Some Bulldozer and Bobcat articles have sprung up

[Daedalus685]

I think they were not talking about a die, but about a module.
And a module doesn't include the uncore functionality, such as the L3 cache, which makes a big difference in terms of any die-area related issues.
There will also be multiple modules on a die, to further affect any die-area talks.
For a module, adding a second integer core, the 50% figure seems correct, and also it is meaningful. You show that the Bulldozer architecture significantly reduces die area per core, a very important metric in the current parallel climate.
For a full die, removing an integer core of every module, 5% may be correct, but what does that prove, really? It seems to be a completely random metric.

The latest info in the ANAND article and the press slides state 12% increased module size though... not 50%.

Certainly the 12% they state might simply be the "if I take it away I take off 12%" ignoring the larger uncore to support it... but that is all they are saying.. The new 'core' is substantially smaller than an old core, and adding a a dual core module is only 12% or so (certainly could be an inaccurate value...) larger than a single core would be assuming the same simple core design.

 

[GaiaHunter]

I think they were not talking about a die, but about a module.
And a module doesn't include the uncore functionality, such as the L3 cache, which makes a big difference in terms of any die-area related issues.
There will also be multiple modules on a die, to further affect any die-area talks.
For a module, adding a second integer core, the 50% figure seems correct, and also it is meaningful. You show that the Bulldozer architecture significantly reduces die area per core, a very important metric in the current parallel climate.
For a full die, removing an integer core of every module, 5% may be correct, but what does that prove, really? It seems to be a completely random metric.

The slide is clear - 12% INCREASE of MODULE SIZE and 5% INCREASE OF DIE AREA (on a 4 Module 8 integer cores).

If it is the truth or simply Marketing BS I don't know. But this a topic discussing BD and Bobcat architecture based on AMD word.

It isn't discussing if AMD word has value or not.

So lets assume what they say is what they want to say.

The value of this is simple.

Lets imagine AMD 1 BD integer core has the same performance hs a single phenom II core, and so similar size (AMD word is that single thread performance is increased).

So if the process was the same (and it is going to be smaller) a BD CPU with 2 modules and 4 cores would be a bit over half the size of a Phenom II quad. So, instead of selling a quad-core for $160, AMD could sell it for a bit over $80 and have the same margin.

Of course we don't know the real performance of it. If it is better AMD could sell it for $160 and double their margins. If it is worse, well, they can sell it really cheap - it is smaller after all. If it is really good they might be able to sell it for a higher price.

 

[AtenRa]

For a module, adding a second integer core, the 50% figure seems correct

I’m GREEK and English is not my mother language, but from the above Slide it states very clear that ONE Int Core (Integer Execution Unit + Scheduler) only takes 12% of the total MODULE. Why you people keep saying you need 50% for what AMD says 12% I really don’t know.

Here is the math. Start with a full die. Remove 1 integer core from each module and 1 integer schedule (everything else stays where it is). Measure the die size. Your new number is 95% of the total die.

John I believe that this is wrong, because if one Bulldozer INT Core takes 5% of the total CPU die then on an 4 Module Bulldozer with 8 INT Cores will take 5% x 8 = 40% witch I think that’s (approximately) the area the whole 4 Modules will occupy (out of the whole CPU die).

I believe the 5% is all the INT Cores in a 2 or 4 Module Bulldozer CPU die. See my post above.

 

[Scali]

Basically we have AMD saying three different things:
- JFAMD
- AMD's corrections to Anand
- AMD's slides

If AMD didn't suck this much in communication, we wouldn't be having this useless bickering.
As jvroig and I have said many times: it's useless anyway, let's concentrate on other topics.

 

[AtenRa]

Pop quiz #1: An "int core" is the whole she-bang that you need to add to an existing single core for it to be a dual core, yes or no?

In a Phenom II design (Deneb) you need the front end, execution unit, L2 etc. In Bulldozer you only need the Integer Core (12% more in the Module).

