Sign in with
Sign up | Sign in
Your question

The "True Quad Core" Question answered the long way...

Last response: in CPUs
Share
April 20, 2007 8:49:20 PM

Intel creates multi-core processors by combining die to form a multi-chip package, while AMD’s yet-to-be-seen quad-core uses a monolithic design. There has been some complaining about it as well.
In the end it is the quality of the product and not whether it is 4 cores on one die or two dual-cores put together that will make a difference.
So here is a description about the difference between the monolithic and multi-chip package (MCP) manufacturing methods.

For the monolithic approach, all cores are manufactured on a single die. The Intel E6700 has two cores on one die. The multi-chip package combines two chip die on a single chip. That is, the dual-core die can be packaged either individually as dual-core processors or together as quad-core processors. Intel's front-side bus (FSB) architecture allows this flexibility. Having a more complicated point-to-point bus architecture, like AMD’s HyperTransport, would result in a more difficult conversion and a prohibitive loss of performance if used for a multi-chip package. That is why AMD is slow to create new products or respond to Intel lately.
“Yield” refers to the number of usable chips on a wafer. Yield is higher for the smaller die sizes used in MCP processors. With the larger die for monolithic processors, you’re likely to discard more chips due to defects during the manufacturing process. MCP chips are more efficient for Intel to produce and allow more flexibility. First, Intel can use cores from the same wafers in dual- and quad-core processors, as in Woodcrest and Clovertown. Second, the difference in production yields increases the cost of a monolithic processor by about 15 percent. Intel has used MCP to get dual- and quad-core processors into production ahead of the competition, as it did with Clovertown and the performance is equal in either case.

The monolithic approach is a choice to be made when the customer requires extremely high performance and efficiency and can tolerate a higher cost for it. Intel's usage was the quick, less expensive, and easy way to get the product out first while keeping quad core performance. If they are designed for faster access to memory and communication between the cores, monolithic processors can offer higher performance within the same power envelope. Some estimate it would be approximately 10-15 percent higher.
April 20, 2007 9:42:41 PM

Quote:
Please post here more.


I wonder where this guy just copied and pasted this information from....Intel??? LOL :lol: 
Related resources
April 20, 2007 10:51:53 PM

Quote:
Please post here more.


I wonder where this guy just copied and pasted this information from....Intel??? LOL :lol: 

I retract my previous statement.
April 20, 2007 10:59:18 PM

don't be so cruel... Maybe it is real information from someone in the know? I suppose it sounds like it though.

BTW, I have been here for a while (years) viewing but never really got involved until recently. See when I joined? Viewing before that.
April 23, 2007 4:36:08 PM

Quote:
don't be so cruel... Maybe it is real information from someone in the know? I suppose it sounds like it though.

BTW, I have been here for a while (years) viewing but never really got involved until recently. See when I joined? Viewing before that.


Hey...I didn't mean to be so cruel. :)  Anyway, whenever you post info like this it is better, that is the ethical norm, to post where you get the info from. In any case, good find.

NP
April 23, 2007 4:57:50 PM

Quote:
don't be so cruel... Maybe it is real information from someone in the know? I suppose it sounds like it though.

BTW, I have been here for a while (years) viewing but never really got involved until recently. See when I joined? Viewing before that.


Hey...I didn't mean to be so cruel. :)  Anyway, whenever you post info like this it is better, that is the ethical norm, to post where you get the info from. In any case, good find.

NP

I don't think that this post was far off from how any engineer that understands IC design would describe the trade-offs between monolithic and multi-chip. It wouldn't require taking the information from any particular source. This is pretty much common knowledge for those in the industry that are able to apply their knowledge to this particular situation.
April 23, 2007 5:01:23 PM

Quote:
I don't think that this post was far off from how any engineer that understands IC design would describe the trade-offs between monolithic and multi-chip. It wouldn't require taking the information from any particular source. This is pretty much common knowledge for those in the industry that are able to apply their knowledge to this particular situation.


I dunno. The last time I saw a monolith I threw a bone at it.

:lol: 

I know that the MCP yield issue has been talked to death around here for months, but does anyone think that AMD may have found that the yields are just too uneconomical to produce especially given their current cashstarved status? I'm just grasping at straws as to why it's past 7 pm in Tunisia and there hasn't been a peep out yet....
April 23, 2007 5:05:38 PM

Quote:
I don't think that this post was far off from how any engineer that understands IC design would describe the trade-offs between monolithic and multi-chip. It wouldn't require taking the information from any particular source. This is pretty much common knowledge for those in the industry that are able to apply their knowledge to this particular situation.


