AMD prepares three-core processors

[KTE]

Some more info here,
http://www.planet3dnow.de/vbulletin/...11#post2893311

A year old?

Some geekdom there, but that's based on nothing and hence no different than me saying Intel has 3-core and 5-core processors ready for November. No matter how theoretically possible the conjectures are.

So, AMD has a tri-core part now. Not on the roadmaps nor any guides for so long?

This must be a last minute post-testing decision. Me thinks speed bins with QC are not going to be what we expected till 45nm K10 at least. It's looking like the 2.6GHz QC part is the 2.8GHz of Opteron K8>>> will get a few more out of it after 3-6 months or so, possibly 2 more SKUs, but the basic limit for speed/power.

While TC... should get 200MHz more than that at least and DC, 400MHz more I'm thinking. Ahh, competition.

[terrace215]

Then Price it right between the Dual and Quad cores.

That's not how the market works.

The DC and QC pricing is not independent of a TC option. Adding it will drive QC pricing down, because it will eat a lot of what would've been QC demand. Now, if they are super-constrained on QC *supply*, then it makes sense.

[WeStSiDePLaYa]

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.

I see your Fanboy gear is in overdrive.

You also seem to dismiss that this move will undercut intel's QCs and take sales from intel.

Face it, QC is high end part. High end parts are NOT where the money is, and its not where the sales are. If a consumer wants QC power, they will buy a QC. a TC will grab the market between DC and QC nicely, since most consumers, where most of the sales come from, cannot justify a QC or its price.

[Donnie27]

A year old?

Some geekdom there, but that's based on nothing and hence no different than me saying Intel has 3-core and 5-core processors ready for November. No matter how theoretically possible the conjectures are.

So, AMD has a tri-core part now. Not on the roadmaps nor any guides for so long?

This must be a last minute post-testing decision. Me thinks speed bins with QC are not going to be what we expected till 45nm K10 at least. It's looking like the 2.6GHz QC part is the 2.8GHz of Opteron K8>>> will get a few more out of it after 3-6 months or so, possibly 2 more SKUs, but the basic limit for speed/power.

While TC... should get 200MHz more than that at least and DC, 400MHz more I'm thinking. Ahh, competition.

Sure but what if that 4th core isn't broke but just real slow? What if they binned it and figured out that 3 Cores are good running at 2.2GHz and 4th can only reach 1.8GHz? Failed doesn't mean just broke. So, do they slow all the cores to 1.8 or sell 3 cores as 2.2GHz? This IMHO is what's *also going on as well. The mix and matching of 4 cores is almost endless. For instance a Dual Core@2.8 Costs more than a Quad Core at 2.2GHz. After binning the Quad Core AMD finds that 2 cores will only do 2.2GHz and the other 2 does 2.8 or better, sure AMD will kill the two slower cores and sell it as a 2.8GHz Dual Core. Wouldn't you?

I think folks want to be able to overclock each core independently because they know one or two of those cores might just be screamers. I agree with them.

[Donnie27]

That's not how the market works.

The DC and QC pricing is not independent of a TC option. Adding it will drive QC pricing down, because it will eat a lot of what would've been QC demand. Now, if they are super-constrained on QC *supply*, then it makes sense.

I only meant it as a possilblity. IMHO Tri-Core will not affect Quad Core in the broader OEM market. Might do that here though. Oh and its not like AMD is going to flood the Market with K10 anyway. All supplies of them will be tight and folks waiting on then will buy whatever they can get the grubby little hands on I might be wrong but I think it is a good Idea but just stupid marketing.

[Donnie27]

I see your Fanboy gear is in overdrive.

You also seem to dismiss that this move will undercut intel's QCs and take sales from intel.

Face it, QC is high end part. High end parts are NOT where the money is, and its not where the sales are. If a consumer wants QC power, they will buy a QC. a TC will grab the market between DC and QC nicely, since most consumers, where most of the sales come from, cannot justify a QC or its price.

Sure they can afford Quad Core, if it's Intel LOL! Sorry, couldn't pass it up. Anyway, I agree and think Tri-Core will sell but clearly I'm no expert.

[adamsleath]

Now, if they are super-constrained on QC *supply*, then it makes sense.

isnt that the only reason for discussing tricores at all.?
obviously phenomx4 yields will be low to start; as with all new processes.
if amd had 'good' x4 yields i doubt this thread would exist.

why complain about having more products in the market? the more choices offered to consumers the better.

many people continue to buy dual and even single cores.
quad cores are NOT mainstream sellers ->yet.
tricores could be another niche below quad cores and should/could increase sales in the mainstream market....and yeah they may hinder quad core sales...they could be a transition thing; tricores might not even eventuate. i'd rather have more choices.