Pop quiz #2: AMD already has a die with 4 cores (normal ones, like Deneb, Propus, etc), if they were to add 4 more cores (making the chip a Bulldozer core), how much more die area would be needed, is it 5% or close?

First mistake, If you add 4 Deneb cores in a Phenom II Quad CPU you don’t get an 8 Core Bulldozer CPU but an 8 Core Deneb CPU. (8 Integer Execution Units + 8 FP Execution Units)

If you add 4 Deneb cores to a Quad Core Phenom II then you add 50% more die because you take 4 times the Front End and the L2 Cache. In Bulldozer you don’t add the front end, nor the L2 thought you don’t add 50%. Go it ???



Hope I was able to help

 

[JFAMD]

Basically we have AMD saying three different things:
- JFAMD
- AMD's corrections to Anand
- AMD's slides

If AMD didn't suck this much in communication, we wouldn't be having this useless bickering.
As jvroig and I have said many times: it's useless anyway, let's concentrate on other topics.

Actually, AMD has been saying exactly the same thing since November. If you look at the corrections they have all been on Anand's part.

As the guy who got the original numbers from engineering, then went back to engineering to have them verified and rechecked, I can assure you that they are correct.

Otherwise we would not have a senior engineer presenting them at hot chips if they were wrong.

AMD only has 1 story, the one we have had since November when we first made the statement. All of the other statements that are wrong have come from someone else, not AMD.

 

[khon]

This whole things would be so much easier if AMD would just release some information on the die size. If they're going to brag about the reduction in size/area, you would think they could offer some sort of proof to cooperate that claim.

Something like: With this new techonology we can fit 8 cores in under X mm^2. Or maybe: Thanks to the new design we've achieved an X % reduction in core compared to Phenom II.
It's taped out anyway, so it's not like they don't have those numbers.

 

[AtenRa]

Lets imagine AMD 1 BD integer core has the same performance hs a single phenom II core, and so similar size (AMD word is that single thread performance is increased).

So if the process was the same (and it is going to be smaller) a BD CPU with 2 modules and 4 cores would be a bit over half the size of a Phenom II quad. So, instead of selling a quad-core for $160, AMD could sell it for a bit over $80 and have the same margin.

Of course we don't know the real performance of it. If it is better AMD could sell it for $160 and double their margins. If it is worse, well, they can sell it really cheap - it is smaller after all. If it is really good they might be able to sell it for a higher price.

I agree

It will be like a 2 Module Bulldozer (4 Cores) will have the same performance as a Quad core Phenom but with a smaller die and it will be able to perform a lot better against a Dual Core SB in multithreaded scenarios (2 Cores + HT vs 4 Int Cores) with approximately the same die size.

 

[Scali]

Actually, AMD has been saying exactly the same thing since November. If you look at the corrections they have all been on Anand's part.

The problem is apparently in HOW you say it. If you say the same, but various sites keep thinking something else, apparently you aren't communicating very well.
I've never seen any such thing with Intel. Apparently their PR can get the message through the first time.
Something to think about.

 

[GaiaHunter]

John I believe that this is wrong, because if one Bulldozer INT Core takes 5% of the total CPU die then on an 4 Module Bulldozer with 8 INT Cores will take 5% x 8 = 40% witch I think that’s (approximately) the area the whole 4 Modules will occupy (out of the whole CPU die).

I believe the 5% is all the INT Cores in a 2 or 4 Module Bulldozer CPU die. See my post above.

Probably 5% of the die is all the extra int cores (that is what JFAMD is saying).

4 modules = 40% of die size.

1 module = 10% of die size.

1 int core 12% of a module - .12 x .1 = 0.012 -> 1.2% of total die size.

1.2% X 4 extra int core = 4.8%.

That is very close to the 5%.

So 1 Int core is 12% of the module size.

The 4 extra int cores are 5% of the die.

