Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Depends on the application, of course. Haswell introduced some really cool new instructions (both the Bit Manipulation set as well as the Software Transactional Memory set). Don't have any numbers off-hand, but these can definitely speed up certain workloads.


Apparently in specific workloads the move from AVX -> AVX2 was very substantial, though I can't seem to find the article that showed this.


The transition is more like from SSE2 -> AVX2, in integer workloads. AVX2 adds 256-bit wide integer instructions and gather-instruction. Gather does multiple loads in one instruction.

AVX was just about floating point.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: