Quad - AMD vs Intel

Black Morty Rackham said:
Look, enough with this "fake quad" bullshit. What Intel are selling has a perfectly reasonable name: MCM, or multi-chip module. It's a single module, with two chips on it. Both chips have two cores.

It's kind of like IBM's POWER series, and nobody in his right mind would call the POWER5 a "fake eight-core" processor.
Click to expand...

Alright its just wat Ive heard, still could wait till summer buhaha:D and if amd will pull it off itll be amd againg, - kinda makesme surprised:eek:
 
Black Morty Rackham said:
Look, enough with this "fake quad" bullshit. What Intel are selling has a perfectly reasonable name: MCM, or multi-chip module. It's a single module, with two chips on it. Both chips have two cores.

It's kind of like IBM's POWER series, and nobody in his right mind would call the POWER5 a "fake eight-core" processor.
Click to expand...
Since this thread has been completely derailed at this point:
The jury about 'native' vs MCM quad-core is still out. Until we have a single-die implementation, we cannot compare whether one is superior to the other, or they are indistinguishable. I assume that the difference, if it does exist, is rather small.

Narisatu said:
If it provides a decent benefit over quad core, im sure more people will go after it,
Click to expand...
It depends what segment we are talking about. It seems that many people on this forum could care less about power draw, at least for their ``main rig''. From an OEM perspective, a lower power consumption makes the whole PC cheaper, since the PSU does not need to be as beefy.

Narisatu said:
If barcelona proves to be faster than Intel, I will be buying a Quad-FX system, put i dont pay ultilities.. :)
Click to expand...
I do pay utilities and by the time Barcelona comes out, I will be paying the high energy prices common in continental Europe. Nonetheless, if Barcelona proves significantly faster than C2D, I will certainly consider to get a system.
 
the overclocking on it wont be as good because it will chuck out more heat!
 
why? even if the amd's are faster if they run hotter you might be able to overclock an intel to be faster.
 
ND40oz said:
That article is over a year old, pretty sure whitefield has been replaced with tigertown, it's due Q3 07. Either way, these are both for MP systems.
Click to expand...

The sooner we see benchmarks- the better 4 amd...
 
Narisatu said:
Intel is selling a Quad, but it isnt a native quad. It is literally a pair of Conroe processors glued next to each other.
Click to expand...

What specific benefit would a "native quad" have over the QX6700?
 
Fox_T said:
If Im correct they ment 2 have 16mb L2c shared
Click to expand...

I'm sorry, but I can't understand your answer. Assuming "ment" means "might", which "they" are you referring to? The "native quad" parts? Why is it perceived that shared cache is better?
 
what it is. qx6700 has 2 lots of 8mb cache. 2 cores share 8mb. therefore the other 2 cores can't access that cache and loses slight performance compared to a native quad which has 16mb for any core to use.
 
Teknokid said:
what it is. qx6700 has 2 lots of 8mb cache. 2 cores share 8mb.
Click to expand...
What what is? The QX6700 chips have two blocks of 4 megs, not two blocks of eight.

Teknokid said:
therefore the other 2 cores can't access that cache and loses slight performance compared to a native quad which has 16mb for any core to use.
Click to expand...
I don't see what's "native" about cache arrangement.

Meanwhile, why do you believe that the shared cache is a benefit? Is there research that says misses happen because of eviction happen more frequently at that cache size than misses because of collisions?
 
sory abit tired.., inquirer or some other site said they r gona to have 16mb if its true, if yes, which is most probable then 16 is beter than 8,- 4mb per core.
 
got no clue to what "shared" is for, but I dont think that its a good thing...
 
but dont think its their main difference, probably clocks will b higher 2
 
hm.. i dont know about that as such. they should be more energy efficient though.
 
well they better b, coz if its just 16mb n low power consumption, than might as well go 4 "quad" now or c2d
 
Teknokid said:
sorry i mant 4mb.. surely 8mb total cache is better than 2 lots of 4?
Click to expand...
Sure. But cache size is independent of core design.

So, we've identified no particular reason for the previous implication that "native quad" is somehow better than the "non-native quad" design that Intel ships. Is that correct?
 
