AMD’s second generation of Epyc processors sticks around after a while. Anyone who manages not to restart a server with the chip for almost three years will encounter the problem.

AMD has discovered a new bug in its Epyc 7002 Rome processors. Under this heading are AMD’s second-generation data center chips. Anyone who relies on the Epyc chips and doesn’t like reboots could be forced to do so after almost three years. AMD reports that cores on a chip encounter a bug that causes them to freeze after about 1,044 days of uptime.

The error is due to a core not waking up from CC6 hibernation after 1,044 days. Exactly when the problem occurs depends on individual configurations. If the error occurs, you need to reset the processor.

No solution planned

AMD has no plans to fix the problem. It is probably not worth troubleshooting via firmware or it has more disadvantages than advantages. In addition, the error does not affect too many users and the solution is simple. Anyone who restarts their servers before the 1,000 mark has nothing to fear anyway. An alternative is to disable CC6 hibernation.

Users will swear here and there. Finally, there are various (niche) scenarios in which it is assumed that critical servers will continue to work for years without a reboot. This is definitely an option for Linux servers, since you can basically patch them live. The error is specifically related to the Epyc Rome chips. AMD’s latest generation doesn’t seem to be affected.