AMD prepares three-core processors

[Cooper]

Compared with who? Who makes such complex quads?
Process is tweaked for quads. AMD has enough single and duals already.

[Starscream]

Those stats dont look right. Specially with the fully defect compared with the others. And there seems to be artificially high number of tricores compared to fully defect and the others.

if your reply is to the planet3dnow.de link.
Ive read the thread and the info the member Dresdenboy posts are pulled out if his ass more or less.
besides that his post is over a year old.

[Shintai]

Compared with who? Who makes such complex quads?
Process is tweaked for quads. AMD has enough single and duals already.

Compare the diesize of places that would kill any option for a CPU with the cache and cores. And even than cache would in most cases not end as a defect.

It just doesnt add up logically. It would _maybe_ be true if any cache defects killed it completely. But it doesnt.

[hollo]

27,3% defekt (55/Wafer)
0,4% Single Core (fast 1/Wafer)
4,6% Dual Core (9/Wafer)
23,2% Triple Core (47/Wafer)
44,5% Quad Core (90/Wafer)

i don't know much about CPU yields, but that looks kinda bad. does anyone know what the yields are for intel's c2d dies for conroe and kentsfield, for comparison?

a single core with 2mb l3 would be a pretty cool freakshow of a CPU for browsing or HTP
but i doubt they'll sell any.

[Starscream]

i don't know much about CPU yields, but that looks kinda bad. does anyone know what the yields are for intel's c2d dies for conroe and kentsfield, for comparison?

a single core with 2mb l3 would be a pretty cool freakshow of a CPU for browsing or HTP
but i doubt they'll sell any.

dont pay any attention to those stats.
they were calculated based on alot of rumored data. and besides that they were calculated in 2006.

[terrace215]

AMD to package up defective quad core Phenoms as "triple-core" Phenom X3 for Q108.

Yields of fully-functional, all-cores-binning-high parts must be really bad, despite AMD's protestations to the contrary.

Leave it to AMD to try to market a failure as a feature.

http://biz.yahoo.com/bw/070917/20070917006584.html?.v=1

http://www.thestreet.com/_yahoo/news...&cm_ite=NA

[terrace215]

dont pay any attention to those stats.
they were calculated based on alot of rumored data. and besides that they were calculated in 2006.

Those stats are entirely reasonable. Grab a yield calculator, and plug in AMD's announced (at July Tech Day) defect density of <= 0.5 / cm^2, Barc die size of 283 mm^2, and see how bad it is yourself. About ~60 functional QC Barcs can be expected per 300mm die, across all bins. They have to do something with the remaining partially-broken ones. Of course, they should've just sold them as DC parts, but to try to get a little extra $$$ they'll sell some of them as Triple-core, and kill their own QC pricing in the process.

[adamsleath]

http://www.betanews.com/article/AMD_...nom/1190069821

and kill their own QC pricing in the process.

prices to be adjusted as process yields change as with all chip manufacture
if it is a way of selling lower bins, then y not?

[terrace215]

http://www.betanews.com/article/AMD_...nom/1190069821

if it is a way of selling lower bins, then y not?

It isn't a way of selling lower BINS, it is a way of selling partially defective K10s. But they should've elected to sell potential tri's as duals, IMO.

There isn't enough room for a tri-core segment in between dual and quad. When AMD launches them, they're going to find it very difficult to charge much more for an equivalently clocked QC.

The only way this makes sense is if yields are so horrendous that K10 wafers yield (or bin high enough) mostly dual and tri-functional dies, with hardly any QC parts. In that case, who cares about QCs, if you don't have many...

Intel, in response, should simply price QC Penryns at the same price as 3-thread-performance-equivalent AMD Phenom X3 parts. Same price, an extra core, and much better overclocking. I don't think they should bother releasing their own tri-core parts, though it should be trivial to fuse off one core on a Harpertown if they wanted to. Not sure where these claims of "it will take them a long time to do it" come from.

The other issue with these parts is that they are quite clearly "partially defective". (If they sold them as Phenom X2s, which they are also making a 2-core die for, there'd be no way to tell short of looking up the OPN, or checking under the heatsink, and most consumers wouldn't do that) Now, in reality, who cares if one core is bad if you aren't using it, but that's never a good association for a product.

[Face]

Originally Posted by Face
FudZilla

Well.. that should be in about........................ now.

Just few more hours to see if the rumor is true or just the usual *.

There is gonna be Press Release regarding triple-core CPUs in Q1 2008.

FUD as usual

Hey, He actually got it right this time

AMD's first triple-core "Toliman" part will be released in the first half of 2008, along with the dual-core "Kuma" chip.
a second "Heka" next gen triple-core chip will follow in 2009.

Source @ ExtremeTech

Interview: AMD's Simon Solotko and the Third Core

Official Press Release

[Starscream]

Those stats are entirely reasonable. Grab a yield calculator, and plug in AMD's announced (at July Tech Day) defect density of <= 0.5 / cm^2, Barc die size of 283 mm^2, and see how bad it is yourself. About ~60 functional QC Barcs can be expected per 300mm die, across all bins. They have to do something with the remaining partially-broken ones. Of course, they should've just sold them as DC parts, but to try to get a little extra $$$ they'll sell some of them as Triple-core, and kill their own QC pricing in the process.

can be expected doesnt make it a fact.

