For the performance of processors, not only the speed of the CPU cores plays a role, but also how fast they communicate with each other.

(Image: Denis Fröhlich / c’t)

Processor performance: Core-to-core latencies of modern CPU cores in detail Ring-Bus versus Mesh Unequal core quartets Article in c’t 2 / 2021 read The construction of modern processors is currently undergoing a change. Until a few years ago, large “monolithic” chips were still popular, combining all CPU cores and functions on a single semiconductor die, but the trend is now towards several individual chips. For example, AMD uses so-called chiplets for the powerful processors of the Ryzen, Ryzen Threadripper and Epyc series, which are several separate silicon dies on a common carrier. The CPU cores, which are responsible for the actual computing work, are located in up to eight CPU core dies (CCD). AMD packs all other functional blocks, such as the memory controller, PCI Express Root Hub and I / O functions, on a separate I / O die.

The chiplet design increases the yield than ever before The larger the area of ​​a semiconductor chip, the higher the probability that an error will creep in during manufacture. In addition, there is a maximum of chip area that can be exposed with today’s technology. It is approximately at 800 mm 2 .

Intel uses monolithic chips for the majority of its CPUs. This not only applies to the quad cores of the 09 – Nanometer mobile processors Core i – 1100 and the desktop CPUs Core i – 10000 up to 10 cores too, but also on the server chips of the Xeon series with up to 28 Cores.

Access to all contents of heise + exclusive tests, advice & background information : independent, critically founded c’t, iX, Technology Review , Mac & i, Make, c’t Read photography directly in the browser register once – read on all devices – can be canceled monthly first Month free, then monthly 9, 95 € Weekly newsletter with personal reading recommendations of the editor-in-chief Start FREE month Start the FREE month now heise + already subscribed?

Sign in and read Register now and read the article immediately More information about heise + Processor performance: Core-to-core latencies of modern CPU cores in detail Ring-Bus versus Mesh Unequal core quartets Article in c’t 2 / 2021 read