Athlon 3000G Overclocking: Is It Worth It?

1

Athlon 3000G iGPU Overclocking

The integrated graphics on the Athlon 3000G are fed by SoC voltage. SoC voltage also feeds the memory controller, and can affect your memory overclock. Safety is also a concern as the memory controller part is less voltage tolerant. Though Ryzen Master offers a separate setting for iGPU voltage, according to AMD the iGPU voltage is derived on-chip from SoC voltage. That means it can be lower than SoC but not higher.

Because of this, we didn’t quite follow a normal process to overclock the Vega 3 graphics. Instead we set SoC and iGPU voltage to a fixed 1.2V, anticipating this would work well with the memory and get good iGPU clocks while staying safe. Normally it’s best not to jump ahead on voltage like this. Nonetheless, with 3DMark Night Raid Stability Test we found that an iGPU clock of 1600MHz seemed stable.

Test Log
iGPU/SoC Voltage Vega 3 iGPU Clock Test Result SoC Power
1.2V 1100MHz Pass 25W
1.2V 1300MHz Pass 26.5W
1.2V 1400MHz Pass 26.7W
1.2V 1500MHz Pass 27.5W
1.2V 1600MHz Pass 29.4W
1.2V 1700MHz Fail
1.2V 1650MHz Fail
Finding a Stable Vega 3 Overclock

Armed with an iGPU clock of 1600MHz at 1.2V we leapt into our benchmark tests… and crashed in PCMark10. This illustrates two things;

  • 3DMark’s “stability tests” are not stress tests. They don’t directly test if your system is stable, they aim to test frame rate “stability” as inconsistency.
  • Temperatures matter for iGPU stability. PCMark10 stressed the CPU as well as the iGPU, adding substantially more heat. At the higher temperature, an otherwise stable clock may have no longer been stable.

The inconvenient truth of overclocking is that fully validating stability can be difficult. Push one part of a system and suddenly others are loaded differently. This is especially true for an APU where graphics and CPU cores share a chip, heating each other up. However, the same problem can affect all systems to some extent.

Our practical solution was to run GPU stress test Furmark alongside three threads of Prime95 Large FFTs. This replicated the crash, and having backed off to 1550MHz we passed both this and our benchmark tests.

1 COMMENT

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.