And if u could read german you would see that he isnt even shure about calculating it the correct way (asin using the correct nrs) and that hes guessing some numbers and is making a few assumptions.
So i wouldnt put to much trust in those numbers.

and again pls check the date of the post in wich the guy makes those calculations 19.09.2006.

[adamsleath]

There isn't enough room for a tri-core segment in between dual and quad

looks like they just made some room

it is a way of selling partially defective K10s

splitting hairs, but yeah i agree.

anyways i'll end up with a phenom and a penryn

[terrace215]

looks like they just made some room

Shouldn't the codename be "Brokealona" ?

[adamsleath]

ahuk; woteva you say; perhaps this thread should just close now; so you can go back to your intel fanclub; and i can go back to counting all the money i am not spending

[terrace215]

can be expected doesnt make it a fact.

And if u could read german you would see that he isnt even shure about calculating it the correct way (asin using the correct nrs) and that hes guessing some numbers and is making a few assumptions.
So i wouldnt put to much trust in those numbers.

and again pls check the date of the post in wich the guy makes those calculations 19.09.2006.

Perhaps I wasn't clear: AMD provided the 0.5 def/cm^2 figure in JULY 2007.

[adamsleath]

ill take two Brokealona's please and cancel the penryn order. cheers

[informal]

Shouldn't the codename be "Brokealona" ?

Very mature and informative..

[terrace215]

Very mature and informative..

Not sure about the "mature" part, but it does capture the "partially defective" nature of the X3 parts, doesn't it?

[hollo]

Shouldn't the codename be "Brokealona" ?

lmao

[terrace215]

ill take two Brokealona's please and cancel the penryn order. cheers

TWO?

You expecting the first one to blow another core shortly after installation, or what?

[adamsleath]

ahuk; yeah that's right; tricores are a hoax and even if they do appear they'll break.
ahuk; id better order three then; but ill have to sell my intel crap first; then i can buy more amd crap.

[WeStSiDePLaYa]

It seems alot of Intel fan boys criticizing without reason.

As the cores go up, so will the die size, and so will the failure rates.

This is simple statistics.

Also, ATI and Nvidia have been crippling their high end parts for years, because the money is in the midrange and lower end, not the high end.

This move lets AMD fill out its product line in the area most profitable, while keeping material waste to a minimum.

This move is smart in many ways. And trust that if intel COULD do it, they would.

[terrace215]

Good article at extremetech.

While AMD says it "isn't about yields", it quite clearly IS about overall yields of fully-functional QC parts, and yields OF SPEED and POWER bins,as they admit in the same breath:

And the additional three-core chip also made smart business sense, AMD executives explained: if one of the four cores failed to achieve its rated speed, the company would either have to clock the entire chip down or simply discard the part. In what an AMD spokesman called a hypothetical example, AMD could launch a 2.8-GHz tri-core Phenom as well as a 2.6-GHz quad-core chip.

... ("hypothetical" example? I think not.)

One of AMD's goals is to keep its latest generation of microprocessor cores within the power constraints of the prior generation, which disabling a core helps achieve.

[Donnie27]

Because AMD's triple-core is essentially a quad-core Phenom chip with one core disabled, the announcement also suggests the possibility that the newest chip is simply a way for AMD to maximize its factory yield: Chips that roll off the manufacturing line with a defective core can be salvaged and sold as a triple-core, rather than wrong tossed into the scrap heap. AMD officials disputed that characterization.

"There's a yield component to anything that happens in a semiconductor company," acknowledged AMD's Brewer. But the primary motivation to release a triple-core chip was to improve the company's menu and management of products, he stressed.

First they wouldn't be tossed in the scrap heap but sold as Dual Core at least.

Different way to spin it, TELL THE TRUTH LOL! How about a partial truth. No one expects AMD to let these things go to waste and 3 will be better than 2 when all is said and done. It would be waste to disable that 3rd core. This does get more from each wafer and each of those four processors are not Cheap. Again, Good move, but slime ball piss poor marketing as usual.

Here, AMD could just ask their customers a Question. "Would you rather have 2 or 3 Cores?" Or "Would you rather we sell you a dual with a perfectly good core disabled or get the value added benefit of that 3rd core?" Then Price it right between the Dual and Quad cores. How refreshing, honesty. More choice is not bad. Oh and he could have corrected the Guy in the quote and not added more spin. He should have said we don't throw them away but sell them as Dual Core.

[terrace215]

And trust that if intel COULD do it, they would.

What makes you think Intel can't fuse off one core on one die of the two-die MCM making up a C2Q, if they wanted to?

They don't need to do it now, because they wisely went the MCM route, instead of the "native" / monolithic die route. So they can yield+assemble many more fully-functional QC parts per wafer.

But in theory, AMD did not have to go this route either, really. They could've sold triple-core parts as dual core (turning off 2 of 4 cores), and stuck to quad and dual. But as I said earlier, apparently they are having such a rough time yielding and binning QC parts that they need to try to sell tri-cores for more than dual, and not care about the impact they'll have on their own QC demand & (hence) QC prices. That only makes sense if they don't have many QC parts relative to Tri-core parts.