I dunno. The last time I saw a monolith I threw a bone at it.

:lol: 

I know that the MCP yield issue has been talked to death around here for months, but does anyone think that AMD may have found that the yields are just too uneconomical to produce especially given their current cashstarved status? I'm just grasping at straws as to why it's past 7 pm in Tunisia and there hasn't been a peep out yet....

65nm paper launch....hmm we're struggling to produce good yields on our parts....i know lets make an even more complex product, thats bound to imporove our process tech.
April 23, 2007 5:12:16 PM

Quote:

65nm paper launch....hmm we're struggling to produce good yields on our parts....i know lets make an even more complex product, thats bound to imporove our process tech.


Could they be that stupid... er... we're talking about Hector, Henri and Co... :oops: 
April 23, 2007 5:36:58 PM

However

if their method is cheaper and works as well. Whats the problem?

in the end, it's all about performance and all about costs. is it worth it to go AMD's route for (hypothetical) 10 times the price for 1/2 the performance gain over Intel's route?

who knows.
April 23, 2007 5:53:31 PM

Quote:
I don't think that this post was far off from how any engineer that understands IC design would describe the trade-offs between monolithic and multi-chip. It wouldn't require taking the information from any particular source. This is pretty much common knowledge for those in the industry that are able to apply their knowledge to this particular situation.


I dunno. The last time I saw a monolith I threw a bone at it.

:lol: 

I know that the MCP yield issue has been talked to death around here for months, but does anyone think that AMD may have found that the yields are just too uneconomical to produce especially given their current cashstarved status? I'm just grasping at straws as to why it's past 7 pm in Tunisia and there hasn't been a peep out yet....

I would agree that it is very possible that AMD has miscalculated and is having problems with their very large, complex die. My recollection is that it doesn't necessarily have more transistors than some of the intel die, but intel's die are mostly memory, which is easy to yield. The huge amount of functional logic on Barcelona has to be difficult to fabricate.

With that said, I don't think that AMD should be having trouble producing a handfull of working chips for demonstration. There have to be other problems getting in the way. It might be that first silicon was garbage and they have been re-spinning the design, possibly multiple times. They may also be reconsidering the viability of the design due to yield problems and don't want to promise something they can't deliver.

To sum up my opinion, something else is going on. AMD shouldn't have problems producing test chips, even if yields are very low. They are not providing a live test either because the design had serious issues, or their results aren't as good as advertised. Neither of these scenarios is good for them.
April 23, 2007 5:57:58 PM

Anything designed a set way from the start is always better, then a modified version. This is similar to the Japanese Cars -vs- American Muscle Cars.

Japanese cars use high revving low displacement engines to create the greater efficiency than most standard American V6's. These engines were designed with efficiency in mind. American engines are low revving high torque engines.

AMD doesn't believe we need to have quad-cores yet, but Intel is pushing its way to 8-core CPUs so AMD is forced to follow, which is why AMD has announced they won't be designing a monolithic octo-core CPU, they will take the Intel approach and make a MCP from the latest quad-core architecture.

Its a mess, a better designed part always better than pieced together part.
April 23, 2007 6:23:49 PM

Quote:
Anything designed a set way from the start is always better, then a modified version.


Anything at all is better than nothing :D 

AMD is a car with no engine ;) 
April 23, 2007 6:42:49 PM

Quote:
To sum up my opinion, something else is going on. AMD shouldn't have problems producing test chips, even if yields are very low. They are not providing a live test either because the design had serious issues, or their results aren't as good as advertised. Neither of these scenarios is good for them.


I've been thinking the same thing for quite a long time, since K10 was first pushed back, and then again when R600 slid and slid and slid. I also believe that there is a possibility that there is some manufacturing disaster that AMD is trying very hard to keep from Wall St. If so, Gawd have mercy on their souls. Everyone but Henri. He deserves to burn! :twisted:
April 23, 2007 6:46:17 PM

Quote:
To sum up my opinion, something else is going on. AMD shouldn't have problems producing test chips, even if yields are very low. They are not providing a live test either because the design had serious issues, or their results aren't as good as advertised. Neither of these scenarios is good for them.


