Question So, Intel DOES "die harvest" after all... according to AT front-page article.

[CHADBOGA]

Proof? Seems like a straw man. Intel literally disables parts/features on chips all of the time. Look at the Xeon lineup or even something like 9700K which has cache disabled. Even the 9600K literally has two cores disabled.

It is known Khaleesi.

 

[LTC8K6]

I'd have thought the 9600K was just a binned 8600K.
Possibly an 8700K with the HT disabled.

 

[Arachnotronic]

I'd have thought the 9600K was just a binned 8600K.
Possibly an 8700K with the HT disabled.

It is not. It is a harvested 8 core die.

 

[beginner99]

It is not. It is a harvested 8 core die.

Harvesting is one aspect, the other are the meltdown/spectre hardware fixes which the 6-core die from the 8600k/8700k did not have yet. Probably cheaper to sell disabled 8-core than make a seprate 6-core die even for intel.

 

[LTC8K6]

It is not. It is a harvested 8 core die.

Perhaps it is, but I don't know how we would know that for sure?

How do we know that the 9600K isn't just a newly made 8000 series chip with the hardware fixes?

Is there a die shot somewhere?

 

[DominionSeraph]

Objective people do not care.

Larry is right about one thing though the intel fanboys rabidly denied that intel harvested chips for the past several decades.

Either point out where anybody has ever said Intel has a separate die for every SKU or admit the lie.
A google search for "Intel has a separate die for every SKU" comes up with bupkis.

 

[VirtualLarry]

Either point out where anybody has ever said Intel has a separate die for every SKU or admit the lie.
A google search for "Intel has a separate die for every SKU" comes up with bupkis.

I've never claimed, nor has anyone, to my knowledge, that Intel: 1) doesn't "bin", and 2) has a "separate die", for "every SKU". Nice strawman, though, it really brought out the hardcore Intel fans.

Binning is different than harvesting.

Binning, is where they take either a 2-core or a 4-core die, and test it (Shmoo plots FTW!), and then mark it, and sell it at certain speed grades. They may also disable features in the lower speed grades.

Harvesting, is where they take entire cores, and disable them, and sell them as a separate product. For example, a 4-core die, with one or two bad cores, being packaged and sold alongside (*intermingled with) 2-core dies, FOR THE SAME SKU. It is THIS that the Intel fans on this board, have previously EXPLICITLY DENIED HAPPENS.

But we know, from the linked AT article in the OP, that THEY DO.

 

[waffleironhead]

Paul Otellini in response to core harvesting and whether they will offer a tri-core processor: "We see a distinctive advantage in having all the cores on one die work."

https://arstechnica.com/gadgets/2007/09/intel-talks-penryn-nehalem-at-idf/

In reality this core harvesting has been going on since Nehalem, probably even earlier cutting dual cores down into single core.

https://arstechnica.com/gadgets/2009/02/intel-details-eight-core-xeon-cache-and-core-recovery/

 

[Rifter]

Either point out where anybody has ever said Intel has a separate die for every SKU or admit the lie.
A google search for "Intel has a separate die for every SKU" comes up with bupkis.

You obviously dont even know the difference between binning and harvesting so responding to this would be pointless.

 

[Arachnotronic]

I've never claimed, nor has anyone, to my knowledge, that Intel: 1) doesn't "bin", and 2) has a "separate die", for "every SKU". Nice strawman, though, it really brought out the hardcore Intel fans.

Binning is different than harvesting.

Binning, is where they take either a 2-core or a 4-core die, and test it (Shmoo plots FTW!), and then mark it, and sell it at certain speed grades. They may also disable features in the lower speed grades.

Harvesting, is where they take entire cores, and disable them, and sell them as a separate product. For example, a 4-core die, with one or two bad cores, being packaged and sold alongside (*intermingled with) 2-core dies, FOR THE SAME SKU. It is THIS that the Intel fans on this board, have previously EXPLICITLY DENIED HAPPENS.

But we know, from the linked AT article in the OP, that THEY DO.

Why would they NOT sell harvested 4 core dies if they can get some money for them? You are also discounting the possibility that Intel's production mix more heavily favors the higher core count parts and that if their demand forecasting was off, they take the higher core count parts and disable cores to meet demand.

You are making a mountain out of a molehill here.

 

[LTC8K6]

Why does anyone care?

If the chip performs satisfactorily for you, you don't care if it's a new die, a harvested die, or a die found in a dumpster.

 

[VirtualLarry]

Why would they NOT sell harvested 4 core dies if they can get some money for them? You are also discounting the possibility that Intel's production mix more heavily favors the higher core count parts and that if their demand forecasting was off, they take the higher core count parts and disable cores to meet demand.

You are making a mountain out of a molehill here.

Perhaps. Personally, I don't care, I would buy a "harvested" CPU. Or a core-disabled one, for marketing purposes. I once bought a tri-core AM2+/AM3 CPU, and unlocked the fourth core. It was like finding Gold. Lots of fun back in the day.

I was just pointing out the "wrongness" of some in the Intel crowd, because they though that "Intel was too 'good' of a company, to disable cores and sell disabled dies." Clearly, that's not true.

Anyways, I made my point. If the mods want to close this thread, due to "run it's course", or leave it open for discussion, and some choice links and quoted excerpts, pro or con to the issue, I'll leave it up to them. But I'll try hard not to respond any further.

 

[jpiniero]

I was just pointing out the "wrongness" of some in the Intel crowd, because they though that "Intel was too 'good' of a company, to disable cores and sell disabled dies." Clearly, that's not true.

This may also be a recent development with the i3's move to quad core. Clearly there will be some dies with defective cores. Before the i3 went to quad, core harvesting was probally not happening much.

