What makes some CPUs perform better than others (other than purposely disabling features?) (Binning?)

[corrado33]

We all know that processors are binned. The best are sold as extreme editions with higher clocks and more features, and the worst are sold (currently) as i3s (probably.) (For intel at least)

 

So... what.... exactly makes some processors better than others? Yes, it's the quality of the die, but what EXACTLY in that? Why do some dies turn out more suited for high clocks than others? CPUs are essentially a bunch of transistors... so I'm assuming a "good" die has more transistors that work compared to a "bad" chip? Or is it simply how much voltage each transistor takes to switch on? For a lower quality chip could it literally be that so many of the transistors failed that only X/Y cores are functioning? (So if a die is built with 4 cores, but then it turns out badly and 2 cores aren't working, would they bin that as a 2 core chip and slow it down so long as those cores work well?)

 

So, for example, if I have a chip that I can OC to 4.5 GHz, but not to 4.6 GHz at a certain voltage, I'd imagine it's because the rising and falling of the clock signal does not reach a high enough voltage (or does not stay there long enough to deliver the amount of current needed) to trigger the transistor? (I'm making this assumption based on the fact that some Mobos give you the option to increase the rising and falling speed of the clock signal.) If this is true, then why not have the clock signal rise and fall as quickly as possible? Why even give users the option? 

 

Also my hypothesis makes sense based on the fact that you can get to a point where the processor will boot, but will fail calculations (AIDA calls it a "hardware failure"), which would mean that SOME or MOST of the transistors can be triggered at this voltage, but some of them in the compute core can't, therefore data gets corrupted. Or rather, since each transistor has a voltage drop associated with it, there isn't enough voltage to trigger all of the transistors in the circuit?

[Jurrunio]

resistance of the transistors is one of the many factors that affects performance. Think of it as a CPU that needs more voltage to hit the same clocks compared to others.

 

while theoretically possible, manufacturers do not spend their time in making their CPUs run clocks suitable for the worst core specifically. That just means selling a ton of SKUs, and that's costly.

[corrado33]

25 minutes ago, Jurrunio said:

while theoretically possible, manufacturers do not spend their time in making their CPUs run clocks suitable for the worst core specifically. That just means selling a ton of SKUs, and that's costly.

What did you mean by this? I assume that intel checks each and every die for performance (and... bins them accordingly)? Or do they wait until it's in a chip and check it then? 

[X_X]

Leakage and thermals

[corrado33]

6 minutes ago, X_X said:

Leakage and thermals

Leakage as in leakage B->E or B->C leakage for the transistors? Or leakage as in one base to another since they're so close? As for thermals, this seems more like a consequence than a cause, and I can't see how it'd be different on otherwise identical dies. 

 

Basically, if you can run the CPU at lower voltages, it uses less current, so it'll heat up less. Heat increases resistance so therefore the hotter the CPU the more voltage the transistors will need. 

[X_X]

Dies are not perfect therefore not identical.

[corrado33]

5 minutes ago, X_X said:

Dies are not perfect therefore not identical.

But what I'm saying is that thermals are related to current (IR heating). In order to have more current you either need to be A. using more transistors or B. needing a higher voltage and therefore more current to use fewer resistors. Therefore, consequence, not cause. 

[X_X]

Thermal resistance between junction(s) and IHS also affect temperature.

 

Try searching for class testing, an example although somewhat old https://www.intel.com/content/dam/www/public/us/en/documents/research/2000-vol04-iss-3-intel-technology-journal.pdf

[Jurrunio]

1 hour ago, corrado33 said:

What did you mean by this? I assume that intel checks each and every die for performance (and... bins them accordingly)? Or do they wait until it's in a chip and check it then? 

1 hour ago, corrado33 said:

and slow it down so long as those cores work well?)

Intel do bin dies, but SKU of the same core/thread count and TDP but different frequency aren't binned separately.

 

[corrado33]

8 minutes ago, Jurrunio said:

Intel do bin dies, but SKU of the same core/thread count and TDP but different frequency aren't binned separately.

 

OOOOooooooo...... Interesting.....