Hacker News Reader

There's a lot of convenient new instructions in avx512 that can work on 128/256-bit vectors. I'm guessing that those wouldn't throttle more than regular avx.

I'm guessing you're right.

It depends on your workload and which Xeon chips you choose — some of them don’t down-clock too much for AVX512.

how often? what suite of code did you benchmark? Why wouldn't it be good for gaming? Put the AVX-512 workload on one core for just that. Or batch the AVX-512 work together so it isn't short bursts.

Nobody is even thinking about leveraging this stuff without understanding things like this anyway.

Show me a game that uses the actual zmm register operations. The big thing is your game has to run at good frame rates without it, so there's little incentive to optimize something to use them, and if you do the speedup is partially mitigated by the throttling anyway. A lot of work for no gain.

In not saying you can't find something in a game to speedup with avx512, but that you wouldn't want to do that in the first place.