By: Anon (no.delete@this.thanks.com), July 12, 2013 10:26 am
Room: Moderated Discussions
A Cortex-A7 supports the entire ARMv7 architecture including NEON in 0.5mm^2 per-core on 28HPM. For sure, it is not as fast as a Krait or Cortex-A15, but the entire ISA is supported in a very small area. The reason for supporting NEON on an A5/A7 class processor is largely to ensure software compatibility across the entire mobile processor family and reduce the fragmentation that Linus is railing against.
I'm certain that Intel could build a cut-down AVX in Atom if they wished too for a lot less than 6mm^2 per core. It wouldn't have the performance of Haswell, but it depends on whether they value software compatibility more than performance.
If it did require 6mm^2 per core (and I don't think it would) it would create a barrier for Intel in a large part of the smartphone market where the entire processor is only 5-8mm^2 (A5, A7, A9) and every fraction of a mm^2 counts. While Intel don't have to play quite the same game on silicon area with their own foundry, they can't entirely escape the laws of yield and chip costs.
I'm certain that Intel could build a cut-down AVX in Atom if they wished too for a lot less than 6mm^2 per core. It wouldn't have the performance of Haswell, but it depends on whether they value software compatibility more than performance.
If it did require 6mm^2 per core (and I don't think it would) it would create a barrier for Intel in a large part of the smartphone market where the entire processor is only 5-8mm^2 (A5, A7, A9) and every fraction of a mm^2 counts. While Intel don't have to play quite the same game on silicon area with their own foundry, they can't entirely escape the laws of yield and chip costs.
