I think that is logically the wrong way of thinking about it. The two companies have different lines of CPUs with different types of architectures. Because the test has changed for the latest DAWbench it’s actually not really comparable because the difference in the test can change the performance of the CPU.
To put it differently: It’s possible that both the Intel benefitted from the new test and AMD’s new 5950x is better than the 3950x, but that the former made a bigger difference than the latter. So because the test wasn’t literally the same we simply can’t know.
But anyway, I took the highest performing Intel CPU I found on Scan’s and Vin’s test (“tafkat”), the 10920x, and compared that to the 3950x/5950x.
In the VI portion at a 64 buffer the delta is 13% in favor of the 3950x versus the 10920x. The 5950x at that same buffer size was 25% ahead, which is a 100% improvement for that generation at that buffer size. For the same test but at 128 samples the lead is only 4% for the 3950x, and about 10% for the 5950x. So this time over 100%. At 512 samples it’s 31% vs 7%, which is odd. Almost as if a different bottleneck was reached in the test system.
For the DSP part of the test (% lead 3950x/5950x over 10920x);
64 buffer - 55% / 57%
128 buffer - 54% / 54%
512 buffer - 26% / 56%
Basically there were zero gains on the DSP test at low/medium buffer and 100% at a “high” buffer, and a whopping 100% (roughly) on the VI test at low/medium buffers and a regression at high buffers which was possibly a bottleneck…
So taking those results at face value: for anyone using VIs and needing a very low latency (64 samples) the difference is significant.
But again though, it’s possible that we’re seeing the results of changes in the test and not changes in CPU capability.