A simple CPU test to determine if the power used by the DDR4 memory on AM4 directly affects (subtracts from) the Platform Power Tracking (PPT) budget allowed by the Processor package. The answer appears to be, no. The test is performed by running 4x 16GB dual rank DIMMs at 1.4V in a CPU-power intensive test (Prime95 Small FFTs) and measuring clock speeds against 2x 4GB single rank DIMMs at 1.2V. For the latter test, the FCLK is set to decoupled mode and remains at 1600 MHz and 1.0V for the SoC, in order to normalise the power used by the Infinity Fabric between both runs. The PPT is set to 75W and the EDC and TDC are set high, to 150A, to ensure the PPT is the limiting factor.
The idea is that if the DRAM power subtracts from the PPT; the CPU cores will have less power to operate within the PPT and will as a result, run at lower frequencies, however, that doesn't appear to be the case. It's important to note that the UMC (unified memory controller) is running at different clock rates as this is tied to the DDR4 Memory clock directly. (1600 vs 1066) so that can account for sub-1W power discrepancies but it's very small.
Board is Gigabyte X570 Aorus Pro. Bios: F33.
Comments