Application Legacy Mode
If you're wondering what half legacy mode does for applications, here is a quick look at that. Firstly memory bandwidth is increased massively and this is because we have two die using NUMA opposed to the 2950X which uses UMA by default. If we changed the 2950X to the NUMA mode we would see the same 75.9 GB/s.
Just to confirm the increased bandwidth here are the AIDA64 results, as you can see peak memory throughput now exceeds 88 GB/s. Please not this figure is higher than the 76 GB/s reported by SiSoftware because here we are reporting peak throughput rather than sustained throughput.
Here we see that the inherent issues with the 2990WX is solved, and that issue is of course bandwidth per core. With more memory bandwidth overall and half as many cores, we see a little over twice the bandwidth now available to each core.
Now although the bandwidth per core is improved, because we have half as many active cores the 2990WX is going to be much slower in workloads where it was previously faster than the 2950X and we see that when testing with Blender.
The same is also true for POVray, basically the 2990XW is delivering 2950X like performance here.
However where the 2990WX was slower than the 2950X we are now seeing comparable performance, as shown here when retesting with HandBrake.
Another example is the Adobe Premiere Pro CC warp stabilizer workload, the 2990WX in half legacy mode doesn't quite match the 2950X but the results are certainly much closer now. So that's how half legacy mode works, it's not a practical or useful solution in my option, but it's there as an option if you get desperate.