I've been thinking the same thing for quite a long time, since K10 was first pushed back, and then again when R600 slid and slid and slid. I also believe that there is a possibility that there is some manufacturing disaster that AMD is trying very hard to keep from Wall St. If so, Gawd have mercy on their souls. Everyone but Henri. He deserves to burn! :twisted:

A fabwide excursion in Fishkill dealing with front-end metal perhaps? :lol:  n00bs.
April 24, 2007 9:05:22 PM

Well it looks like someone looked at my profile and knew why I was speaking the way I was. And I was also kidding about being cruel, no harm. My intent was to illustrate the differences between the two items and why each party chose the paths they are on. When the initial announcement of quad-core was made by Intel AMD went "Ok, we will get our quad core ready as well." While AMD was rethinking the whole archetecture probably under the impression that intel was doing the same thing, Intel took the quick and easy road and pasted two dual cores together. I do think that if AMD does it right (and I think that is what is taking them so long) then the performance differences will be in their favor on a Mhz to Mhz basis. Direct communication between cpu's or components vs. using busses and controllers has always been the ideal condition. AMD had the upper hand for a while due to this regard and how their processors talked to the memory and various components. Intel is currently on the offensive and developing some new technology to keep AMD at bay, so the next two years will be interesting. BTW, Bearlake and beyond will not have any legacy components so be prepared to upgrade!!
April 24, 2007 9:42:22 PM

Quote:
The monolithic approach is a choice to be made when the customer requires extremely high performance and efficiency and can tolerate a higher cost for it. Intel's usage was the quick, less expensive, and easy way to get the product out first while keeping quad core performance. If they are designed for faster access to memory and communication between the cores, monolithic processors can offer higher performance within the same power envelope. Some estimate it would be approximately 10-15 percent higher.


This is an empty statement with no basis. The FSB is not a bottleneck between the two dies in a kentsfield, so the performance would be the same as if it were on the same die. Th espeed between dies is currently a non-issue.

There are some benefits to a monolithic core, such as being able to completely shut down up to 3 cores at idle. There are some other benefits that may yield some performance boosts, but I don't know where you're getting your 10%-15% figure from.

I believe Intel has it right this time: the costs of a monolithic core outweight the benefits. AMD prefers to stay 2-steps ahead of the competiton by releasing technologies before they are required, to get "one leg up" on the competition (e.g. HyperTransport and IMC). The monolithic core is just AMD doing what AMD does, and it doesn't provide any direct or real benefits over two dies at this time.
April 24, 2007 10:08:17 PM

I did not say bottleneck. I was referring to the faster of the two and no more. everyone said the HyperTransport made a difference, albeit small.

I agree, the performance differences will be neglegible at best, but those are some numbers I have seen about the theoreticle differences on paper so I do not want to count them out.

Either way, in the end the same old game will be played out move for move leap-froging. I do not think AMD will ever disappear, just get wounded in the battle as Intel did last year.
April 24, 2007 10:23:32 PM

i disagree. the FSB may have the bandwidth for the 2 die's however, latency also becomes an issue. especially if they do that with 8 cores. just my opinion.
April 24, 2007 10:28:17 PM

Quote:
IMO, intel's way aint true anything as if their way is right, then why not just call, a dual socket sytem a dual or qaud core. it is the same thing ain't it? by that logic, we have had dual, quad and octo cores for god knows how many years.

just because they have so many cores on one socket does not necessarily mean it can be described as a multi-core CPU.

yes, other disagree but that is my opinion.


I don't get your reasoning. Maybe if you referred to hyperthreading that would be the case. But my QX6700 is made up of 4 cores, hence the multi-core CPU. That's 4 physical cores, 4 cores. If you look up the definition of multi-core it refers to dual-core cpu's but its meaning can be used for more than 2 cores: Definitions of multi-core on the Web:

A dual-core CPU combines two independent processors and their respective caches and cache controllers onto a single silicon chip, or integrated circuit.

Your opinion it may be, but your opinion doesn't count when the facts state otherwise. There's a lot of other technical infor available to help explain what multi-core is. Just visit AMD or Intel's websites to see.