[terrace215]

obviously phenomx4 yields will be low to start; as with all new processes.

The process isn't new, nor are low yields a function of all new processes. (Think non-AMD processes here. )

The problem is a function of the very large monolithic die together with a seemingly mediocre defect density (so large numbers of parts have at least 1 completely defective core) and what seems to be a lot of small scale variation in AMD's 65nm process (meaning that neighboring cores, even on the same die, are sometimes (often?) quite different from each other in terms of speed and power bin, which creates a huge binning problem for QC parts, as the lowest bin core "wins", and any high-leakage cores drive up the overall TDP).

[terrace215]

I see your Fanboy gear is in overdrive.

You also seem to dismiss that this move will undercut intel's QCs and take sales from intel.

It will? What if Intel prices its QC Penryn parts at the same or lower price as AMD prices a Phenom X3 with equivalent performance on 3 threads?

Then, Intel is offering a "free core" for the same price.

Why can Intel afford to do this? 45nm + MCM (instead of monolithic) means significantly lower costs per part, even comparing Intel "assembled" QC parts to AMD X3 parts.

[Cracker]

Three cores is stupid.

The only way I want a 3-core chip is if one of my quad cores lays down.

[Donnie27]

The process isn't new, nor are low yields a function of all new processes. (Think non-AMD processes here. )

The problem is a function of the very large monolithic die together with a seemingly mediocre defect density (so large numbers of parts have at least 1 completely defective core) and what seems to be a lot of small scale variation in AMD's 65nm process (meaning that neighboring cores, even on the same die, are sometimes (often?) quite different from each other in terms of speed and power bin, which creates a huge binning problem for QC parts, as the lowest bin core "wins", and any high-leakage cores drive up the overall TDP).

QFT! on this one and #111

[DeepThought86]

Wow, 4 pages of argument because AMD is planning to introduce an intermediate bin product, same as CPU and GPU manufacturers have been doing for years!

I guess the reason some people can't learn from history is because when you're 16 years old there is no history to refer to...

[KTE]

Sure but what if that 4th core isn't broke but just real slow? What if they binned it and figured out that 3 Cores are good running at 2.2GHz and 4th can only reach 1.8GHz? Failed doesn't mean just broke. So, do they slow all the cores to 1.8 or sell 3 cores as 2.2GHz? This IMHO is what's *also going on as well. The mix and matching of 4 cores is almost endless. For instance a Dual Core@2.8 Costs more than a Quad Core at 2.2GHz. After binning the Quad Core AMD finds that 2 cores will only do 2.2GHz and the other 2 does 2.8 or better, sure AMD will kill the two slower cores and sell it as a 2.8GHz Dual Core. Wouldn't you?

You know, INTC decisions for selling us higher FSB and L2 chips at higher costs was pure marketing and a profit making scheme, whereas it gave us hardly anything more worth the costs. We were already able to gain those FSB speeds from first launch 2GHz parts.

While AMD has done exactly the same here, a marketing and profit making scheme, it's head and shoulders above what INTC did, for us.

1: We have the option of cheaper SKUs than QC, for those that cannot afford them or have no need of them, and yet with more than DC potential performance.
2: Higher frequencies than QC, lower TDP, potentially better overclocking.
3: QC desktop market was only 2% by this year - now, we have more options, a new "in-between" and this'll drive Intel and AMD DC prices a little lower.
4: More competition is always healthy, and now even if Wolfdale desktop is too good for Kuma desktop, throwing X3 to compete with it will be a keen showing.
5: Clever by AMD to offer this, something new and not just following the leader, whilst IMHO for any microprocessor business, this is one hellova potential cost saving+profit+segment gain.
6: AMDs QC speed bin yields are low > above 2500MHz. B2 production for Q1 will just about suffice. How? Many meet the frequencies but not the TDP envelope they require. That you can be assured about before 45nm. Keep in mind, they can roll out a 3.4GHz X2 even now IMHO, but not meeting a competitive price offering/power envelope/volume numbers to be worth it. Bulk speed gains in K10, aswell as SSE and more IPC optimizations are being delayed for the 45nm move as they consume too much power/time for now - that concerns HT 3.0 & DCA 2.0 aswell. The IMC is a power hog with higher speeds at this fab level. However, to then salvage and reoffer those otherwise perfect QC parts as something only 3/4 the possible performance, rather than 0/4 (dead chip+huge waste of time/costs/tools), is simply intelligent decision making and not just blind following the book your competitor follows.
7: AMD must have already tested 3 cores and seen it offer a good advantage over DC, and yet reach higher speed bins than QC to offer it.
8: Its caught INTC of guard at the IDF. We might get 3 cores or something wacky from them now, it 'aint too difficult if your architecture is flexible.
9: Cost worries me - rumors of 700-1000 for X3 as they mention maybe perfect for many if performance is zehr gut, but it'll have to be competitively priced with the already low Q6600, which it won't be at those prices by far. The least they can do is offer a 2.4GHz X3 for 250-270, which will be otherwise "free profit" but still, it needs competitive performance.
10: If ones aware of microprocessor business, and industrial costs, realize that these chips were otherwise debts. Now they're purely profit. This means they're flexible and if all else goes to plan for AMD K10, they can drop the prices on these to ruin sales of Intel QCs (or have them dropped very low) and Intel DCs (or have them dropped low in price again).
11: It allows higher QC speed in attempts. Because the failures can still be salvaged as in-market, or even higher speed X3s and X2s, allowing to make a profit, whereas otherwise the cost/time would not have allowed for this.