The problem with these compairisions, is a "native" quad core chip is normally a different architecture than two dual cores slapped onto the same die. The Pentium D was intels 'dual core', while the Core Duo was a native quad core. The architectures where different, one was two P4s slapped next to each other, while the Core duo was a modified design of the pentium Ms, and had more than twice the number of transistors than two pentium Ms, which means that the P-M design underwent optimizations for running in dual core.

Same with AMD and how i have heard some people talking about Barcelona as just a die shrink with two current x2s on the same die.. the Barcelona has about 60 million more transistors than 2 x2s... Now, in my opinion, i doubt that 60 million transistors are needed to connect two cores.. So thats A LOT of optimizations and modifications. of course, I have heard, and am pretty sure most will agree, that Barcelona is a completely new Architecture. Speaking of which, i really cant wait to see it.
 
Narisatu said:
The problem with these compairisions, is a "native" quad core chip is normally a different architecture than two dual cores slapped onto the same die
Click to expand...
How does the architecture differ, specifically?

I don't think the transistor count comparison between the Penium M and the Core Duo series holds. The Core Duo contained implementation changes local to the core, such as improvements in SIMD instruction handling. The difference in transistor count might be anywhere on the chip.
 
kush said:
I would *prefer* "true" quad as AMD will be releasing soon, or as Intel will likely do in the not too distant future (ie: the PD[/CoreDuo?] -> Core 2 Duo transition)..
Click to expand...
Why, specifically?

I'm really interested in learning why "true" or "native" quad-core is perceived to be so much better.
 
mikeblas said:
Why, specifically?

I'm really interested in learning why "true" or "native" quad-core is perceived to be so much better.
Click to expand...
Because they can ship data between all four cores without going via the FSB. It's faster. Still, the difference probably isn't all that big.
 
mikeblas said:
What data would move from one core to the other, and when?
Click to expand...
All data, all the time. A single task is generally not tied to a single CPU core. It moves back and forth all the time, which means the related data in the cache will be shipped around as well.
 
Black Morty Rackham said:
All data, all the time. A single task is generally not tied to a single CPU core. It moves back and forth all the time, which means the related data in the cache will be shipped around as well.
Click to expand...
Such data doesn't move from core to core; it moves from cache to cache. (I guess you'll argue about wether the cache is a part of the processor core or not). Either way, the data isn't "Shipped around". If the cores involved in a context switch share cache, then the data is just there in the cache. If the cores don't share cache and the new core has to re-populate the cache with the fetched data.

The scheduler in the OS knows that the cost of moving from one processor to another is more or less based on its analysis of the processor architecture. It can decide, then, to try to keep a given thread on cores that share cache and only decide to move to a different cache if it thinks the cost of the cache misses outweigh the ability to get the quantum scheduled sooner.

While avoiding repopulation at task switch time is one advantage of a shared cache architecture, I don't think there's any research about about its benefit over an isolated cache approach. An isolated cache doesn't have the problem of cores competing for cache lines, and makes it more appealing for data-intensive applications.
 
mikeblas said:
I'm sorry, but I can't understand your answer.
Click to expand...
I think you are not alone in that. I cannot understand about a good quarter of what was said in this thread.

mikeblas said:
kush said:
I would *prefer* "true" quad as AMD will be releasing soon, or as Intel will likely do in the not too distant future (ie: the PD[/CoreDuo?] -> Core 2 Duo transition)
Click to expand...
Why, specifically?
Click to expand...
Rumor has it that AMD's next chip design, Barcelona, is between 10% and 40% faster than a Core 2 chip. I am not sure if AMD's claims are clock-for-clock or just in general.
mikeblas said:
I'm really interested in learning why "true" or "native" quad-core is perceived to be so much better.
Click to expand...
Because AMD's marketing department has done an excellent job.
 
drizzt81 said:
Rumor has it that AMD's next chip design, Barcelona, is between 10% and 40% faster than a Core 2 chip. I am not sure if AMD's claims are clock-for-clock or just in general.
Click to expand...
That's possible, but I haven't read anything that would attribute the increase in performance to this "native" multi-core idea.
 
You must log in or register to reply here.
Back
Top