Are cheaper CPUs the same as expensive CPUs that don't perform as well?

[corrado33]

I always wondered, how much binning occurs at say intel or AMD? Say, for example, are the i5-6600s just badly performing i5-6600ks? Or is the i5-6600k the same as the i7-6700k but just effectively neutered by a resistor? Has there ever been cases of this? Or is every number LITERALLY a different die? 

[Shimejii]

Binning or defects. 6600 is generally a lower yield 6700k where the sample didnt come out as good, so they disable Multi threading and sell it for cheaper. Generally its the same Die, unless its drastically different power wise.

[corrado33]

2 minutes ago, Shimejii said:

Binning or defects. 6600 is generally a lower yield 6700k where the sample didnt come out as good, so they disable Multi threading and sell it for cheaper. Generally its the same Die, unless its drastically different power wise.

So they don't purposely produce 6600s, they just always produce the highest tier chip and bin them then assign what they are? Where does it stop? Are the i3s the same die as the i7s?

[Shimejii]

2 minutes ago, corrado33 said:

So they don't purposely produce 6600s, they just always produce the highest tier chip and bin them then assign what they are? Where does it stop? Are the i3s the same die as the i7s?

Its how they save money, it would be a LOT more expensive to have each one on its own die. Yeah generally they are pretty similar or the same, although thats not always the case. 

[Jurrunio]

actually Intel don't bin CPUs for each model, they only bin for core count. Back when the sky-OC thing broke out, I've seen people overclocking 6400T (which one would think to be the absolute worse quality of CPUs) to hit 5GHz. It's at 1.45V no less, but for Skylake that's decent.

[corrado33]

4 minutes ago, Jurrunio said:

actually Intel don't bin CPUs for each model, they only bin for core count. Back when the sky-OC thing broke out, I've seen people overclocking 6400T (which one would think to be the absolute worse quality of CPUs) to hit 5GHz. It's at 1.45V no less, but for Skylake that's decent.

So are you saying that every Intel cpu (of a certain generation) that has the same core count is the same die? 

[Jurrunio]

14 minutes ago, corrado33 said:

So are you saying that every Intel cpu (of a certain generation) that has the same core count is the same die? 

Same die, different silicon quality.

[Curufinwe_wins]

40 minutes ago, Jurrunio said:

actually Intel don't bin CPUs for each model, they only bin for core count. Back when the sky-OC thing broke out, I've seen people overclocking 6400T (which one would think to be the absolute worse quality of CPUs) to hit 5GHz. It's at 1.45V no less, but for Skylake that's decent.

This is strictly speaking not true. Lower power processors are actually HIGHER binned chips, as those are binned for efficiency above all else.

(the rest of the explanation is more for OP)

 

And there is some binning involved as well at every level. However they bin from top down. So they take the whole pile of Skylake 4 core (GT2) dies, test a set of chips to be 6700k (as the highest margin part, this gets first bin) based on the expected market. If they find a dead core, it goes straight down to the next core bin (in this case i3/Pentium), otherwise more subtle defects go down to the next testing bin of any sort based on how sever the flaw is.

 

Then they test the other i7 -X bins (-S/-T/HQ if applicable), then finally the i7-6700, which is the lowest bin of the i7 parts in that generation.

 

They then take the whole pile of leftovers (many of which might be capable of meeting 6700k standards), bin the i5's in the same manner. Repeat with the whole leftovers for the i3 and then Pentiums. The next time they need to differentiate more chips, they take whatever undeliniated stock they have and repeat the process. Thus chips that are found to be defective end up in lower bins, but often the lower bins were perfectly capable chips (purely based on the increased demand for the lower bins overall).

 

 

 

36 minutes ago, corrado33 said:

So are you saying that every Intel cpu (of a certain generation) that has the same core count is the same die? 

Not quite, but for desktop parts this is almost always true. If they have the same core count, are part of the same family of processors, and use the same integrated graphics configuration (where applicable) they are the same die. As just about every desktop part sold is GT2 (if it has one at all), those ones tend to be of the same die. However, there are cpus with Iris graphics (GT3 and above) which use a different die.

 

[Jurrunio]

8 minutes ago, Curufinwe_wins said:

Repeat with the whole leftovers for the i3 and then Pentiums.

I thought Intel use different dies for CPUs with different core count (in LGA 115x anyway)?

[Curufinwe_wins]

25 minutes ago, Jurrunio said:

I thought Intel use different dies for CPUs with different core count (in LGA 115x anyway)?

For desktops generally not, unless they have a different GT configuration. More often the excess cores are lasered off (because the 4-core defected models aren't going to be wasted after all). This is the skylake example, which has arguably some of the most dies in a long time.

 

Top left is for -U (and -Y) processors, Top center is for -U with Iris, Top right is for normal mobile and embedded pentiums (though it count potentially be used for some desktop parts as well depending on the amount made of each). Bottom right goes in everything from 6700k to cutdown 6100 (almost every desktop and almost every non-dual core mobile). Bottom right is the unicorns for Xeon/mobile chips with 4C and Iris.

 

As as more clear cut examples of cores themselves being cutdown, Skylake-X only has 3 dies. The first is LLC which covers up to 10 core models on the platform. The second is HCC which covers 12-18 cores, and the third XCC covers 20-28 core chips.

 

https://en.wikichip.org/wiki/intel/microarchitectures/skylake_(server)