Just another proof that the fast benchmarks are not a real measurement of power...
I would say say it's proof on Corona rather than the CPUs. Not sure if the Infinity-Fabric is "memory starved" in Corona as the Reddit drones about but it does scale less favorably than Intel.
It does scale properly in Vray though, so I would say there is definitely room for improvement.
CineBench isn't synthetic benchmark either in true sense I would say it's way more valid benchmark than the odd multi-process Corona Benchmark 1.3 which hardly starts with full-cores engagement.
Interesting thing is that when my 2698v4 Node and i9 7980XE are set for same CB 4400 score, i9 is still 6-10perc. faster in actual rendering in both Vray and Corona, but surprisingly, not in the 1.3 benchmark. I also attribute that to fast memory access. (and the AVX shanenigans)
I have currently much higher faith in overclocking 2990WX than my 7980XE which despite delidding can't transfer the heat properly through the small die, while the massive multi-die 2990WX seems like breeze to every reviewer.
Also, my stupid ROG Strix x299 has fucking 8 VRM phases total, while MSI MEG x399 has massive 19 phases (16 for the cpu). I would need to get another board for Intel, but even the best overclocking board has 12 phases for CPU, which isn't enough for the 18c to reach past 4.0 Ghz AVX (4.5 non AVX).
Not arguing that dual-platz won't reach better result, no reason why it shouldn't. But that's the only thing it can do :- ). I kind of feel I should grab pair of them as well but I bought so much crap in past weeks I need to get my hardware&gadgets buying habits in check.
