Hello.
I have just recently installed a Ryzen 9 5950X on my PC, and have been playing around with it.
I have a custom liquid cooling loop on my X470 Crosshair VII Hero WIFI motherboard and was looking into what kind of stable manual clockspeed I could get using 1.3V as the core voltage. I was using Cinebench R23 as a stability test. My first initial test of all cores at 4.6GHz failed despite temps being in the mid 70C range. It was a hard crash as soon as the Multicore test was started. I'll highlight some additional tests in the table below.
What has been interesting is when I started to clock my two 8 core CCDs independently. I was aware that Zen 2 had shipped with CCDs that were of vastly different quality in processors that contained more than one. I figured Zen 3 was likely similar. What I noticed was that Cinebench R23 fails as soon as I raise CCD1 above 4.4 GHz with 1.3V. At 4.5 GHz I get an error mid run, at 4.6 GHz the system crashes. However, CCD0 is able to run at 4.6, 4.7 and 4.8 GHz at 1.3V with no issues.
So the results indicate that in a standard all core manual clock setting, I would be limited to 4.4GHz without increasing the voltage. However, since the Crosshair VII allows the CCDs to be clocked independently in the UEFI I can make some solid gains without increasing voltage. There are obviously a few cores in CCD1 which just aren't all that efficient. With CCD0 I now have 8 cores (16 threads) that are running at 4.8 GHz, which is only 100 MHz below the stated boost clock of 4.9 GHz for the processor on a single core.
So with CCD specific clock speeds I was able to gain a considerable amount of performance in gaming workloads (that rarely use more than 8 cores) and also a bit in multicore as well. Something to consider for individuals that are have lower performing samples and don't want to just throw voltage at the problem.