 

[AtenRa]

Here is the math. Start with a full die. Remove 1 integer core from each module and 1 integer schedule (everything else stays where it is). Measure the die size. Your new number is 95% of the total die.

You are right

Got it wrong sorry

 

[Idontcare]

Actually, AMD has been saying exactly the same thing since November. If you look at the corrections they have all been on Anand's part.

As the guy who got the original numbers from engineering, then went back to engineering to have them verified and rechecked, I can assure you that they are correct.

Otherwise we would not have a senior engineer presenting them at hot chips if they were wrong.

AMD only has 1 story, the one we have had since November when we first made the statement. All of the other statements that are wrong have come from someone else, not AMD.

Personally I really appreciated that you came here, all the way back then, and cleared up the confusion because we all had wildly differing interpretations of the meaning of the terms that were being used.

You didn't have to do that but it really helped defray a bulk of the angst that was developing among posters in this forum at the time.

Likewise your persistence here to continue to do what you can to ensure the numbers and meaning of the numbers aren't misinterpreted. For those who actually want to know more we are learning more.

There will always be those handful of individuals that could care less about the value of the info contained in the message and instead just want to make hay over the messenger.

They too, I'm quite sure, feel that they are delivering value with their message and I have no doubt you too see that value in their feedback.

 

[Kuzi]

Personally I really appreciated that you came here, all the way back then, and cleared up the confusion because we all had wildly differing interpretations of the meaning of the terms that were being used.

You didn't have to do that but it really helped defray a bulk of the angst that was developing among posters in this forum at the time.

Likewise your persistence here to continue to do what you can to ensure the numbers and meaning of the numbers aren't misinterpreted. For those who actually want to know more we are learning more.

There will always be those handful of individuals that could care less about the value of the info contained in the message and instead just want to make hay over the messenger.

They too, I'm quite sure, feel that they are delivering value with their message and I have no doubt you too see that value in their feedback.

+1

 

[Phynaz]

Personally I really appreciated that you came here, all the way back then, and cleared up the confusion because we all had wildly differing interpretations of the meaning of the terms that were being used.

You didn't have to do that but it really helped defray a bulk of the angst that was developing among posters in this forum at the time.

Likewise your persistence here to continue to do what you can to ensure the numbers and meaning of the numbers aren't misinterpreted. For those who actually want to know more we are learning more.

There will always be those handful of individuals that could care less about the value of the info contained in the message and instead just want to make hay over the messenger.

They too, I'm quite sure, feel that they are delivering value with their message and I have no doubt you too see that value in their feedback.

Here goes IDC being all logical again.

 

[Vesku]

Scali, reading your posts in this thread is giving me a headache. JFAMD lays AMDs claims out very well and showed you their slide. Even after that you insist that the integer core must be close to 50% of a module if not the whole chip. Slide clearly outlines a much smaller part of the module. I think it is more you are confusing yourself rather than poor communication from AMD on their claims. The small percentage of die occupied by their integer units just illustrates how much of the x86 CPU die has become dominated by cache and multicore interconnects, something else that has been repeatedly pointed out in this discussion.

Also, IDC summed it up, thank you for the information JFAMD.

 

[Scali]

Scali, reading your posts in this thread is giving me a headache. JFAMD lays AMDs claims out very well and showed you their slide. Even after that you insist that the integer core must be close to 50% of a module if not the whole chip. Slide clearly outlines a much smaller part of the module. I think it is more you are confusing yourself rather than poor communication from AMD on their claims.

It's not me who was confused, it was Anand and various other sites.
I was just pointing out that Anand was corrected by AMD, with a figure of 50% rather than 5%.
JFAMD still hasn't explained where the 50% came from.

And I stick to my point (which jvroig also seems to share), that although perhaps technically there is some merit to the 5% figure (wherever it came from, JFAMD's explanation could well be after-the-fact, damage control), it doesn't make a lot of sense.
Why do you want to argue from a NEW core, and then pose some hypothetical situation where you REMOVE some cores (but ignore the fact that you had to insert a lot of extra logic in order to make these cores work in a shared fashion)... It doesn't make sense.

What you want to know is this:
- Phenom II required X amount of extra transistors for every 2 cores added.
- Bulldozer requires Y amount of extra transistors for every 2 cores added.

Note that I'm not talking about die area, especially not in percentages, because that's rather meaningless when the die area is not the same in the first place (and I don't think the actual die sizes of Bulldozer chips have been disclosed yet?)

So really, everyone should be on my side, because we want useful info, don't we?

 

[Tsavo]

^You are completely dense.

 

[GlacierFreeze]

^You helped ruin the thread.

Fixed.

 

[heyheybooboo]

Personally I really appreciated that you came here, all the way back then, and cleared up the confusion because we all had wildly differing interpretations of the meaning of the terms that were being used.

You didn't have to do that but it really helped defray a bulk of the angst that was developing among posters in this forum at the time.

Likewise your persistence here to continue to do what you can to ensure the numbers and meaning of the numbers aren't misinterpreted. For those who actually want to know more we are learning more.

There will always be those handful of individuals that could care less about the value of the info contained in the message and instead just want to make hay over the messenger.

They too, I'm quite sure, feel that they are delivering value with their message and I have no doubt you too see that value in their feedback.

:thumbsup:


Thanks for your input, patience and professionalism, John.

(I would have gone 'Jack Bauer' on a few folks by now () )





--

 

[JFAMD]

Personally I really appreciated that you came here, all the way back then, and cleared up the confusion because we all had wildly differing interpretations of the meaning of the terms that were being used.

You didn't have to do that but it really helped defray a bulk of the angst that was developing among posters in this forum at the time.

Likewise your persistence here to continue to do what you can to ensure the numbers and meaning of the numbers aren't misinterpreted. For those who actually want to know more we are learning more.

There will always be those handful of individuals that could care less about the value of the info contained in the message and instead just want to make hay over the messenger.

They too, I'm quite sure, feel that they are delivering value with their message and I have no doubt you too see that value in their feedback.

Thanks!

All of you guys are potential customers, I just want to make sure you all have the right information before you buy.

Maybe you buy my products, maybe you don't. As long as you are informed, you can make a better decision.

 

[JFAMD]

:thumbsup:


Thanks for your input, patience and professionalism, John.

(I would have gone 'Jack Bauer' on a few folks by now () )

--

That is the biggest pain of all of this. Because I actually tell everyone who I am, I don't get to tell people to ^*^*&((*&%#

Being anonymous has its advantages, but creditibility isn't one of them.

 

[Vesku]

One thing it does seem I'll have to be wary of is "core count". I seem to recall a "% more cores for % more performance." Can you refresh my memory on that John?

 

[cbn]

Lets imagine AMD 1 BD integer core has the same performance hs a single phenom II core, and so similar size (AMD word is that single thread performance is increased).

So if the process was the same (and it is going to be smaller) a BD CPU with 2 modules and 4 cores would be a bit over half the size of a Phenom II quad. So, instead of selling a quad-core for $160, AMD could sell it for a bit over $80 and have the same margin.

Of course we don't know the real performance of it. If it is better AMD could sell it for $160 and double their margins. If it is worse, well, they can sell it really cheap - it is smaller after all. If it is really good they might be able to sell it for a higher price.

Good point.

I am also wondering *how much* these changes will help AMD's Bulldozer laptop market (from a power efficiency standpoint).

Intel core with more execution units and SMT vs Bulldozer in an efficiency sensitivity mobile environment?

 

[busydude]

One thing it does seem I'll have to be wary of is "core count". I seem to recall a "% more cores for % more performance." Can you refresh my memory on that John?

Its 50% increase in performance with 33% more cores.

 

[Vesku]

Found the quote I was looking for, 33% more cores with 50% more throughput of the previous generation. So a 16 core bulldozer would be a direct replacement of the current 12 core magny cours with up to 50% more performance.