Intel Gets AVX-512 Support Going In GCC

[phoronix]

Phoronix: Intel Gets AVX-512 Support Going In GCC

The forthcoming release of the GCC 4.9 compiler will feature support for Advanced Vector Extensions 512 (AVX-512) although this instruction set extension won't be appearing on Intel CPUs for a while...

http://www.phoronix.com/vr.php?view=MTU5OTE

[uid313]

AMD

It would be really fun if AMD hurried up and released a CPU with AVX-512 instructions before Intel launches theirs.

[bridgman]

It would be really fun if AMD hurried up and released a CPU with AVX-512 instructions before Intel launches theirs.

I guess... we already have 512- and 1024-bit SIMD instructions in the GPUs though.

[carewolf]

It would be really fun if AMD hurried up and released a CPU with AVX-512 instructions before Intel launches theirs.

I would be okay if they extended XOP to 256-bit and extended their internal SIMD ALU to 2x256-bit. AVX-512 is a rather interesting piece of nonsense. It's only benefit really comes down to cases where instruction decoding is the bottle-neck and not internal processing power or memory I/O, and when is that? It does close a lot of holes with new instructions, but the same holes has already been closed by AMD using XOP.
AMD might be able to do better with a better GPGPU integration. But considering their reluctance to share any specifics on mantle and they haven't talked about anything else, I am not even sure they have anything up their sleeves.

[wizard69]

It would be really fun if AMD hurried up and released a CPU with AVX-512 instructions before Intel launches theirs.

AMD has a better solution in GPU integration. In fact I have to wonder why Intel is going this route as it seems to duplicate hardware already in the GPU. Of course GPUs don't have a huge instruction set to work with but it would make more sense to extend the GPU instruction set than to go this route.

Time will tell but I really beleive AMD is on the right track. Basically they are all in with HSA whereas Intel is very very wishy wash when it comes to GPU compute.

[Marc Driftmeyer]

AMD has a better solution in GPU integration. In fact I have to wonder why Intel is going this route as it seems to duplicate hardware already in the GPU. Of course GPUs don't have a huge instruction set to work with but it would make more sense to extend the GPU instruction set than to go this route.

Time will tell but I really beleive AMD is on the right track. Basically they are all in with HSA whereas Intel is very very wishy wash when it comes to GPU compute.

Phi Co-Processor product.

http://wccftech.com/xeon-phi-hpc-codenamed-knights-corner-revealed/