Good morning,
on my Setup (Ryzen 5900x, 2x16GB DR GSkill b-dies, B450 Tomahawk Max newest BIOS/Agesa) I prefer a consistent high performance over HIGHEST performance. So I cleared my CMOS and tested some default BIOS settings with SPD Profile (gdm0, cr2, 2133 cl20) from my Memory, to get the most accurate CPU Performance for my benchmark.
OS environment is Win10 which have low context switches and low dpc latency of a driver base, with minimal services to get as less interference as possible.
Benchmark software is PassMark PerformanceTest
Quick check at (all) Memory Tests for 3 runs each with reboot if Memory is consistent. Which I can confirm. I waited till OS is booted completely, CPU runs at idle and temperature is <35°C. Also after each reboot.
Now test subject CPU, tested different FCLK frequencys at Auto voltage, with PBO, without PBO, with and without maxed curve optimizations, fixed Allcore freq at different ratios between 36 and 41 and the performance score of the CPU tests (all) has always big variance. I tested 6 runs each with reboot and the variance is always +- 100 points, sometimes more/less. Even FLCL of 1600 has such big variance.
Are the zen3 architecture always so inconsistent or is some of my bios setting causing this? OS I think is fine because memory and disk benchmarks are +- the same...
Greetings - Powl
PS: I didn't test the maximum temperature of these tests but I don't think this cause this behavior, cuz even core cycler for 24h (avx, avx2, sse) at maxed tuned curve are below 80°