"True" Dual/Quad Core

[SocrPlyr]

Alright after reading a couple of other threads over time I have some questions here. Why do people consider an AMD X2 processor a "true" dual core, while not considering let's say the dual P4? I have some ideas, but I want to hear what other people have to say first.

Edit:
Removed topic summary ("oh how i love the ignorance"), I didn't mean it to be there the way I wound up writing the original post. Sorry.

 

[AmberClad]

The AMD dual/quads are true/native because they have all the cores on a single die, not a couple of cobbled together single/dual core dies in the case of Intel's stuff. From what I've read, the native multicore solution is a bit more efficient as far as communication between the various cores and memory latency.

So, with everything else being equal and assuming hypothetically that AMD's CPUs preform on par with Intel ones, then it seems to me that a "true" or "native" multicore would be slightly preferable.

I'm totally open to other views on the matter though.

 

[coldpower27]

Marketese, for the MCM vs Native debate. It's like as someone said, all other factors being equal then a Native solution is superior to the MCM version with regards to performance. The problem of course that most people don't understand of fail to consider is that all other factors aren't equal. They are all indeed Dual Core processors like the Pentium D/Core 2 Duo/Athlon 64x2, they are just implemented differently, the "true" term was a way of trying to differentiate the Pentium D from the Athlon 64x2/Core 2 Duo.

 

[Viditor]

Originally posted by: SocrPlyr
Alright after reading a couple of other threads over time I have some questions here. Why do people consider an AMD X2 processor a "true" dual core, while not considering let's say the dual P4? I have some ideas, but I want to hear what other people have to say first.

The only real effective difference between an MCM (Multi-Core/Chip-Module) and a native solution will be what's called Cache Coherency (though this is changing for things like Torrenza and Fusion, but I digress...).

Say a cache line in the L2 of Core 0 needs to get to the L2 of Core 1...
For a native dual core like X2, the path is directly "next door" through the crossbar switch...or with Core 2 Duo, it's even closer as it's a shared L2 cache.
For an MCM, the cache line must first go through the FSB before it gets to the core "next door", this adds significant latency to cache coherency. Keep in mind though that we're not talking about a huge amount here, but it also adds "traffic" to the FSB. You can imagine that as you keep adding cores, at some point this can bog things down a little bit (though again, not a huge amount).
For all intents and purposes, that's it in a nutshell...

 

[TC91]

technically speaking, a true / native dual/quad core with the same architechture (like a hypothetical native core 2 quad) would have less heat production and would have slightly better performance due to having one shared l2 cache but that difference is not huge or really noticeable for most people apart from say multi socket servers. the mcm chips however are easy to make compared to the native cores and can lead to lower prices provided there is some competition in the market.

 

[swtethan]

I really dont care if its native or not, as long as it shows 4 cores and performs well.

 

[SocrPlyr]

Originally posted by: swtethan
I really dont care if its native or not, as long as it shows 4 cores and performs well.

finally a post that matters
That doesn't mean i disagree with some of you but we are consumers, thus the end result is all that really matters.

Question for the Athlon X2:
When it requests stuff from the cache of the other cores does it go through the HT link?

Josh

 

[jonmcc33]

Originally posted by: AmberClad
The AMD dual/quads are true/native because they have all the cores on a single die, not a couple of cobbled together single/dual core dies in the case of Intel's stuff. From what I've read, the native multicore solution is a bit more efficient as far as communication between the various cores and memory latency.

So, with everything else being equal and assuming hypothetically that AMD's CPUs preform on par with Intel ones, then it seems to me that a "true" or "native" multicore would be slightly preferable.

I'm totally open to other views on the matter though.

The Core 2 Duo are native dual core. The Pentium D is a pair of separate die on the same package.

 

[taltamir]

intel tends to first make an MCM and then evolve into a "true" multi cpu core... AMD waits until they get it in "true core" mode...
There is a also a big power/heat reduction in having it as "true" multi core... I say multi core and not duel because intel made their first quad cores by putting two duel core dies on a single chip...

With that being said though... the end result is what matters. The AMD X2 beat the pants of the pentium D. But the Core 2 processors, duel or quad, have been kicking amd in the cahoonas (with the core2 due being one duel die, and the quad being two duel dies on a single chip). AMD is about to release their "true" quad core CPU... but so far it doesn't look like they will be retaking the crown from intel with this one...

When it comes down to it, you should care about the "Trueness" of multi core asmuch as you care about the amount of pipelines or pixel shaders or ALUs or Mhz the the thing is tauting... It is all completely insiginificant to the customer. It is all engineering decisions. The only thing that should matter to a customer is
1. How much does it cost
2. How well does it perform in real world applications

with performance including both speed of processing, aswell as heat / power issues... depending on the user.

