Why doesnt Intel make a 3 core cpu...?

[mhahnheuser]

AMD doesn't make triple core CPUs, it makes dual, quad, and hexa cores, and sometimes disables a core on their quad core to make a triple core. This was originally due to yield problems (the disabled core was defective), and later due to marketing decisions.

Intel is not in the "budget" market, it sells to the mid-upper range and it makes a much higher profit, economically it does not currently make sense for them to sell such a product while for AMD it does.

..if that is the case.....then why does Intel even bother with the bottom end at all? If it's sold as a three core and it has three cores then it's a tripple core....made by AMD. It's a fully independant core, fully featured multi-core processor and if three are good why not sell three instead of just two? That's a great marketing decision for the consumer.

..If it's established a market for them, then the consumer is the winner..and if Intel doesn't want to compete in this market then so be it....it's probably a relatively small niche one anyway. Of course the i series drops cores as it turbos up so at its max TDP it's not acting as a quad either as it makes room to let off some steam. And it's sold as a quad not a tripple, or dual or a mono-core.

...but of course crank up a Phenom and all the cores 2,3,4 or 6 all fire up...oh well that's the beauty of horses for courses....we all see differences from a different perspective.

 

[taltamir]

because selling a triple core means disabling one core in a quad core (a waste of a perfectly good core which cost money to produce), or it means designing a dedicated tri core chip (a waste of money and rather inefficient in using die space)...

it all comes down to manufacturing... intel doesn't want to ignore the mid-low end market... it just wants to sell the slowest most cut down processor possible to those market to avoid cannibalizing their high end. Their dual cores are much smaller (and thus cheaper to produce) then their quad cores.

Disabling an otherwise working core is simply a waste of money, but is preferable to the alternative as the underdog, but isn't preferable to the alternative to the biggest seller.
Basically there are a bunch of factors involved in here, all of them economical.

 

[GLeeM]

Disabling an otherwise working core is simply a waste of money

Isn't Intel testing/trying something where a buyer can enable more cache and/or HT by paying money for an unlock code?
And how would that be different in principle to triple core? Just asking

 

[VirtualLarry]

Isn't Intel testing/trying something where a buyer can enable more cache and/or HT by paying money for an unlock code?
And how would that be different in principle to triple core? Just asking

You know, you have a point there. Perhaps Intel would introduce quad-cores, sold as upgradable triple-cores, to unlock the fourth core. Interesting idea.

 

[justinm]

Because Athlon is an AMD Trademark.

(TriAthlon)

:biggrin:

Sorry, could not resist.

When 6GHz is low end they'll have a triple six and finally those folks with the "Satan Inside" stickers will have something to brag about! :biggrin:

hahaha

I have wondered about Intel 3-core CPUs but what would they name it?? Core2Trio would work but really does not flow well...

 

[taltamir]

Isn't Intel testing/trying something where a buyer can enable more cache and/or HT by paying money for an unlock code?
And how would that be different in principle to triple core? Just asking

An entire core is a lot more die space (and thus more expensive). And they are testing it, IIRC it increases clockspeed, enables HT, I don't remember if it does cache or not.

 

[JFAMD]

because selling a triple core means disabling one core in a quad core (a waste of a perfectly good core which cost money to produce), or it means designing a dedicated tri core chip (a waste of money and rather inefficient in using die space)...

it all comes down to manufacturing... intel doesn't want to ignore the mid-low end market... it just wants to sell the slowest most cut down processor possible to those market to avoid cannibalizing their high end. Their dual cores are much smaller (and thus cheaper to produce) then their quad cores.

Disabling an otherwise working core is simply a waste of money, but is preferable to the alternative as the underdog, but isn't preferable to the alternative to the biggest seller.
Basically there are a bunch of factors involved in here, all of them economical.

Interesting that you chose economics. If intel makes a quad, and one core is not functioning, they throw away the chip in your opinion, right? Well, that pushes up the average cost per chip because you are not recovering revenue on that unit. That is a poor business decision since they should be maximizing the yield to ship ratios.

Also consider that triple core allows for a price point that can be reached without having to compromise the quad core price points. There are solid, business reasons why triple core makes sense. I would think intel would be very interested in die harvesting and is probably doing that today with quads that don't make the grade as 6-core. (and they probably pushed quads down to duals. )

 

[Bartman39]

You know, you have a point there. Perhaps Intel would introduce quad-cores, sold as upgradable triple-cores, to unlock the fourth core. Interesting idea.

