Triple-core's

[c0d1f1ed]

Hi all,

I was wondering if anyone expects quad-cores with just one defect to be sold as triple-cores. They could just market it like a kind of Celeron, instead of throwing them away.

With the Cell processor and GPU's it's also very common to have disabled components.

This would give people on a budget the chance to closely follow the multi-core revolution, instead of falling behind by at least a factor two.

Your thoughts?

Nicolas

 

[n7]

I'm guessing no.

But i suppose we'll see.

 

[Idontcare]

Can it be done - yes, to be sure it can be done.

Is it practical to do it - not likely very practical from a business model sense. To launch a product you have to be reasonably confident in your ability to supply that product.

No one wants to project to their shareholders a business plan that is relying on a steady-stream of broken parts.

Sure you can project a business plan that says initially you will recycle scrap chips to augment your supply of this new product line...but customers don't like the perception they are buying broken stuff or someone else's garbage so you better control the PR...BUT it must be part of a grander business plan that entails supporting that product over the long run.

In other words most companies don't look at what they have TODAY and say "let's package this doocky and sell it as a new product". Rather they sit down today and define next years product portfolio..."let's launch 3-core celerons next year"...and then you work backwards from the launch milestone to today on the calendar and chart your path on how you plan to get to the point of launching 3-core celerons.

ONLY in that environment would you say "hey I can save 6 months in my launch timeline if we start with recycling broken chips while the layout guys finalize their plans on the dedicated triple-core celerons".

That is how you get celeron products at launch that are merely cache-disabled versions of the premium product but then 2-3 quarters later the "real celerons" show up with die size reductions that can be leveraged into cost savings AND the shareholder feels your business model is sustainable.

 

[JeffreyLebowski]

I doubt we will see odd number of cores on things. I expect the Quad cores to become the mainstream and dual core will be the budget segment.

 

[simonnance]

surely as the C2Qs are 2 dies with 2 cores per die, either both cores on the die work, or none?

So you could sell them as Dual cores if they dont work propperly. Tripe core is unlikely IMO

 

[Kesshi]

Forgive me if I'm wrong, but wasn't this how the 486SX/DX thing came about? One was basically the other, but with a defective or failed part?

It's been a while since I've ever touched a 486, please forgive me if I'm incorrect.

 

[f4phantom2500]

Originally posted by: Idontcare
Can it be done - yes, to be sure it can be done.

Is it practical to do it - not likely very practical from a business model sense. To launch a product you have to be reasonably confident in your ability to supply that product.

No one wants to project to their shareholders a business plan that is relying on a steady-stream of broken parts.

Sure you can project a business plan that says initially you will recycle scrap chips to augment your supply of this new product line...but customers don't like the perception they are buying broken stuff or someone else's garbage so you better control the PR...BUT it must be part of a grander business plan that entails supporting that product over the long run.

In other words most companies don't look at what they have TODAY and say "let's package this doocky and sell it as a new product". Rather they sit down today and define next years product portfolio..."let's launch 3-core celerons next year"...and then you work backwards from the launch milestone to today on the calendar and chart your path on how you plan to get to the point of launching 3-core celerons.

ONLY in that environment would you say "hey I can save 6 months in my launch timeline if we start with recycling broken chips while the layout guys finalize their plans on the dedicated triple-core celerons".

That is how you get celeron products at launch that are merely cache-disabled versions of the premium product but then 2-3 quarters later the "real celerons" show up with die size reductions that can be leveraged into cost savings AND the shareholder feels your business model is sustainable.

Actually, the Core Solo's are Core Duo's with a defective core. Whenever they have a processor with some defective cache they just disable a portion of it and sell it as a cheaper chip. In the case of current quad cores (2 dual cores put together), if 1 core was defective they just wouldn't put it into a quad core; they would probably just disable it and sell it as a single core chip, then get a different dual core to sandwich with the other good dual core for a quad core.

In the case of native quad cores, though, instead of disabling the 4th core and selling it as a triple core, I think they'd probably just disable it and a 2nd core and sell it as a dual core because they already have dual core parts, why introduce a new product just to offload the few quad cores that have 1 damaged core when they'll already be offloading the ones with 2 damaged cores as dual cores? it's a lot easier to just sell them all as dual cores. The only way I could see them disabling 1 of those cores and selling it as a tri core chip is if they were getting a whole lot of native quads with 1 damaged core (although it seems pretty unlikely that they'd have enough to warrant coming out with a new product; if they couldn't meet demand they'd either have to scrap the product from their lineup or take perfectly good quad cores, disable 1, and sell those as tri cores).

Get it?

 

[f4phantom2500]

OP: I recall reading somewhere that Intel is planning on selling C2D's with a defective core as Celerons at some point (or maybe Core 2 Solo?) That's the cheapest thing I could see them doing anytime soon, I think they were planning on selling them for like $40 or so each.

 

[c0d1f1ed]

Originally posted by: Idontcare
No one wants to project to their shareholders a business plan that is relying on a steady-stream of broken parts.

Look at the GeForce 8800 GTS. It's based on the G80 chip with four clusters of 32 stream processors, but one of them is disabled. And it's insanely popular.

