-
Unfortunately had a brain fart mixing up my remaining Ryzen 9 PRO 3900 and the 5750G, could test the 3900 PRO on the Taichi, the 5750G is in my passive ASUS system, its heat-pipe-to-case-heatsink design is a pain to take apart.
-
But I’d bet that the 4750G and 5750G behave the same.
-
I overclocked the 2 x 16 GB Samsung ECC UDIMM DDR4-2400@3400, 1.20 V, am surprised how well that old memory clocks
STILL NO REPORTED ERRORS, which is extremely unlikely, comparing it to the 4750G’s results where the same memory modules reported errors at DDR-2933.
Based on your tests with the 5700X and my tests with the 4750G, 3700X and PRO 3900 ASRock messed up AMD’s firmware implementation for ECC error reporting regarding the parts that are used for chiplet-based CPUs (Matisse Ryzen 3000, Vermeer Ryzen 5000). Both Ryzen generations use the same IO-Die that is handling the system’s memory.
-
AM4 APUs (Ryzen PRO 4x50G/5x50G) are monolithic designs that use a different firmware module, these seem to currently work on the Taichi with ECC error reporting to the operating system.
-
Note: On none-PRO APUs AMD made the product segmentation decision to disable ECC support there, otherwise every Zen2/3 CPU has ECC enabled from AMD’s site and it’s up to the motherboard manufacturers to not mess it up.
Maybe you can relay that the 3900 PRO’s ECC error reporting is also broken.
The fix for that will likely fix the issues with your 5700X as well, since as mentioned these use the same IO-Die.
You should also mention that ECC error reporting works on ASUS motherboards with the very same CPUs, maybe that would add a little pressure to fix this.
I know that ASRock has been a sponsor but are you able to shoot them a message so that this might actually get fixed? Their AGESA 120A UEFI update is also quite late.