just seeing reviews of Intel 7900x seems has new AVX512 instructions and each core has 2 AVX512 FMA units at least on Core i9 there is a doubt if 7800x and 7820x has only 1FMA AVX512..
anyway seems even Knights landing had AVX512 there was no OpenCL support on this platform (correct me if wrong) and seeing also no new Intel OpenCL 2017 SDK even we are almost in H2 question is if current Intel OpenCL 2016 SDK CPU runtime has support for AVX512 and is optimized for exploiting it on this 7900x CPU or better waiting on Intel OpenCL 2017 CPU runtime for better optimizations..
For example Luxmark 3.1 benchmark using OpenCL CPU as renderer will be faster with some OpenCL 2017 runtime or is almost optimal by now?
Also some ETA on 2017 SDK reléase.. (hope has VS2017 support +CanonLake IGPU support+ OpenCL 2.2 CPU support)
The Core i9 7900x doesn't have the ER extension :(
I believe this affects its ability/performance when meeting OpenCL ulp requirements compared to GPU's
i.e. I think and might be wrong that native reciprocal square root (The one I care about) on Nvidia and AMD GPU beats out non ER AVX512
I'd be happy if someone from Intel can correct me on this....
Let me see if I can get more info on CPU implementation performance expectations, including AVX-512.
There is a new SDK coming soon, as well as a new GPU/CPU driver package. Neither has a set date yet but both should be coming in weeks instead of months.
Kaby Lake is the newest architecture covered. Latest info I have is that Visual Studio 2017 support is coming but OpenCL 2.2 is not. Please watch for upcoming announcements on new features, which will be a more authoritative list.