And it doesn't have to rely on a stead stream of partially defective parts. If you have more working quad-cores that's a good thing, and you can lower your quad-core price so more people buy it and you don't run out of triple-cores.

...but customers don't like the perception they are buying broken stuff or someone else's garbage so you better control the PR...

Most people won't know a triple-core has four cores with one disabled. And again I refer to the 8800 GTS. Nobody cares it has 1/4 disabled. They rather buy this one than a chip with two working clusters. As long as the price/performance is right they won't feel like buying broken stuff.

BUT it must be part of a grander business plan that entails supporting that product over the long run.

Why? In my local store I've seen Celerons and Semprons come and go. As long as I can buy a well-performing CPU for my budget I'm happy. I'd rather buy a triple-core than spend too much on a quad-core or get only half the performance with a dual-core.

ONLY in that environment would you say "hey I can save 6 months in my launch timeline if we start with recycling broken chips while the layout guys finalize their plans on the dedicated triple-core celerons".

If it's correct that broken dual-cores are now being sold as single-cores I doubt they won't have thought about how to properly disable a broken core in a quad-core.

That is how you get celeron products at launch that are merely cache-disabled versions of the premium product but then 2-3 quarters later the "real celerons" show up with die size reductions that can be leveraged into cost savings AND the shareholder feels your business model is sustainable.

Actually with new silicon processes and even higher number of cores I believe you can 'sustain' that business model. The PlayStation 3 is equipped with a Cell processor with only 7 out of 8 vector processing cores enabled (even if all eight work). Intel and AMD have a lot more freedom by just pricing them differently.

 

[firewolfsm]

If there is a chip with a faulty core (one of two) then it will either be dumped or sold as a single core. Only two working chips would go to make a quad. Triple cores don't make sense.

This is Intel's argument of why non-monolithic chips are better, they can be mixed and matched according to yields.

 

[BladeVenom]

Originally posted by: firewolfsm
If there is a chip with a faulty core (one of two) then it will either be dumped or sold as a single core. Only two working chips would go to make a quad. Triple cores don't make sense.

I agree with that. I don't see Intel making triple cores this generation. Maybe when there's four integrated cores and one is bad it will happen.

 

[c0d1f1ed]

Originally posted by: firewolfsm
If there is a chip with a faulty core (one of two) then it will either be dumped or sold as a single core. Only two working chips would go to make a quad. Triple cores don't make sense.

This is Intel's argument of why non-monolithic chips are better, they can be mixed and matched according to yields.

I'm not talking about dual-dual-core (should have made that clearer from the start). Both Intel and AMD will be producing single-die quad-cores in the near future. Yields will obviously not be very high, and the percentage of single-fault dies they would throw away would be gigantic. So it only makes sense to also sell triple-cores.

Soon after that I also expect them to start producing dual-quad-cores. And then they could also offer a six-core (dual-triple). Then, when they produce true octa-cores they could sell the ones with one or two failing cores as another six-core. Etc...

The clockfrequency race is over, but with applications quickly becoming multi-threaded I believe there would be more than enough demand for chips with an amount of cores between the high-end and budget solutions.

 

[tallman45]

Originally posted by: Kesshi
Forgive me if I'm wrong, but wasn't this how the 486SX/DX thing came about? One was basically the other, but with a defective or failed part?

It's been a while since I've ever touched a 486, please forgive me if I'm incorrect.

No, an SX core back then did not have a math coprocessor. So a 486 DX was a 486/487, the 486SX did not includes the 487 co-processor. They used the same process with the 386/387DX and 386SX

 

[JackPack]

Originally posted by: tallman45
No, an SX core back then did not have a math coprocessor. So a 486 DX was a 486/487, the 486SX did not includes the 487 co-processor. They used the same process with the 386/387DX and 386SX

Actually, the 486SX did have the coprocessor - it's a fairly well known fact. It was simply disabled. Bob Colwell's guest lecture at Stanford confirmed this.

 

[tallman45]

Point is the 486SX was not able to utilize a 487 coprocessor, wether it was disabled or ommited from construction. It was not a defective 486DX that was rebranded as an SX Chip

 

[lopri]

Originally posted by: firewolfsm
If there is a chip with a faulty core (one of two) then it will either be dumped or sold as a single core. Only two working chips would go to make a quad. Triple cores don't make sense.

This is Intel's argument of why non-monolithic chips are better, they can be mixed and matched according to yields.

 

[lopri]

Oh and Intel's first 'native' quad-core part is Nehalem, isn't it? That's like 2 years from now.

 

[TanisHalfElven]

Originally posted by: lopri
Oh and Intel's first 'native' quad-core part is Nehalem, isn't it? That's like 2 years from now.

that would be about the time celerons are dualcore or higher so sounds about right.

 

[c0d1f1ed]

Originally posted by: lopri
Oh and Intel's first 'native' quad-core part is Nehalem, isn't it? That's like 2 years from now.

AMD will have single-die quad-cores much sooner. I've read that Barcelone will have four separate power planes and clocks, which should make it easy to completely disable a core.

Yields of quad-cores will be terrible at first I think. So instead of throwing all those chips away they could price them at Intel's dual-cores and gain (back) popularity...