Do you know when this extension will release for intel iCore ?
I am working on a 3D engine with 3D matrix calculation on full CPU since 2014 and I actually use the AVX with its 256 bit registers.
And the performance is around 5-15 fps for 1 million vertex.
I heard about the new update of AVX instructions that operate on 512-bit registers since 2014 and now I see that it 's release on Xeon CPU.
But because I am not working with Xeon CPU, I wait it on Intel iCore CPU.
I haven't work on this project since 2015 because 5-15 fps is too tiny to make something good with the engine and I wait the AVX-512 to get back on this project.
- Intel® Advanced Vector Extensions (Intel® AVX)
- Intel® Streaming SIMD Extensions
- Parallel Computing
About performance, yes it will be the case because I build my project with the intel assembly language and comparing to 64 bit register operation, the use of 256-bit register is much more performant.
That's why I want more register size, because I am not using a compiler.
And beside the performance fact that can be resolve by using a compiler, it's the evolution fact, I personally found interesting and amazing this add of new register size operation than classic 32-64 bit size.
I am not seeking for performance but for technology evolution, that's why I also try to make a game engine with 3D calculations only on CPU.
Basically, all the info that everyone has is that you can google over the internet. Intel typically doesn't comment on future products. In current client CPUs AVX-512 is (reportedly) available in Cannonlake, but unfortunately it is only available in mobile sector and even there it's pretty hard to find. A more wide adoption is expected to come with Ice Lake sometime in the second half of 2019, according to the latest rumors. Ice Lake is supposed to come out in the desktop sector, as well as mobile and server/workstation, though the time frames in the different sectors will probably be different.
You should take into account that the benefit from AVX-512 in Skylake-X turned out to be controversial in some applications. Because the CPU produces more heat when working with larger vectors, it tends to reduce CPU clock more aggressively, which often results in reduced overall performance of the system. It remains to be seen whether this problem is solved in Cannonlake or Ice Lake.
Finally I was wrong about the argument for avx-512.
My Engine made on C have 60 fps with 128-bit register from SSE, but if I use the classic registers, my Engine is at 20 fps even using x64.
It's not a normal 3D engine, but a voxel engine built only on CPU: https://www.codeproject.com/Articles/1247960/Learn-D-Math-made-by-the-GPU-by-Creating-a-D-Engine.
I haven't made the calculation using 256-bit because I haven't time for this, but I hope I will have.
So logically if I use 512-bit, my engine will be more powerful and I will be able to display more voxel than 1 millions actually with decent FPS.