Hurdles and Solutions: AMD's APU Ryzen 5 2400G tested with different storage settings

1 - Aufgabenstellung, Testsystem und ein kniffliges Puzzle 2 - Wir lösen auf: die stabilsten Timings für die Taktraten 3 - Performance-Tests, Zusammenfassung und Fazit

The test system

In our article "White and Fast: KFA2 Hall Of Fame Extreme DDR4-4000 in Test" we have already seen that the Core i7-8700K and the Ryzen 7 2700X benefit from an increase in memory frequency and also offer increased CPU and game performance. However, we also wanted to know whether this also applies to a Ryzen APU and its integrated Vega graphic and, above all, to what extent. For this purpose, a Ryzen 5 2400G APU with four cores (SMT, eight threads) with integrated Vega-11 graphics unit was used, because there is currently no faster.

When it comes to memory, we rely on the G.Sill Sniper X DDR4-3400 CL16, which AMD supplied for the sample, whose exact data I also quickly summarized in a table:

RAM Timings on Ryzen: We Explain the Puzzle

I wanted to do these tests on the well-known X470 platform, but even the first runs proved unexpectedly problematic: even at low DRAM frequencies, the platform (with the latest BIOS or the previous version) was unstable in a cow on stilts as soon as I really demanded the graphic. First I had the Ryzen 5 2400G and then the G.Skill kit in my suspicions, but it turned out that it was the motherboard that created the problems. The walk to the warehouse was logical and so I switched to another model, an Asus ROG Strix X370-F Gaming….

The G.Skill Sniper X Kit is ryzen compatible by self-declaration, which is also confirmed by a bold sticker on the packaging. However, it still seems that individual Ryzen platforms are still quite moody as a diva before brushing their teeth, which of course could also be due to the XMP profiles of the memory modules (an Intel technology that is only used as a small one as a reminder). However, both MSI and Asus also provide automatic DRAM timing management using the A-XMP and D.O.C.P. capabilities in their respective UEFIs. at least.

In practice, I still had to snap the ns duration for each latency type and then manually convert it into bagged timings using a kind of "magic formula" in a simple calculation (timing = latency in ns / clock cycle time). For a better overview, I have here tabularatable a summary of the latency in ns and the corresponding calculated times for the G.Skill Sniper X DDR4-3600 CL16 Kit, whereby then, taking into account certain rules, of course also on the next suitable unit must be rounded (e.g. tRC must be greater than or equal to the sum of tRAS + tRP).

Once we have calculated these timings, all we have to do is configure the UEFI accordingly. The timings have been set to achieve approximately the same latency in ns and only the clock frequency of the memory becomes the only variable parameter during the individual runs. Anyone who is now afraid of too much theory is reassured. On the next page I test the whole thing with some motherboards and the usual replica settings in deciphered form.