My whole thought...:thumbsup: Seems Intel is losing some bizz...? Or do they just dont care...?

 

[WildW]

I seem to recall (but now can't find reference to) Intel's boss Mr Otellini saying something along the lines of thinking it important that all the cores on the CPUs they sell should work. Whether that's just marketing talk I'm not sure, but it seems to be a bit of a point of principle not to sell "defective" chips as cut-downs.

 

[amenx]

I seem to recall (but now can't find reference to) Intel's boss Mr Otellini saying something along the lines of thinking it important that all the cores on the CPUs they sell should work. Whether that's just marketing talk I'm not sure, but it seems to be a bit of a point of principle not to sell "defective" chips as cut-downs.

If Intel was as financially pressed as AMD, I'm sure they'd come up with a new logic as to why 3 cores is a good solution to offset bad yields.

 

[exar333]

I seem to recall (but now can't find reference to) Intel's boss Mr Otellini saying something along the lines of thinking it important that all the cores on the CPUs they sell should work. Whether that's just marketing talk I'm not sure, but it seems to be a bit of a point of principle not to sell "defective" chips as cut-downs.

I love AMD and Intel, but this comment from Intel was a little perplexing. There is more to a CPU than just the core, and both Intel and AMD "cut-down" chips all the time due to non-functioning cache or poorly performaing clock speeds. If anything, Intel has been in a better position to sell a triple-core than AMD because they could keep the transisters present and completely inactive (this has been available since nehalem launched). If Intel had yield issues, you might see this, but they are obviously not having an issue with quads+ so they don't have a nig reason to do it.

 

[aigomorla]

wasnt the original tri cores failures of X4's which they bin'd down to recycle wasted parts?

That why i heard the originals had a chance to unlock the 4th core.

But i heard it was to save costs and optimize there foundry.

I could be completely wrong...

Intel on the other hand, doesnt worry about stuff like that.

 

[Phynaz]

Interesting that you chose economics. If intel makes a quad, and one core is not functioning, they throw away the chip in your opinion, right? Well, that pushes up the average cost per chip because you are not recovering revenue on that unit. That is a poor business decision since they should be maximizing the yield to ship ratios.

Also consider that triple core allows for a price point that can be reached without having to compromise the quad core price points. There are solid, business reasons why triple core makes sense. I would think intel would be very interested in die harvesting and is probably doing that today with quads that don't make the grade as 6-core. (and they probably pushed quads down to duals. )

Failed quads could become a dual and single (celeron?), assuming there is enough redundant logic on the chip to carve it into two independent cpu's.

Harvested quads could also make excellent mobile units.

Edit:
Now that think I about it, the first scenario in my post is very unlikely, too much logic duplication would be needed.

JF, could you express an opinion?

 

[JFAMD]

I seem to recall (but now can't find reference to) Intel's boss Mr Otellini saying something along the lines of thinking it important that all the cores on the CPUs they sell should work. Whether that's just marketing talk I'm not sure, but it seems to be a bit of a point of principle not to sell "defective" chips as cut-downs.

Hmmmm, on Xeon they cut down the QPI and memory speed on some models. Wouldn't it be even worse to take a functioning product and cut the capability?

 

[JFAMD]

Failed quads could become a dual and single (celeron?), assuming there is enough redundant logic on the chip to carve it into two independent cpu's.

Harvested quads could also make excellent mobile units.

Edit:
Now that think I about it, the first scenario in my post is very unlikely, too much logic duplication would be needed.

JF, could you express an opinion?

Smart people harvest dies across the board. All of the parts come out of a single design. There is only one Opteron core die that every single 4100 and 6100 comes from. It is all a manufacturing exercise.

I don't know enough about their parts to know how harvesting would work for them, but you would have to be insane to think that it is not happening because they are smart as well.

 

[Idontcare]

I seem to recall (but now can't find reference to) Intel's boss Mr Otellini saying something along the lines of thinking it important that all the cores on the CPUs they sell should work. Whether that's just marketing talk I'm not sure, but it seems to be a bit of a point of principle not to sell "defective" chips as cut-downs.

Hmmmm, on Xeon they cut down the QPI and memory speed on some models. Wouldn't it be even worse to take a functioning product and cut the capability?

