(used to have 2, one died)
That would make me immediately look to the RAM as the possible source or corruption. If it used to be a matched pair and one stick died, the odds of the other being on its way out are MUCH higher than normal. I would never trust that matched stick.
Similar issues even with just 2 DIMMs with some XMP/EXPO profiles not working on AMD systems because of board/CPU limits. It should technically work, but for whatever reason it just can’t handle it and speeds need to be dropped or the timings loosened a bit even though the RMA itself is rated for that.
Not that the higher speeds are even necessary for 90% of users outside extreme overclocking. DDR5 6000 is basically where you reach diminishing returns anyway, and that’s often where that limit seems to appear.
Yeah AMD’s memory controllers, especially DDR5 seem to have a lot more difficulty at high speed with 4 slots filled. I used to plan upgrades around populating 2 slots and doubling if needed a few years later, instead now you really need to plan to ignore those slots if you are needing memory performance for things like gaming versus raw capacity.
Dug into it, got into Memtest’s source code and discovered that the first pass is shorter on purpose so that it quickly flags obviously bad RAM. Apparently if you want to detect less obvious issues, you have to run multiple passes.
I thought it was common knowledge that Memtest needed to be run for multiple passes to truly verify there are no issues. Seems that’s one of those things that stopped being passed down in the community over the years. Back when I was first learning about overclocking around 2005 that was emphasized HEAVILY, with the recommendation to run it at least overnight, and a minimum of 10 passes.
Crazy, thought for sure it would fail testing.
Still wouldn’t trust it personally after a failed stick from a matched pair regardless of what the test says though.