IDK of its performance, but for us consumers, its a good thing, and if it performs decently above a DC, it'll be a hit for AMD too. This comes to mind:

SCOTT FULTON: We're not to make any assumptions now about whether this type of design moves over to the mobile space or not?

SIMON SOLOTKO: The only note I'd make is, the Griffin mobile processor for mobile is a substantially different architecture. So there's a very different set of dynamics, and we have substantial architectural differences between our mobile and desktop products. They are more than the architectural differences we have with Barcelona.

SCOTT FULTON: Same goes with server lines, right? We shouldn't make any presumptions there, right?

SIMON SOLOTKO: I certainly wouldn't.

[adamsleath]

The process isn't new, nor are low yields a function of all new processes. (Think non-AMD processes here. )

The problem is a function of the very large monolithic die together with a seemingly mediocre defect density (so large numbers of parts have at least 1 completely defective core) and what seems to be a lot of small scale variation in AMD's 65nm process (meaning that neighboring cores, even on the same die, are sometimes (often?) quite different from each other in terms of speed and power bin, which creates a huge binning problem for QC parts, as the lowest bin core "wins", and any high-leakage cores drive up the overall TDP).

isnt that what i said?

the proportion of higher clocking chips is always smaller than the rest, and yields increase over time usually as more and more chips are produced and new steppings are introduced.

if not tricore then ill have to wait for amd to iron out their fab of x4's.
thx again.

[Motiv]

It's just seems to me AMD are exploiting a market segment.

If Intel came out with this first I'm sure all the Intel fans would be jumping around saying what a good move.

when you sit back and look, from AMDs point of view it's a good move. It gives them valuable PR and makes Intel follow (which no doubt they will).

Although not showing now, the strength of having the 4 cores on 1 die will show at 45nm. This will give AMD the ability to have 6 or 8 core MCM packages.

[terrace215]

isnt that what i said?

No.

[zakelwe]

Assuming that 3 cores are handled well by software I don't see what the issue is, it gives more choice and surely that is never a bad thing? Do people complain BWM do 4, 6, 8 and 10 cylinder engines rather than just 4 and 8?

I guess this makes the future naming convention even more confusing though

I guess in the dual socket boards you could have the following number of cores - 2, 3 , 4, 5, 6, 7 and 8 . Or maybe not.

Regards

Andy

[evoic]

We live in a world that ripped open their wallets to buy pet rocks and Chia pets.

No one ever said that it had to make perfect sense to the masses in order to be consumed by them at clips of a thousand per day.

At first I thought of all the possibilities of why AMD shouldn't do this and how ridiculous it would potentially be.

After thinking it over......digesting some of the posts.....and considering the information presented - I'm starting to think it could be a minor saving grace for a company that could definitely use a minor saving grace or two.

[zerazax]

It's a smart move precisely because it has great marketing potential. People out there used to buy P4's because they could run at 3+ GHz even when K8's were kicking their asses at much lower clocks because the market likes buying big numbers.

The name of the game today, however, are core numbers. Why get that puny dual core when you get a triple core! Also, they are turning what would've been essentially wasted quad cores into triple cores that can save some of the costs. Also, a core being disabled doesn't mean it was because it died - as stated by others, if 3 cores can hit 2.8 GHz but one can only hit 1.8 GHz, disable the 1.8 GHz core and sell the triple core as 2.8 GHz, which might be priced HIGHER than a 1.8 GHz quad core. Voila, you just made more money.

[Cooper]

Thread closed