I haven't talked with any folks in Intel's QRA in a long time, so I can't say with any certainty that my experience and insight here (from TI) overlaps with the decision matrix employed at Intel but at TI we were guarded against the prospects of releasing known partially defective dies with reduced functionality into field out of concern of the root-cause of the partial defectivity not being time-invariant and it would potentially cause failures in neighboring regions (circuits) of the chip.

For us it was just a matter of sampling rate versus inferred severity of the impact of the defect. Just because the defect takes out core 1 today doesn't mean that it won't "grow" to take out cores 2 and 0 tomorrow after a few hundred thermal cycles and applied voltage gradient.

Not that this isn't manageable, its just the management of this particular risk vector requires active investment into engineering both the products as well as the test/QRA process to build confidence around the ability to contain the liability.

What I'm getting at is that there is real math and science behind the business decisions surrounding risk mitigation and containment and this difference between Intel and AMD when it comes to core harvesting may just be nothing more complicated than a difference in how each has invested in their risk mitigation infrastructure at the DFM and QRA levels of the business.

 

[Phynaz]

hmmmmm.... I never thought of that scenario IDC.

 

[Bartman39]

I haven't talked with any folks in Intel's QRA in a long time, so I can't say with any certainty that my experience and insight here (from TI) overlaps with the decision matrix employed at Intel but at TI we were guarded against the prospects of releasing known partially defective dies with reduced functionality into field out of concern of the root-cause of the partial defectivity not being time-invariant and it would potentially cause failures in neighboring regions (circuits) of the chip.

For us it was just a matter of sampling rate versus inferred severity of the impact of the defect. Just because the defect takes out core 1 today doesn't mean that it won't "grow" to take out cores 2 and 0 tomorrow after a few hundred thermal cycles and applied voltage gradient.

Not that this isn't manageable, its just the management of this particular risk vector requires active investment into engineering both the products as well as the test/QRA process to build confidence around the ability to contain the liability.

What I'm getting at is that there is real math and science behind the business decisions surrounding risk mitigation and containment and this difference between Intel and AMD when it comes to core harvesting may just be nothing more complicated than a difference in how each has invested in their risk mitigation infrastructure at the DFM and QRA levels of the business.

Evidently AMD has taken this into consideration much more than Intel has done...? While AMD is of course the smaller player here possibly thats the reason they have taken more of a risk persay...? It seems to have worked out quite well as I see alot of mainstream end users buying and putting these what might be considered defective quad cores... Intel in the early years had its hands full with its processor the pent 90, but of course later years saw alot better quality and growth. I`m just surprised Intel seems have never considered this market but hey maybe they have a bigger better plan for things...? But if we did not have AMD and others then Intel would`nt be where they are now which is a simple plain fact.

 

[Idontcare]

Evidently AMD has taken this into consideration much more than Intel has done...? While AMD is of course the smaller player here possibly thats the reason they have taken more of a risk persay...? It seems to have worked out quite well as I see alot of mainstream end users buying and putting these what might be considered defective quad cores... Intel in the early years had its hands full with its processor the pent 90, but of course later years saw alot better quality and growth. I`m just surprised Intel seems have never considered this market but hey maybe they have a bigger better plan for things...? But if we did not have AMD and others then Intel would`nt be where they are now which is a simple plain fact.

It's not necessarily that AMD has taken more risk, it's only more risky if they did not institute risk mitigation procedures.

Instituting these procedures involved more expense, but that's what business decisions are all about, determining the best way to maximize ROI.

I'll give you another example, GloFo's CTI (continuous transistor improvement) policy and APC (advanced process control). The reason many IDM's don't make significant xtor changes midstream in production is because of all the risk it poses in terms of reducing the device lifetime reliability.

AMD (at the time) obviously made the business decision to invest the time and effort ($) into establishing a robust re-qualification process to ensure the risk is managed such that CTI does not come to represent a liability in terms of reliability and in-field fails.

It may well be that AMD's aggressive (by industry standards) harvesting techniques are made possible because of the existing risk-quantification and mitigation policies that were developed long ago as necessitated by CTI.

 

[JFAMD]

1. different architectures act differently.

2. nobody (who would actually know) has definitively said that intel does not do this. Based on their product selections (with core counts and cache sizes), it is a definite possibility.