kentsfield not a "true quad" -- real world meaning of this?

Toggle sidebar Toggle sidebar
G

graysky

Distinguished
Jan 22, 2006
546
0
18,980
Can someone explain all these articles flying around stating that Kentsfield cores aren't true quad core chips? What are the implications of this? I have seen the speed increases when encoding x264 on a kentsfield core vs. a conroe and they are ~45% faster @ the same clockrate in the test here.

Thanks.
 
It's because Kentsfield is pretty much two Conroes on a single chip that communicate over the FSB. The point is, though, it works pretty darn good. All the complaining about them not being "true quad core" is mostly just AMD fans trying to justify why their beloved company isn't keeping up to the big mean Intel.
 
Can someone explain all these articles flying around stating that Kentsfield cores aren't true quad core chips? What are the implications of this? I have seen the speed increases when encoding x264 on a kentsfield core vs. a conroe and they are ~45% faster in the test here.

Thanks.
Click to expand...

A: Marketing

A: Intel was able to sell Quad cores for 6 months before AMD and greatly increase their product line performance to jump way ahead of AMD.
 
jeff basically nailed why it's being said that way.

For Intel to acheive 4 cores within the same CPU package so quickly, it took 2 existing dual cores, threw them together into 1 package, and linked them via FSB and called it Quadcore.

it's basically the equivelant of 2 sockets with dual cores in each. Each pair of dualcores inside may share resources between themselves but all data that needs to be shared to the other 2 cores need to use the FSB instead of dedicated links and shared resources.

it works. It wll give you 4 cores of 4 core performance. But it's not 4 pure cores sharing resources like cache and controllers.
 
Can someone explain all these articles flying around stating that Kentsfield cores aren't true quad core chips? What are the implications of this? I have seen the speed increases when encoding x264 on a kentsfield core vs. a conroe and they are ~45% faster @ the same clockrate in the test here.

Thanks.
Click to expand...

The Kentsfields consist of two dual-core Conroe dies that sit under the same heat spreader, use the same LGA contacts, and share the same FSB. Since there are two physically separate dies on the CPU, it is called a multi-chip module (MCM.) The Pentium Ds were MCMs, as well as the Pentium Pro. The Pentium Pro had its processing core + L1 on one die and its L2 on another die, which made it an MCM. This also led to the funky oblong Socket 8 with different pin spacings on each side of center, but I digress.

There was never an official definition for what constitutes a "dual-core" or "quad-core" CPU. One definition is that a chip must have all processing cores integrated into the same silicon die to be considered a dual-core or quad-core CPU. That definition would not define any MCM chip as a dual- or quad-core chip. Others define dual- or quad-core by how many CPU cores are fit in a single socket, regardless of how many dies are underneath. All of AMD's multi-core CPUs use a single die due to their memory controller arrangement, whereas Intel has some MCM chips (Pentium D, Core 2 Quad) and some single-die chips (Core Duo, Core 2 Duo.) As such, AMD is in the single-die definition camp whereas Intel is in the other.

I personally agree with Intel in this matter as the main attraction of multi-core chips is being able to use multiple processing cores without buying a multi-socket motherboard. So it does not matter if the processing cores are not all on one die as long as they will fit in a single socket. The MCM route is seen as a quick-and-dirty approach as it takes very little engineering for Intel to simply stick two dies on a PCB and call it a multi-core CPU. Single-die multi-core CPUs do have advantages over MCMs in bandwidth and power control, but they also take longer to bring to market and suffer from lower yields due to a bigger single die. I think that AMD is playing a little bit of sour grapes in the "not a true n-core CPU" bit, especially since they are supposed to be bringing out an MCM 8-core CPU in a couple of years.
 
The main reason for this is AMD's mud slinging marketing, since they were caught enjoying the fruits off their laurels, Intel backslapped them with a quad core that performed... and the fight should have been on... but AMD has yet to respond...
 
It takes Intel little engineering to put two dies on a CPU because their FSB can be shared between the two dies as easily as it can feed one single die. Intel's engineering work had been done when the current AGTL+ FSB spec was designed, and they can sit back and use it without modification in newer chips. Intel just has to put a little bit of circuitry to facilitate the FSB sharing in the CPU package, make sure that the dies are all wired up and won't exceed a reasonable TDP, and away she goes. This requires the chipset to support FSB sharing, but an MCM sharing a FSB is the same as two CPUs sharing the FSB, and Intel has been doing that ever since the first Pentium SMP setups. FSB sharing support has been included in all Intel desktop chipsets since the 865, excepting the 915 and 925.

AMD would have to do a lot more work as their current on-die memory controller pretty much requires that all cores be on a single die or they would need to make a new socket for two IMCs or slave one die off the other using HTT, which probably would not perform very well. Or they could redesign the CPU to have MCM operation in mind. The Barcelona has two independent 64-bit memory controllers in it. The official word is that the two independent memory controllers are used to allow for synchronous reads and writes, which is not possible with the current 128-bit IMC, and that would improve performance. They are probably right, but it also lets them shut off one IMC so that they can put two dies in a package, each with one IMC working.

So in summary, it would take a CPU redesign due to the IMC and the Barcelona is designed to allow for an MCM. AMD even says they will release the "Shanghai" 8-core MCM chip.
 
Cool, thanks for all the great replies, all. I suppose the question I need to answer is: can I run a q6600 @ 9x333 and cool it on air (tower or ultra-120 likely) safely? I've heard that chip @ 9x333 would likely hit 65-70C easily under a full load.
 
You should be able to. The Core 2 Quad QX6800 runs at 2.93 GHz, which is just a hair shy of the 3.00 GHz that your Q6600 at 9x333 runs. The QX6800 is supposed to have a TDP of 150 watts or thereabouts. Intel is thinking that its stock HSF should be enough to cool it, so my guess is that a big air cooler will work fine for you.

One thing that can help is to make sure you have plenty of air going through your case so that the Tcase is decent. If the air inside the case is warm, you have a much harder time cooling things inside off, no matter how big the heatsink is. Also be mindful of your room's temperature. I wouldn't run the chip overclocked that much if you have a room up in an attic or you don't have A/C. A chip that runs fine in a 22 C room might not run so well in a 32 C room.

And if you do this and air isn't enough, there's always water, phase-change, and TEC (Peltier.)
 
I have my QX6700 with no IHS on the cpu..and water cooled with the
big water 745 (risky to have a heavy fan on a cpu with no IHS)..
so far my temps are idling at 39 C (was 36 C before i installed an 8800 GTX)..and on load its 55-58C. Performance wise its a killer.
Rendering on LIGHTWAVE is almost cut down to 1/4th of the normal time from my 4600X2 ..so yes..its worth it and props to intel to put 4 cores on one chip.
 
Internal Heat Spreader...the metal piece that covers the CPU's insides
Its good for protection..but in the quads taking it off can take the temps off like 15 degress ..(I love taco's) helped me alot with info on setting up the water cooling for it. I did some overclocking just to test it out..
took it to 3.0 ghz and it stood on 45C idle..which is still very nice and cool
compared to the temps i read before...

p.s. taking off the IHS seems like an impossible risky thing to do..
from what i heard 4 outta 5 quads die if the IHS is taking off wrong..
so i guess im lucky with mine.
 
= amd speak for: we dont have a quad core but when we do it will be 4 cores on one die not 2 cores on 2 dies

big deal what you rather have - a kick azz intel or nice (but slower) and very cheap amd x2?
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom