AMD tri core

[BaronMatrix]

You do realize that it will also affect AMD's quad core prices as well, which will in turn, make their already non-existent profit margins get only smaller.
Let's see, Intel has a quad selling for under $300, so AMD's 3 core would have to be lower than that, let's say $190. Where would that price AMD's quads? $250? $299? Then AMD will have to not price them so low, as to not compete with their own dual core products.
Not much profit there.
Add to the mix is that Intel will introduce Penryn in November, which would likely lower quads again a bit, so more profit margin is gone from AMD. How much money has Intel loss during this price war? How much as AMD? Who's smacking who around again?

Yeah, so let's undercut our own quad core sales with our 3 cores, and make even less per CPU. That's brilliant.

Since Anand basically showed that K10 is on average ahead by 300MHz ( in his tests), then they can charge more for quad and a little less for tri. Another issue none of you seem to realize is that AMD AUTOMATICALLY saves money because most of their stock is 90nm desktops which are fabricated on 200mm wafers at 90nm.

Using 300mm wafers at 65nm gives them a little more than 2X the amount of chips per wafer (at the same die size).
If you take worse case scenario of 199mm for X2 and 283mm for Agena, that's about 30% larger with 120% + wafer area. If you take the beast case of Opteron at 223mm K10 is only 25% larger gain with 120% more wafer area.

By cutting back production at 90nm that should save even more money. Shipping Brisbane\Turion out to Chartered saves more.

 

[yomamafor1]

That's not the only thing from 1993 I don't exactly remember.

Well, why don't you come up with your own links to justify your claim? Asking others to google it is simply outright saying, "I don't know what I'm saying, but I'm sure this happened. Go look it up".

[quotmsg]
I said there were conflicting reports. XBit has a story, ExtremeTech has a story. Fudzilla has ten.[/quotemsg]
Again, why don't you put the links up?

And no I don't believe your 30% yields. Like I said before the data in the link was 1996 data. I would think that the process has improved by now. I scanned through it but didn't see an algorithm.

I said it numerous time already. Yield is a function of die size and defect density. The figure is just a graphic model of the yield function, with 1996 DRAM data points. They even provide the function at the bottom of the figure. Even the time has changed, the function is not going to change.

EDIT: For simplicity's sake, and for your own ignorance to see the data, I'll explain it step by step.

The yield function is

Y = [(1-e^(-AD))/(AD)]^2, where A is area of the die, in cm^2, and D is defect density, in cm^2

Now, according to AMD's slide:

It clearly stated that the defect density is 0.5cm^2. We also know that Barcelona has a 283mm^2 die size, which is 2.83cm^2.

Y = [(1-e^(-(2.83)(0.5))/((2.83)(0.5))]^2 = 0.286 ~ 0.29.

That means, with its monolithic approach, AMD can only yield them at 29%.

Now, BM, your turn.

 

[BaronMatrix]

Look at today's retail market - Intel Quad is $279 for Q6600 (2.4GHz). AMD's top dual 6400+X2 is $244, and the 6000+X2 is $170. So, for $30 more, you can get a Quad over the 6400+, and for about $100 you can get a Quad over the 6000+. You squeeze the X3 between both, and you are now eliminating the top dual core product from AMD, and possibly undercutting all the other dual cores (which could lead to price cuts, which won't help AMD). Even if the 3X is priced at $200, that's still only $80 more for a Quad.

AMD selling these things is fine, but I am just trying to see where they would be placed to be both competitive and not undercut themselves. And if rumors of bad yields is true, they cannot price these too low, or they will cost more to make than to sell.


But remember that Q6600 is only $50 more than the E6600, so for every C2Q they sell, they lose about 70% of the price of an E6600. Check out Newegg.

With these AMD will have EVERY market segment covered. A64 Sempron - entry level single core; A64 X2 entry level dual; Kuma - mainstream to high end dual; X3 - higher end gaming, image stuff; X4 - extreme high end gaming.

K10 has enough oomph to push most of AMDs prices up. The 8350 is a little more than the 8224SE so the 8360 and up will go for up to $2500. AMD did say in the YouTube video that K10 would be released at the original Opteron dual core prices.

 

[BaronMatrix]

It clearly stated that the defect density is 0.5cm^2. We also know that Barcelona has a 283mm^2 die size, which is 2.83cm^2.

Y = [(1-e^(-(2.83)(0.5))/((2.83)(0.5))]^2 = 0.286 ~ 0.29.

That means, with its monolithic approach, AMD can only yield them at 29%.

Now, BM, your turn.

283mm is not 2.83cm unless math changed. It's 28.3. The slide says BELOW .5 not at .5.

Here is a link with a different yield equation.

http://www.csee.umbc.edu/~cpatel2/links/640/itrs_presentations/sai_11_23.pdf

Do you really think AMD is yielding at 30%? This is their second 65nm design. I don't know but somehow I don't think it's that low.

 

[warezme]

AMD is smart, 3-cores is an opportunity.

1.) sell those defective quads
2.) get publicity by from as of now unheard core "design"

The biggest reason is for us end users is what if some smart hacking type person figures out how to "enable" that fourth core!!!

Turn it on make it run, suddenly there is a cheap as dirt Quad core everyone wants. After all, what is overclocking?? If not taking a slower rated chip that failed at higher speeds at default voltage.

Thats why most overclockers can just juice those suckers up and viola, a very fast stable processor. There's nothing to say a little bios patch and some extra juice won't bring that 4th core to life!!

Huh? are you guys with me???

 

[yomamafor1]

283mm is not 2.83cm unless math changed. It's 28.3.

In case you're so eager to prove this wrong, its mm^2 to cm^2. So 283mm^2 is 2.83cm^2

The slide says BELOW .5 not at .5.

The slides says BELOW 0.5 RANGE. If AMD's defect density is 0.4cm^2, do you think they will write "below 0.5cm^2 range"?

Here is a link with a different yield equation.

http://www.csee.umbc.edu/~cpatel2/links/640/itrs_presentations/sai_11_23.pdf

Yet you failed to come up with a different yield. Want to provide the function as well as the calculations?

Do you really think AMD is yielding at 30%? This is their second 65nm design. I don't know but somehow I don't think it's that low.

From the technical data, it says 30%. Phenom x3 exists for a reason, and definitely not because of "market demand". This was never part of their roadmap, so AMD didn't design a native tri-core.

Unless you can give me a different yield, with calculations and explanations, I'll believe 30%.

 

[chunkymonster]

In case you're so eager to prove this wrong, its mm^2 to cm^2. So 283mm^2 is 2.83cm^2

Hmmm, not to butt in here as it is really entertaining, I gotta agree with BM on that 283mm is 28.3cm. It's a simple conversion.

It clearly stated that the defect density is 0.5cm^2. We also know that Barcelona has a 283mm^2 die size, which is 2.83cm^2.



Y = [(1-e^(-(2.83)(0.5))/((2.83)(0.5))]^2 = 0.286 ~ 0.29.



That means, with its monolithic approach, AMD can only yield them at 29%.



Now, BM, your turn.

And, since converting 283mm is actually 28.3cm, if I did the math correct using the above equation (and I could be wrong as it's been a while for me), I come up with a minimum yield of about 86%.

Can someone else do the math? Someone else aside from yomamafor1 and BM, perhaps?

 

[spongebob]

Hmmm, not to butt in here as it is really entertaining, I gotta agree with BM on that 283mm is 28.3cm. It's a simple conversion.

No, that's a linear conversion. Since we're talking square surface area, we need to divide not by 10, but by the square of 10.

Think about it, 10cm * 10cm is 100cm squared. 100mm * 100mm isn't 1000mm squared, it's 10000mm squared.

yomamafor1's math is correct.

 

[reconviperone1]

I prefer some Oakland or Macomb names:

The Warren
The Troy
The Royal Oak
The Clawson
The Southfield
The Sterling Heights

defineately not the warren, the royal oak, or the troy we can agree on. Maybe the madison heights even(i,m moving to concord towers in the microcenter parking lot in madison heights, which may be a bad thing lol)

 

[reconviperone1]

Yet another instance where the technically minded say "bah, asymmetric=bad" and the consumer says "3 > 2" but in this case the consumer is correct. Though, honestly, I would be surprised if the yield is bad enough that they would have a lot of them to sell to OEMs in sufficient quantity for them to justify sticking them into their product lines. I expect them more in the custom builder channel. If Intel were building monolithic quad cores, they would be doing the exact same thing.

Though I'm intrigued by the prospect of a 6 core system, oddly enough. If you could find a 2P board that took AM2/2+ chips that would be sweet. Should be a hell of a lot cheaper than the socket 1207 stuff.

I'm kind of intriqued by the tri core and the fasn8(i would never own one of course), can't wait to see it come out.

 

[cyberjock]

I posted a comment in slashdot.org about this topic too. From what I'm reading(contrary to pictures on sites and the actual article) the triple core is actually designed that way. It's not a 4 core hand-me-down. It's actually designed to be 3 cores from the getgo. If they are doing 4 core hand me downs, I'm not sure it's really gonna help them much for the reason I list below. If they designed it to be 3 core, they are morons. They wasted R&D money on a processor that has little to no consumer demand. No consumer demand = product that cost the company money and they won't even break even marketing it.

So why is 3 core CPUs bad? People fall into 2 categories for the most part:

1. Joe Schmuck - Surfs the net, uses MS Office, maybe some Solitaire or a game or 2 but nothing very CPU/3D intensive. Dual core is more than enough for this task.

2. OMFGGAMERZ - Games hardcore. They want the fastest they can get for the buck(if they even care about the cost of it). They want the quad core whether it'll help them game or not because it's fast.

Where's the comsumer that's gonna want both worlds? I don't see people wanting to buy a 3 core CPU vice a 2 core CPU to surf the net, but don't want to 'splurge' for the 4 core.

 

[jaywalker256]

I posted a comment in slashdot.org about this topic too. From what I'm reading(contrary to pictures on sites and the actual article) the triple core is actually designed that way. It's not a 4 core hand-me-down. It's actually designed to be 3 cores from the getgo. If they are doing 4 core hand me downs, I'm not sure it's really gonna help them much for the reason I list below. If they designed it to be 3 core, they are morons. They wasted R&D money on a processor that has little to no consumer demand. No consumer demand = product that cost the company money and they won't even break even marketing it.

So why is 3 core CPUs bad? People fall into 2 categories for the most part:

1. Joe Schmuck - Surfs the net, uses MS Office, maybe some Solitaire or a game or 2 but nothing very CPU/3D intensive. Dual core is more than enough for this task.

2. OMFGGAMERZ - Games hardcore. They want the fastest they can get for the buck(if they even care about the cost of it). They want the quad core whether it'll help them game or not because it's fast.

Where's the comsumer that's gonna want both worlds? I don't see people wanting to buy a 3 core CPU vice a 2 core CPU to surf the net, but don't want to 'splurge' for the 4 core.

im pretty sure there are more groups in between the average joe, and super enthusiasts with money to blow.
for example, people like me. a learning software developer who often has 20+ applications open (multiple copies of vb, winamp, a few web browsers, talking on skype, chatting on xfire+aim, downloading movies and WoW running in the backround, for that break every few minutes), and in doing all this at once, a single core can get kind of sluggish. so depending on how they are priced, a tri-core may just be my next cpu.

 

[TechnologyCoordinator]

defineately not the warren, the royal oak, or the troy we can agree on. Maybe the madison heights even(i,m moving to concord towers in the microcenter parking lot in madison heights, which may be a bad thing lol)

I've been watching them repaint those on my way home, they keep getting further and further. I grew up in Clawson and now live in Warren. I worked for Madison Schools for 17 months, where I had the title that I currently post under.

 

[Badarthur]

One of the most important word's to CPU mfg is " Yield " and if the yield is low then the number of failed procesor's is high meaming that the price on non failure cpu's is going to be higher. More failure's equle's less yield equle's higher prices for you and me. Just like most of the GS type Video card's a failed quad or Tri core might just be a heck of a deal for us. And there is allway's the possibility of unlocking the failed 4th core.
I didnt hear anything all that good about AMD's quad processor's so now we are hearing about the 3core because of low yield on the quad.
I have been opresonally waiting on a good Quad from AMD for a wile now and yester day I quit waiting and bought a Intell Quad MB and ram to match. I actually bought a P5K like Tom's used in there mid ranged build I saw today. And am loking forward to haw well this thing worki's aginst my 170 Opty clocked up to 2,8. this si the first intell build in my life and ive been building pc's since 1999 we will have to wait and see how well it work's. BA

 

[dragonsprayer]

This is simple - since amd is only making mistakes this must be a mistake! if it was not then it would be a brillant move - it can not be a brillant move since amd is only making mistakes.

simple - another bad marketing plan!


if amd was smart they sell these as a dual core with a free core!!!

now that would sel! New marketing plan: buy 2 cores an get one free!

I can see it now "amd offers free core:

if this is not another dumb move then it could be the start of a whole new line of good moves! since amd so far for the past 6-9 months just makes claims that turn out not to be true!

 

[sedaine]

You do realize that it will also affect AMD's quad core prices as well, which will in turn, make their already non-existent profit margins get only smaller.
Let's see, Intel has a quad selling for under $300, so AMD's 3 core would have to be lower than that, let's say $190. Where would that price AMD's quads? $250? $299? Then AMD will have to not price them so low, as to not compete with their own dual core products.
Not much profit there.
Add to the mix is that Intel will introduce Penryn in November, which would likely lower quads again a bit, so more profit margin is gone from AMD. How much money has Intel loss during this price war? How much as AMD? Who's smacking who around again?

Yeah, so let's undercut our own quad core sales with our 3 cores, and make even less per CPU. That's brilliant.

I think there are a number of factors to consider.

Firstly, it seems that these X3's will only end up on the desktop end i.e. Phenom X3.

Secondly, rumour is that the X3's clock better than the X4's.

Also - a point others have made is that an X3 puts each core a max of one hop away from the next, where as with an X4 the max is 2 hops. So better data sharing between cores - faster.

Tri-Cores for the FX scene?! Giving you 6 cores on one of those FX MOB.


Ok, with all that to consider - we now need to look at pricing. If they clock all the X3 from 2.5Ghz+ then they have a home run. If it comes in at 3.0Ghz I'll buy one. Any of this 2.0Ghz nonsense then Hector had better be ready to quit - since effectively AMD will be done for!

So they supposedly have 30% yeild of Quads. So that would put 40-50% of the failed CPU's being X3. So technically speaking they are around 70-80% in terms of what they will sell. I just assumed most failures would result in Tri-Cores and that the yeild would be the highest. Of course some X2 will also come out of this - perhaps 20-30% at most.

I'd bet AMD considered disabling the 3rd core and just calling them X2 - I also suspect this is the reason those two AMD execs quit, how were they going to stay afloat with all these defects that were going to become X2's.

Yes - it is an expensive way to get a X3 because 1/4 of the silicon to make your processor is wasted. But it's better than disabling 2 cores to get an X2 when you have one perfectly good one.

So let me put this to everyone. If it turns out that Phenom X2 performs better than Core 2 Duo, would you spring the extra $40-50 for an even better performing X3. I suspect AMD will price their X2 a little cheaper than Core 2 Duo's at same performance level - then price X3 at same level as Core 2 Duo so as to make it a no brainer. All indications suggest that Phenom X2 will perform about the same or marginally bettey than current Core 2 Duo.

So this will be interesting.

 

[sedaine]

In case you're so eager to prove this wrong, its mm^2 to cm^2. So 283mm^2 is 2.83cm^2

The slide says BELOW .5 not at .5.

The slides says BELOW 0.5 RANGE. If AMD's defect density is 0.4cm^2, do you think they will write "below 0.5cm^2 range"?

Here is a link with a different yield equation.

http://www.csee.umbc.edu/~cpatel2/links/640/itrs_presentations/sai_11_23.pdf

Yet you failed to come up with a different yield. Want to provide the function as well as the calculations?


From the technical data, it says 30%. Phenom x3 exists for a reason, and definitely not because of "market demand". This was never part of their roadmap, so AMD didn't design a native tri-core.

Unless you can give me a different yield, with calculations and explanations, I'll believe 30%.

I hate to butt in but I feel I must. I'll use approcimate values to make it easy.

20 mm = 2.0 cm
15 mm = 1.5 cm

I think we can all agree on those figures.

20mm x 15mm = 300mm^2 this is the approx. size of X4.

Which also is 2cm X 1.5cm = 3cm^2


Not sure what anyone would do with a CPU bigger than my hand (28.3 cm^2). that thing would measure 7cm X 4 cm. One hell of a big CPU!!! 😀

 

[Harrisson]

IMO its both ways doable, i.e. converting defective X4 to X3, and even making X3 as it is when process matures and if there is loads of demand.

Some here mentioned there is too little gap between 2x and 4x cpu's so there isnt price range to fit 3x into. Not really. For example Intel sells Q6600 for E6850 price, thats two speed bumps or ~140$ in my country. Plenty of space for X3 parts.

One may argue, who will buy X3 if X4 isnt that much more expensive? There is another question IMO, who will buy X2 or C2D if X3 isnt much more expensive either? ;-) That said, there is always market for decent products with a smart pricing, and we as customers only win when we have more choices.

 

[dasickninja_lives]

There could be a market for this, but not if Intel decides to make quads even cheaper.

And by the way... yo.

 

[jwolf24601]

Not sure what anyone would do with a CPU bigger than my hand (28.3 cm^2). that thing would measure 7cm X 4 cm. One hell of a big CPU!!! 😀

Ahh reminds me of my old 333mhz Pentium 2 Slot connecting chip. What a silly thing that was...

 

[yomamafor1]

There could be a market for this, but not if Intel decides to make quads even cheaper.

And by the way... yo.

you alive?

 

[dasickninja_lives]

you alive?

Well mostly. Just wanted to say hi.

 

[bfellow]

It clearly stated that the defect density is 0.5cm^2. We also know that Barcelona has a 283mm^2 die size, which is 2.83cm^2.

Y = [(1-e^(-(2.83)(0.5))/((2.83)(0.5))]^2 = 0.286 ~ 0.29.

That means, with its monolithic approach, AMD can only yield them at 29%.

Now, BM, your turn.

283mm is not 2.83cm unless math changed. It's 28.3. The slide says BELOW .5 not at .5.

Here is a link with a different yield equation.

http://www.csee.umbc.edu/~cpatel2/links/640/itrs_presentations/sai_11_23.pdf

OMG I can't help but laugh at simple square area mathematics!

Do you really think AMD is yielding at 30%? This is their second 65nm design. I don't know but somehow I don't think it's that low.

 

[yomamafor1]

BM, in regards to your newer Yield function, contains a new variable, a, as defect clustering parameter.

Assuming that there is no clustering, since I don't have AMD's clustering parameter value, the function I posted earlier is still valid.

So 30% yield for AMD's Barcelona, at least for now.

 

[reconviperone1]

There could be a market for this, but not if Intel decides to make quads even cheaper.

And by the way... yo.

Good to see you in here ninja.