PS speed of processing does not mean GHZ or what not, it means how fast it performs comparative tasks... like rendering the same file on two different machines, or opening the same document, or getting so and so FPS on otherwise identical machines... thats what I mean when I say speed of processing in regards to real world performance

 

[TC91]

i personally prefer a native multi core cpu instead of a mcm but that is just my personal opinion, since it feels more "right" to me lol.

 

[taltamir]

I admit it "FEELS" more right... there is a psychological aspect here... but if they give me something that outperform that then who am I to complain?

 

[TC91]

Originally posted by: taltamir
I admit it "FEELS" more right... there is a psychological aspect here... but if they give me something that outperform that then who am I to complain?

i couldnt agree more

 

[jonmcc33]

Originally posted by: taltamir
With that being said though... the end result is what matters. The AMD X2 beat the pants of the pentium D. But the Core 2 processors, duel or quad, have been kicking amd in the cahoonas (with the core2 due being one duel die, and the quad being two duel dies on a single chip). AMD is about to release their "true" quad core CPU... but so far it doesn't look like they will be retaking the crown from intel with this one...

That's because the Pentium D was nothing more than a pair of Prescott processors on the same package. The Prescott was already getting roasted by AMD's single core line. So it has nothing to do with the fact that the Pentium D is a pair of die on the same package.

What Intel did that was smart was drop the prices of dual core. I got my AMD Opteron 165 for $330 at the time of purchase. Because Intel's dual cores were so low in price it forced AMD to drop prices too.

So going non-native dual core was a huge benefit to everyone, wouldn't you say?

 

[sonoran]

Originally posted by: SocrPlyr
Alright after reading a couple of other threads over time I have some questions here. Why do people consider an AMD X2 processor a "true" dual core, while not considering let's say the dual P4? I have some ideas, but I want to hear what other people have to say first.

To me the definition of "true dual core" is the chip has two full x86 cores in it. Native vs MCM is largely marketing fluff, since as others have pointed out performance and price are all that really matter to the consumer.

Now in the future, saying how many "cores" a chip has may be a lot tougher, since the cores may not all be full-fledged x86 cores. Chips may end up being more akin to the cell processor in design.

 

[Viditor]

Originally posted by: SocrPlyr

Originally posted by: swtethan
I really dont care if its native or not, as long as it shows 4 cores and performs well.

finally a post that matters
That doesn't mean i disagree with some of you but we are consumers, thus the end result is all that really matters.

Question for the Athlon X2:
When it requests stuff from the cache of the other cores does it go through the HT link?

Josh

No...here's a block diagram of it...
Data flows through the crossbar switch.

 

[bradley]

True native multi-core solutions are more than just a novel concept, unlike something like Hyperthreading. Otherwise it wouldn't be part of Intel's roadmap. They know that eventually diminishing returns would catch up with them. Eventually it would have meant as much on the desktop as it currently does on the server side. And it unquestionably matters on the server side, certainly multiple power planes will too.

 

[TuxDave]

Originally posted by: AmberClad
The AMD dual/quads are true/native because they have all the cores on a single die, not a couple of cobbled together single/dual core dies in the case of Intel's stuff. From what I've read, the native multicore solution is a bit more efficient as far as communication between the various cores and memory latency.

So, with everything else being equal and assuming hypothetically that AMD's CPUs preform on par with Intel ones, then it seems to me that a "true" or "native" multicore would be slightly preferable.

I'm totally open to other views on the matter though.

Which is why marketing exists. To make people have a bias when there's no reason to.

 

[Viditor]

Originally posted by: TuxDave

Originally posted by: AmberClad
The AMD dual/quads are true/native because they have all the cores on a single die, not a couple of cobbled together single/dual core dies in the case of Intel's stuff. From what I've read, the native multicore solution is a bit more efficient as far as communication between the various cores and memory latency.

So, with everything else being equal and assuming hypothetically that AMD's CPUs preform on par with Intel ones, then it seems to me that a "true" or "native" multicore would be slightly preferable.

I'm totally open to other views on the matter though.

Which is why marketing exists. To make people have a bias when there's no reason to.

To be fair, native vs MCM does indeed have an effect...and that effect grows as you add more cores. The thing is that we just don't see it until you get up to 8 cores or more, so desktop systems will not have any measurable difference, but Enterprise servers certainly do.
If there wasn't a difference, I can't imagine why Intel would spend so much R&D money to go native on their quads and beyond.

