CCX Latency Testing – Pinging between threads

With our custom-made thread to thread communications benchmark, we can measure the “ping time” of inter-thread latency. This allowed us to discover and teach the community about the CCX to CCX latencies and now we can see how the increases in memory frequency lowers those latencies.

Though it is hard to see in this visualization, the Ryzen 5 1600X shows thread to thread communication for threads on the same CCX remaining mostly unchanged as we increase memory speed but, more importantly, the latency on inter-CCX communications is dropping as the memory frequency increases.

For a better view:

This takes ONLY the CCX to CCX latency results and averages them across all runs on a single thread tested, and then averages the averages across the threads. At DDR4-2133 memory speeds the Ryzen 5 1600X has a 140 nanoseconds latency for thread to thread communications if those threads are on separate CCX modules. When running at DDR4-3200 that latency drops to 111 nanoseconds, a 26% improvement! Even though this is just a component of a very complex series of systems that make up the performance of a CPU, this clearly shows the inherent advantage of raising memory speed and internal fabric interconnect speed.

« PreviousNext »