Retailer
Always RMA the CPU directly to AMD as the retailer will just run some benchmarks that don't seem to reproduce the light load crashes so they just end up returning the CPU.Is this a rma situation for the cpu?
These kind of crashes seem due to instability at light SC loads and as someone eluded to earlier in the thread that you can reproduce them on a healthy CPU by pushing either the boost clock speed higher in PBO or using too much negative offset in curve optimizer.
If it's behaving like that at stock then it's definetly an RMA job but fortunately AMD seem to recognise this issue and will just replace the CPU unlike the retailers who don't seem as clued up and are more used to testing for all core stability on Intel CPUs.