 

[bradley]

Stuff like native design and on-die memory controllers only become important when implemented by Intel, or so it seems. Otherwise, native processes will probably be as imperative to furthering multi-core cpu design as process shrinkage was to single cores. AMD probably made a mistake by looking too far into the future. However, they also will have worked out all the kinks way before Nehalem arrives. Perhaps that gives them a slight edge in a year. My guess is probably as good as anyone's.

 

[TuxDave]

Originally posted by: Viditor

Originally posted by: TuxDave

Originally posted by: AmberClad
The AMD dual/quads are true/native because they have all the cores on a single die, not a couple of cobbled together single/dual core dies in the case of Intel's stuff. From what I've read, the native multicore solution is a bit more efficient as far as communication between the various cores and memory latency.

So, with everything else being equal and assuming hypothetically that AMD's CPUs preform on par with Intel ones, then it seems to me that a "true" or "native" multicore would be slightly preferable.

I'm totally open to other views on the matter though.

Which is why marketing exists. To make people have a bias when there's no reason to.

To be fair, native vs MCM does indeed have an effect...and that effect grows as you add more cores. The thing is that we just don't see it until you get up to 8 cores or more, so desktop systems will not have any measurable difference, but Enterprise servers certainly do.
If there wasn't a difference, I can't imagine why Intel would spend so much R&D money to go native on their quads and beyond.

But then you'd be breaking the initial premsis of 'all else equal' which was what I was pointing out.

 

[firewolfsm]

They're all right, but what you fail to realize is that in the most apps (say...en/decoding) P4 dual cores had gains in the upper 90%s. Sometimes 98% gains showed. And that was with an 800MHz FSB. It never made too much of a difference.

In fact...it might just be better, because of better yields, we could get cheaper processors.

 

[Viditor]

Originally posted by: TuxDave

Originally posted by: Viditor

Originally posted by: TuxDave

Originally posted by: AmberClad
The AMD dual/quads are true/native because they have all the cores on a single die, not a couple of cobbled together single/dual core dies in the case of Intel's stuff. From what I've read, the native multicore solution is a bit more efficient as far as communication between the various cores and memory latency.

So, with everything else being equal and assuming hypothetically that AMD's CPUs preform on par with Intel ones, then it seems to me that a "true" or "native" multicore would be slightly preferable.

I'm totally open to other views on the matter though.

Which is why marketing exists. To make people have a bias when there's no reason to.

To be fair, native vs MCM does indeed have an effect...and that effect grows as you add more cores. The thing is that we just don't see it until you get up to 8 cores or more, so desktop systems will not have any measurable difference, but Enterprise servers certainly do.
If there wasn't a difference, I can't imagine why Intel would spend so much R&D money to go native on their quads and beyond.

But then you'd be breaking the initial premsis of 'all else equal' which was what I was pointing out.

Woops...fair call.
I guess I better move that Optometrist appointment up...:roll:

 

[manowar821]

While the arguments presented in this thread have been good ones for the most part, I'm going to go ahead and say the obvious... The OP is a dirty troll.

 

[SocrPlyr]

how does that make me a troll? I just wanted to see what other thought. I myself believe there should not be any distinction. In the end it doesn't matter to the consumer how it was made. Maybe it would matter if say gluing them together had a higher failure rate or something like that. As for how things are actually implemented between AMD and Intel, at least at first glance, they are actually very similar. The difference is that AMD uses a "cross bar switch" (Posted by Viditor) and Intel uses the FSB. Clearly the cross bar switch should be superior and thus why AMD scales better, but in my mind that doesn't make intel's method any less valid. You take into account the location of RAM on the intel system and theirs makes even more sense for them. This whole question was to see what other people though because i ran into a post in General Hardware that was asking what processor to get and it was actually used as an excuse that it wasn't "true" quad core not to get a Q6600, which I didn't understand why a consumer would even care about this. I wanted to see if this was a common thought. Now if I were building a server with 2 processors and going to scale it to 4 then it would matter. Now these are only one processor with several cores, you know the performance and don't have to worry about scaling with other things. Basically, for the most part I believe there is no such thing as "true" dual/quad cores, because there is probably always someway that you can better integrate them. So the argument can go on and on. As for the best way to do it is probably what Intel did w/ the Core Duo, but that takes a lot to integrate at that level, followed by what AMD did with the X2 (I don't know how Barcelona is implemented). However, none of this matters because you can't pick and choose and if the Intel w/ an inferior method outperforms the AMD why care.
I do take offense to being called a troll the meaning of which is not clear in your statement, I just wanted to see what other thought.
Josh

 

[Shortass]

Originally posted by: SocrPlyr
Wall of text

It's a troll because your subtitle is "oh how i love the ignorance" and the only thing you responded to was a subjective opinion of the matter, ignoring the 3-4 technical explanations that actually contribute to the thread.