X38 chipsets won't be seeing a native quad core. However, you can easily buy a quad core (Q6600 for example) for your P5W-DH! What's better is that the Q6600 will be dropping to around $280-300 (US) by the end of July!

 

[Eurasianman]

Yea... I knew that. But I wasn't sure about the whole performance issue and technical data between a native quad core and a non native quad core.

I'll probably get a Quad Core right before they phase out :tongue:

Perhaps than, they'll be like the the Intel 9x0 series, where you can buy one for around ~$150, give or take

 

[petevsdrm]

I just wanted to post that this is the most informative and interesting thread I have read around here in a while. :trophy:

I actually learned something, thanks to all who contributed.

 

[Eurasianman]

I agree! Some of this information should be posted on wikipedia! :tongue:

But yes, I'm sure there are more details that we could go into about the difference between a "native" quad core and a "non-native" quad core.

It just sort of surprises me that I've googled and searched the forums and couldn't really find anything that explains the differences, even though quad core CPUs have been out for quite some time now. *shudders*