According to PassMark software, DDR4 memory latencies on AMD CPUs are more than 4 times higher than latencies on Intel CPUs. This has been confirmed on my Corsair Vengeance LED 2x16GB DDR4 PC4-24000 [CMU32GX4M2C3000C15R], operating at 2666MHz due to motherboard limitations (the RAM itself supports 3000MHz), with timings 14-15-15-15-31: the CPU is Ryzen 1800X and the latency is about 70 in PassMark tests.
So is this an issue in the test? Or is DDR4 memory on Ryzen CPUs really more than 4 times slower on some data access patterns? What are these patterns?
By the way, Intel Burn Test also showed strange result: I understood that it heavily relies on memory, and for 8-core 16-thread Ryzen 1800X@3.6GHz with DDR4@2133MHz it showed only 75GFlops, while old 6-core Phenom II 1100T@4Ghz (overclocked from 3.3GHz stock) with DDR3@1600MHz with tight timings (like CAS 8) and lots of L1&L2 cache showed almost the same result: 73.8GFlops.
So can someone spot when specifically such new system as Ryzen 1800X based performs no better than a several generations older Phenom II based system?
Best Answer
Ratio of latencies is not ratio of speeds, e.g. having half latency does not mean having double speed. It has a few percent affection on speed.
Given that we are comparing same types of memory (e.g. DDR vs DDR), clock rate is the primary factor of speed while timing (CAS latency,...) is the secondary one.
Even now, Intel traditionally makes better memory controllers than AMD. Ryzen competes well in performance with Core i, however, I don't know how much its (on die) memory controller is approved.