 

[Arachnotronic]

I was just pointing out the "wrongness" of some in the Intel crowd, because they though that "Intel was too 'good' of a company, to disable cores and sell disabled dies." Clearly, that's not true.

The countless CPUs Intel has released over the years with disabled features/cores hasn't been enough to prove that point?

 

[jpiniero]

The countless CPUs Intel has released over the years with disabled features/cores hasn't been enough to prove that point?

Disabling cores has been extremely rare up to this point on mainstream, and I think it's mostly been some mobile parts. That's changing of course.

 

[krumme]

Disabling cores has been extremely rare up to this point on mainstream, and I think it's mostly been some mobile parts. That's changing of course.

Yep. It's easier to go from 8c to 6c than from 2c to 1c or 1c to 0c.

 

[DominionSeraph]

You obviously dont even know the difference between binning and harvesting so responding to this would be pointless.

What is "binned" about ECC support vs not?
What is "binned" about VT-x support vs not?
What is "binned" about VT-d support vs not?
What is "binned about TSX-NI support or not?
What is "binned" about hyperthreading vs not?
What is "binned" about L3 cache amount?
What is "binned" about integrated graphics or not?
What is "binned" about integrated graphics core count?

Intel has a range of products using the same die across a range of SKUs. Now prove that anyone has ever said that Intel uses a separate die for every SKU or admit the lie.

 

[Insert_Nickname]

What is "binned" about ECC support vs not?
What is "binned" about VT-x support vs not?
What is "binned" about VT-d support vs not?
What is "binned about TSX-NI support or not?
What is "binned" about hyperthreading vs not?
What is "binned" about L3 cache amount?
What is "binned" about integrated graphics or not?
What is "binned" about integrated graphics core count?

OK, I'll bite.

That has zero to do with either binning or harvesting, and everything to do with artificial product segmentation. Something Intel are masters of.



I'll get my coat.

 

[hojnikb]

why the hell would they make 2/4/6/8 core dies ?

honestly, im surprised they even bother with 2 core dies rather than just harvest these from 4 core...

 

[Topweasel]

With Sandybridge they announced that for 2c CPU's they were moving to an independent die for 2c production, as those were high volume sales and in doing so they would save a bunch in wafer costs. (thereby admitting that they harvested for first gen i3's).

That is what it comes down to. Over all cost maintaining a 2c design back in those days was so valuable because they were a major source of unit sales and the cost of running different production, designing another chip and so on was worth it. May still be to some much smaller degree today. Now they are really really really close to having to do cut down 4c's anyways just to make packaging work (need a CPU big enough to support all the traces they need) it's one of the reasons they kept pushing up the iGPU for the gens pre SL. They wanted to keep the core count at 4, but needed something to fill out the size of the die for the connections.

 

[Rifter]

OK, I'll bite.

That has zero to do with either binning or harvesting, and everything to do with artificial product segmentation. Something Intel are masters of.



I'll get my coat.

This guy gets it

 

[DrMrLordX]

With Sandybridge they announced that for 2c CPU's they were moving to an independent die for 2c production, as those were high volume sales and in doing so they would save a bunch in wafer costs. (thereby admitting that they harvested for first gen i3's).

First-gen i3s? You mean Clarkdale? I didn't think those were die-harvested from Lynnfield.

Anyway I'm gonna throw in with @VirtualLarry here. I too remember the forum banter from when AMD started selling die-harvested K10 chips (x3 720, etc). Intel had a separate die for everything, or so the line went. About the only area where it was obvious that Intel was die-harvesting was in their Xeon line, and even then they were just disabling cores for segmentation purposes rather than selling dice with defects. Or so the Otellini line went!

 

[Topweasel]

First-gen i3s? You mean Clarkdale? I didn't think those were die-harvested from Lynnfield.

Anyway I'm gonna throw in with @VirtualLarry here. I too remember the forum banter from when AMD started selling die-harvested K10 chips (x3 720, etc). Intel had a separate die for everything, or so the line went. About the only area where it was obvious that Intel was die-harvesting was in their Xeon line, and even then they were just disabling cores for segmentation purposes rather than selling dice with defects. Or so the Otellini line went!

Double checked, miss remembering slightly. It was with Sandy Bridge that they stated they were going to be doing dedicated dies. But Clarksdale was specifically a chip made for the mobile market that they also sold as dekstop I5's and I3's (also included a an igpu that Lynnfield didn't have). That's really where the tree seperated and the birth of HEDT started. Lynnfield really moved onto SB-E where Clarksdale moved up to 4 cores with the two new dies and all Desktop CPU's were really laptop CPU's after that.

But it was with SB that they designed dual cores to be closer to AMD's modules for quick design reasons and said they were doing a dedicated 2c die for wafer savings.

 

[realibrad]

Why would they NOT sell harvested 4 core dies if they can get some money for them? You are also discounting the possibility that Intel's production mix more heavily favors the higher core count parts and that if their demand forecasting was off, they take the higher core count parts and disable cores to meet demand.

You are making a mountain out of a molehill here.

The issues is not that they do it, its that others claimed Intel was not doing it when they clearly are.

 

[Mopetar]

honestly, im surprised they even bother with 2 core dies rather than just harvest these from 4 core...

It depends on how many they expect to sell. At some amount, it becomes more cost effective to make a 2-core design than to artificially hobble some of your 4-core parts. It doesn't prevent you from using defective 4-core parts as long as they still function as a 2-core part.

It would seem that's what Intel has done in this case. Given the sheer volume of chips that Intel sells, apparently the economics of it worked out for them.