This is a question that everyone is curious about, how the X3D SKUs with 2 CCDs fare versus the non-3D SKUs with 2 CCDs. After all, now you will find yourself in the situation where the 7900X3D / 7950X3D gets a high frequency CCD and a CCD with the larger cache, but a frequency limitation, while the 7900X/7950X does not have the advantage of the larger cache, but also they do not have the disadvantage of a frequency-limited CCD.
So what you’re likely to see is that the X3D models dramatically increase in workloads that really benefit from the added 64MB of cache, but those workloads that don’t benefit from more cache, but rely on frequency, show no improvement or perhaps even regression. After all, you’ve seen it with the 5800X3D versus the 5800X.
Another unknown (until the reviews come in) is how well the Windows (or Linux) scheduler will handle this. Until now, AMD has only had SKUs where each CCD was equipped with 3D V-Cache when present on the SKU, so that each core always had the same amount of cache available.
Now for the first time we will see SKUs where this is not the case, so not all cores have the same amount of cache and the scheduler then has to choose, do I run this process on the slowest core with a large cache, or do I run the process on the fast core with less cache. If AMD and Microsoft fail to do this, it could be, just as you saw with Intel’s E and P cores, that some workloads will show very strange performance results.
