By: Axel (no.delete@this.spam.org), January 20, 2011 11:33 am
Room: Moderated Discussions
Mark Roulo (nothanks@xxx.com) on 1/16/11 wrote:
---------------------------
>Brett (ggtgp@yahoo.com) on 1/15/11 wrote:
>---------------------------
>>EduardoS (no@spam.com) on 1/15/11 wrote:
>>99% of AVX is in Bulldozer
>>
Furthermore, I once red that even heavy FPU code only uses 20-30% of all FPU pipes, maybe Bulldozer can schedule even 256AVX instructions in a clever way and achieve higher throughput. As long as there are not two 256bit instruction every clock cycle this should be possible.
---------------------------
>Brett (ggtgp@yahoo.com) on 1/15/11 wrote:
>---------------------------
>>EduardoS (no@spam.com) on 1/15/11 wrote:
>>99% of AVX is in Bulldozer
>>
Furthermore, I once red that even heavy FPU code only uses 20-30% of all FPU pipes, maybe Bulldozer can schedule even 256AVX instructions in a clever way and achieve higher throughput. As long as there are not two 256bit instruction every clock cycle this should be possible.
