Memory testing on AMD's Ryzen 2400G APU
Now that we have been able to stabilize the Ryzen platform, we can also see together what the impact of each setting is on performance. We already know that the performance of the Zen(+) architecture depends very much on the RAM frequency – let's see if the IGP performance increases in the same way. We used our classic game collection with two different settings (ultra and medium). At least the games that somehow make sense to run them in full HD.
Game | Ultra-Settings | Medium Settings |
Ashes of the Singularity: Escalation | Extreme, GPU Test (DX12) | Standard GPU Test (DX12) |
Tom Clancy's The Division | Ultra, VSync OFF (DX12) | Medium, VSync OFF (DX12) |
Far Cry 5 | Ultra, Blur OFF, VSync OFF | Normal, Blur OFF, VSync OFF |
Rise of the Tomb Raider | Very High, SMAA, HBAO+ off (DX12 ) | Normal, FXAA (DX12) |
Civilization VI | Ultra/Ultra, MSAA 8x, VSync OFF, Blur OFF, GPU test | Medium/Medium, MSAA 2x, VSync OFF, Blur OFF, GPU Test |
Total War: Warhammer II | Ultra, Test Scene | Medium, Test Scene |
In absolute terms, all games gain in FPS with increasing memory frequency, including the classic error limit in this type of testing. It's just stupid that some games have cheekily refused to complete or even start the benchmark runs with an DDR4 frequency of 3400 MHz without errors. No wonder, by the way, because AMD only guarantees the operation of the Ryzen 2 with a maximum DRAM frequency of 2933 MHz. The rest is lottery and luck. In the following gallery, we have listed the individual games with the comparison of the settings and depending on the memory clock rate:
After elections, people always look at the bars of winnings and losses, as well as the nice average rating. The averaging of the number of FPS achieved in the six games tested gives us at least a real idea of the average percentage gain (or loss) compared to the DDR4-2933, the fastest memory officially used with the Raven-Ridge APUs. achievable. DDR4-3200 offers on average only 3.2% to 5% full HD performance, depending on the level of detail.
Conversely, the difference between DDR4-2133 and DDR4-2933 is about 12%! For DDR4-2133 and DDR4-2400 (as suspected and not surprising): both are clearly to be avoided, since the difference in graphics performance compared to the higher speeds is simply too big. Of course, I did not like to deprive anyone of these blanket results either. In the following gallery we see this again from above, only just generalized in four graphics and also subtly standardized in the percentages:
Summary and conclusion
With AMD's APUs, it's much more complicated to find the right RAM with optimal settings, because the Ryzen memory controller is much more moody. APU performance also does not scale linearly to the RAM frequency. In the end, the performance gains are much lower overall, while the difficulties with the correct (and therefore really stable) configuration of memory increase exponentially. One could even say that at the moment, and despite the announced certifications, the rates higher than DDR4-2933 can hardly be honestly guaranteed if one really attaches importance to real stability. It will always be necessary to tinker with the timings and tensions and patiently find his own attitude, which is then also "rockstable".
Kits with DDR4-2133 and DDR4-2400, which so frustratingly limit the performance of these processors, are nothing for the APUs. The best compromise is actually to stick to AMD's maximum standard frequency: DDR4-2933. That's the purse, every (possible) little bit more speed is then bought expensively with money. And time to invest in testing. I see the biggest bug devil still in the BIOS of the motherboards, here AMD has to improve with better AGESA versions, because with the pure Ryzen 2 CPUs without graphics it is also possible.