I'm not trying to be rude, but, honestly, when someone's opinion is flawed, you should tell them.
April 24, 2007 10:53:00 PM

Hence the original reason for this post. It is two different ways to achieve the same thing and one is merely easier to accomplish than the other. A dual SOCKET does not share the same key components on the single chip and uses the system buss for communication, all on individual and separate terms.

Multi core it is no matter how it is stacked, glued, paired, or otherwise. two dual core dies put on the same chip is still four physical cores. It is just one way to do it. And when AMD releases their quad core you better bet they will be spouting theirs is the only "True Quad Core" processor.
In the end it will depend how well these differences scale down to 54nm and 32nm. At that point the advantages and disadvantages will become apparent and by then everyone will be "True Octacore Processors".
April 25, 2007 2:26:37 PM

Quote:

Either way, in the end the same old game will be played out move for move leap-froging. I do not think AMD will ever disappear, just get wounded in the battle as Intel did last year.


Yeah; I certainly hope so. The strong competition is really pushing the technology forward, and driving the prices down. It is possoble for Intel to maintain the lead, but we as consumers would inevidebly lose.

Quote:
i disagree. the FSB may have the bandwidth for the 2 die's however, latency also becomes an issue. especially if they do that with 8 cores. just my opinion.



I remember reading something about how they tested the communication between the cores, and showed that the FSB latency/bandwidth had little to no impact on performance. I'll have to dig up the link....

I remember it came up with a lot of AMD fanboys when Conroe was released, because the PD seemed FSB bottlenecked.


greenmachineiijh


Welcome to the forums, and sorry to rip into you so early ;) 

Statements like thses make you an easy target:

Quote:
If they are designed for faster access to memory and communication between the cores, monolithic processors can offer higher performance within the same power envelope.


Quote:
Some estimate it would be approximately 10-15 percent higher.



The first is actually completely false from everything I've read. The second seems to be a number pulled from he sky. Be sure you can site your sources, or your posts here will be torn apart.
April 25, 2007 3:43:31 PM

I agree Whizzard9992, but I must state those are the numbers I have seen but I can not say where. Lets say it is from an inside source in the industry. It is all speculation anyway and the TRUE performance differences remain to be seen, if any...

BTW, Thanks for the warm and fuzzy welcome :lol:  . I have been here for a long time but have not posted anything until recently, hence the Newbie title. I will be showing my "face" around here a lot more.
April 13, 2009 9:24:43 PM

I purchased a Q6600. I looked at AMD and intel before buying it at the most recent core stopping and the lowest price since first production. $186.00. The whole thing is 6 of one and a half dozen of the other. or like the curved = signs in math where they refer to roughly equal; meaning little difference. "Greenmachine” explained it best. Intel took two Core 2's and 'SOMEHOW' put them together to make a quad core or 2(core2) = quad Core... Intel even named it Core 2 quad... They did the same thing in the Pentium D 940 which I also own. 2 P4's jammed together. That's why the chip ran hot and instead of bitching, I bought a larger heat sink and cooler and had NO problems. Communication is the key here. Here's a crazy analogy; If I have 2 Siamese twins, do I have 2 or four kids? Let’s complicate this even more; What if they were born at the same time? Are they quads or conjoined twins. The questions the Dr's ask are can they live if separated? Can they think for each other? I'll stop here because I think we all get Greenmachine's point now... The system bus approach to communication when using two CPU's, P4's or dual cores, link the processor together. efficiency is determined by the link and how you solve the different problems the Bus/link causes. Throughput varies with application. From my experience the P-D 940 is a better game chip than the Core2. Most gamers’ disagree because the chip can shut down when you're running a heavy game and it becomes hot. I guess the little nanopeople carrying the info between the two cores got tired and exhausted from carrying the gaming info. between the two P4 chips in such a heated environment...
April 13, 2009 9:28:29 PM

The little Naopeople should have take a bus so they wouldn't get tired...
April 13, 2009 9:41:00 PM

New feature request for the forums - automatically lock threads that are over a year old.

April 13, 2009 9:44:52 PM

I'm sure the nanopeople have gotten a little out-of-shape in the two years that this thread has been dormant.
a c 127 à CPUs
April 13, 2009 10:39:19 PM

Woot for necroing. I agree with spongebob.

Too bad the OPs prediction wasn't right.
a b à CPUs
April 14, 2009 1:28:28 AM

Ask Justin, he won't say "yes" to my complaint, but if we complain together...

!