Search the Community
Showing results for tags 'cachemem'.
Hello! I founded strange behaviour in AIDA64 Cache & Memory Benchmark. L3 cache bandwidth on processor look lower then it's actually is. It's don't matter which settings i use in BIOS (even optimized default). Here is details: System spec: Hardware: CPU: AMD Ryzen 7 5800X (2049SUS); Cooler: Phanteks PH-TC14PE; Tim: IC Graphite Thermal Pad; MB: GIGABYTE B450M DS3H v2 Rev. 1.0 (BIOS: F60h, AGESA Combo V2 22.214.171.124, SMU 56.45.00); RAM: 2x 8GB Team Group Dark Pro DDR4-3600 (16-16-16-16-36-1T, FCLK=1800MHz); Software: AIDA64 v6.32.5600 AMD Chipset Driver v126.96.36.1991 So, here is two tests (low L3 cache bandwidth and normal): LOW I may click on each L3 box (Read\Write\Copy), but it never go above ~350 GB\s. At this stage problem was PBO setting in AMD CBS (in BIOS). After i disable PBO in AMD CBS and set it on Auto in AMD Overclocking (in BIOS) - i got ~400 GB\s. At this point CPU got correct all it's limits (PPT\TDC\EDC). Normal Now i got this. How? As you may guess - i set fixed frequency on all cores - 4425 MHz. It's only one setting from BIOS (well... actually voltage too, but anyway). Question: Why AIDA64 Cache & Memory Benchmark can't meassure normal bandwidth with auto setting on cores frequency? My guess is next: Cores are in "sleep" (i know it for sure, since i run AIDA64 with opened Ryzen Master window) and benchmark end faster then cores wake up. When used fixed cores frequency - benchmark sent data on "ready" core and data more correct.