I just ran the Aida CPU benches in the review (Julia, SinJulia, VP8 and Mandel) and I'm impressed by the Ryzen's performance. And more than a little upset. In all but the Mandel my Skylake at 4.7 GHz got its butt kicked-badly. Are the Ryzens that much better??? It looks (to me) like AMD is back, with a vengeance.
My 6700k at stock (4.0 GHz with 4.2 one core turbo) beats the Ryzens apart from SinJulia.
For comparison I got:
vp8 7298
julia 33973
mandel 18291
sinjulia 4816
For SinJulia, if I look at the example results given in Aida64, my score is within tolerance of the 6700k they have there, but they're suggesting a 3770k at 3.5 GHz base 3.9 turbo is faster?
What does each do?
https://www.aida64.com/products/features/benchmarking
In short:
VP8 - who knows
Julia - 32-bit (single precision)
Mandel - 64-bit (double precision)
SinJulia - 80-bit (old x87)
Now, my concern here is for Mandel they list "x87, SSE2, AVX, AVX2, FMA, and FMA4" as instructions used. I don't see FMA3 in there. The thing is, FMA3 was introduced with AVX2 in Haswell, and I'm bad at using them as if they're the same thing, but I'm not sure they technically are. If FMA4 is of benefit, FMA3 should be too. If FMA3 isn't used, it'll be missing out on a significant boost for Haswell or newer, and Ryzen can also try to take advantage of it. It supports it, but I'm not convinced it is effective. Similar story with FMA4 on older AMD processors.
I've not really looked into x87 as there are few things that really need it now. But in what little I have done, is yet another prime number finding software called genefer. This is NOT based on the Prime95 codebase, but does have some similarities in that it does FFTs also. For a particular type of number, due to its size the extra precision of x87 was required over 64-bit, yet it used small FFTs. This should allow running out of cache, so isn't ram limited, and in that I saw a similar gap in IPC between Ryzen and Intel as for 64-bit. So SinJulia is behaving nothing like that. Further looking at the Aida64 example results, the 4770 and even 2600 (both non-K) are not far behind the 6700k, despite their lower clocks and older architectures. The only commonality I can see is they're all DDR3 systems. I'm wondering if there is some other feature of the CPU that helps these results more than the cores, and I'm thinking ram access in some way.