Athlon 3000G Memory Overclocking
For memory, we picked out a set of old OEM sticks rated for only DDR4-2400. We fully expected these to be unimpressive. After all, not many users are likely to spend £200 on memory for their sub-£50 APU. To our surprise, the Athlon 3000G and Gigabyte B450M-DS3H still managed to push them to DDR4-3800. This speed is likely limited on the memory controller side. We still think most modern memory should do at least as well, taking timings into account.
There’s an awful lot to say about memory tuning, but for this article we’ll stick to sharing our test log. If you want to learn more about memory overclocking, we recommend the extensive guide by integralfx on github.
|3200||26-26-26-60-84||1.2V||Stable (short test)|
|3400||26-26-26-60-84||1.3V||Stable (short test)|
|3466||26-26-26-60-84||1.3V||Stable (short test)|
|3533||26-26-26-60-84||1.3V||Stable (short test)|
|3600||26-26-26-60-84||1.3V||Stable (short test)|
|3666||26-26-26-60-84||1.3V||Stable (short test)|
|3733||26-26-26-60-84||1.3V||Stable (short test)|
|3800||26-26-26-60-84||1.35V||Stable (short test)|
|3800||18-24-16-24-62 (tRCDWR 10)||1.35V||Stable (long test)|
|3800||18-24-16-24-62 (tRCDWR 10, GDM OFF)||1.35V||Crash on boot|
|3800||18-24-16-24-56 (tRCDWR 10)||1.35V||Boot fail|
|3800||18-24-16-24-60 (tRCDWR 10)||1.35V||Crash on boot|
|3800||16-24-16-24-62 (tRCDWR 10)||1.35V||Boot fail|
|3800||18-22-16-21-62 (tRCDWR 8)||1.35V||Memtest errors|
Because almost all DDR4 can hit DDR4-3200, we started out jumping to that speed. Since this is 50% higher speed, we set 50% higher primary timings. However because this left us close to the limits we then kept the timings the same as we raised speed. We ended up needing a voltage bump for DDR4-3800, and decided to stop there for speed as this was already way above all of our expectations.
For timings… we cheated. These same sticks were tested at DDR4-3333 before the author joined play3r. Since DDR4-3800 is 14% faster, we took those results and loosened the timings by around 14%, rounding up, to get a good set of timings for these specific sticks at DDR4-3800. Dropping timings individually in small steps would have got similar results but taken a full day. We did try to tweak them a little more, but didn’t get anywhere past the calculated settings.