How the hell do CPU yields and fault tolerance work? Is it literally just one transistor doesn't work in a core and that core has to be disabled? And how does the CPU do calculations correctly if a transistor isn't working? And how to they determine if everything works and identify where the defects are?
Bump for interest
>>62107029
The people who work at Intel AND come here are not the kind of people who can answer that.
>>62107029
Pretty much spot on. There also might not be transistors That dont work in a section of cache, which would mean that a block of cache gets disabled. That's why you get some i5's with 3M and i7's with 4M (L3) in mobile chips.
>>62107029
it doesnt work like that. So think of it like this. 80% of a CPU is execution units. Each like its own calculator. They all are in clusters and are repeated in a huge array like fashion. You have a man away from this directing the incoming instructions to whatever cluster of execution units is free and closest now. when a single transistor is not formed correctly only that execution unit fails to works. but when a CPU is built it has several hundred failed transistors scattered among the millions of working transistors. So if a hundred fails its still a core i7. if the man directing the transistor itself fails its a core i5 and so on. and apart from this some transistors have slightly narrow bus connecting it. sometimes a main bus itself is narrow. in that case it cant carry as much current across it. thats why some processors are clocked lower even while overclocking
>>62107029
>And how does the CPU do calculations correctly if a transistor isn't working?
A transistor "not working" is usually a probability distribution rather than a yes or no question.
They detect errors by doing redundant calculations. Like record whether there's an even or odd number of bits in a word and then verify that at every stage. Or if you're fancy, have two cores do exactly the same thing all the time and roll back results that don't match.