I never found any settings that would consistently train to stability with raptor lake and 2x32 GiB A-die @ 6800 MT/sI did however manage to get to what seems like properly stable ram 45 days ago. As an experiment i am simply going to keep my current memory training for as long as i can and see if it eventually becomes unstable (i did read people claim it could become unstable due to "drifting" but i saw no evidence provided).I did change my the msi fast boot setting to "no training" which means that no matter what setting i change in bios it will not retrain and the bios will simply ignore me if i try to change some setting that would require new training to implement.I have verified that the motherboard will keep the old training even if i reboot or power off the PC. My guess is that no new training will occur unless there is a power outage (or if i completely cut power to the motherboard).The inconsistency with regard to stability after training was mostly in tests that are highly stressful for the imc such as mprime large FFT, some y-cruncher tests, stressapptest.I did actually run into stability problems later but that was caused by improper CPU tuning (was found thanks to the y-cruncher pi benchmark).
>>108629046what the heck is memory training? never had to train my memory to do anything
>>108629074The bios configures the ram based on the settings in bios. It does that without you asking for it and it typically results in longer boot time when it happens. Some motherboards display a code to indicate memory training but even then you will not notice that unless you pay attention.
>running a memory test from the OSWhy are winfags so fucking retarded?
>>108630157Memtest86+ isn't particularly efficient as a memtest and will miss various instability issues.
>>108630981Skill issue
>>108631084There is no strategy you can used to compensate for the stresstest being ineffective besides trying to run it for longer (which often isn't even going to work anyway).
>>108629074computers always did memory training. the old computer bios used to count the memory on screen.
i prefer thhe kernel cmdline memtest=N
>>108629074the connection between your CPU and main memory is a ridiculously fast and complicated bus. training is just making sure you are talking at the right speeds.
>>108630157memtest86 / 86+ is some unc shit, my system would be up for over 30 days without problem but a week longer and it'd freeze, it did this three fucking times til I bumped up my VDDQ/VDDIO voltages a tad since auto at 1.1v was too low.
>>108630157The only way to reliably test with modern chips. Things are so dynamic, without loaded drivers your results can be way off, work fine in mt86 but not in the OS.Ironic calling others retarded.
>>108632384Most ram overclockers do not bother to run anything close to proper rigorous stability testing. They might run like 1 test for 2 hours and then think that everything is fine. Perhaps that is blissful ignorance for them but having your PC randomly freeze or windows get corrupted does not seem like some great bliss to me.
