Renowned tech blogger Roman “Der8auer” Hartung decided to see firsthand and show everyone that there is not so much difference between 14nm transistors in Intel processors and 7nm transistors TSMC in AMD processors, as marketers want us to believe. Common sense says that the numbers “14” and “7” are two times different, but in practice, for technical processes, everything is different.
Punching samples from processors for examination under a scanning electron microscope (Der8auer)
–
–
–
As test subjects for Der8auer chose the best Intel and AMD have in mainstream PC products today. He took samples with transistors from a 14-nm Intel Core i9-10900K processor (14 +++ process technology) and a 7-nm AMD Ryzen 9 3950X processor from the Taiwanese company TSMC. Sections of processors with L2 cache were selected as samples for study. Transistors in logic blocks have a scatter in the sizes of gates and edges, while in the composition of memory cells they are more or less the same, and represent a regular (repeating) sequence that is convenient for comparison.
Comparison of gate pitch and vertical edge of transistors in Intel and AMD processors (Der8auer)
–
–
–
Scanning each processor sample under a scanning electron microscope revealed that 14nm Intel transistors have a gate width of 24nm, while 7nm AMD / TSMC transistors have 22nm gate widths (gate heights are also about the same). As we can see, we are not talking about 14 or, moreover, 7 nm. To justify modern marketers, let’s say that this discrepancy began after the 90 nm process technology and accelerated after the transition from transistors with planar to vertical gates.
–
–
Nevertheless, TSMC’s 7nm process technology turned out to be slightly better than Intel’s 14nm process technology with the ability to accommodate 90 million transistors on one square millimeter. In the case of Intel’s 10nm process technology, this company is a little ahead, as it promises to place 100.8 million transistors for every square millimeter. But this comparison also has its own nuances.
So, three years ago, for the 10-nm process technology, Intel suggested not just counting transistors from one or another block on the processor, but choosing them purposefully and using weighting factors. For the Intel metric, transistors are taken from elementary logic. First, there are two-input NAND transistors (not to be confused with NAND flash memory), which are assigned a weight factor of 0.6. Secondly, transistors from triggers with a minimum of 25 gates are used, which are given a factor of 0.4. From this data, the density of the transistors is derived, which, for example, as in the Der8auer comparison, will differ from the real one for transistors in the cache memory.
And again, someone will have to take samples of processors, dissect them and literally manually count the real number of transistors. And so it will be until the industry comes to an agreement on a new metric, and if marketers do not stick their nose in there.
–
If you notice an error, select it with the mouse and press CTRL + ENTER.