11 ms·

I'm a little shocked at how so few people outside of the column store industry are aware of the impact SIMD and compressed bitmap indexes has had. If in doubt,

I'm a little shocked at how so few people outside of the column store industry are aware of the impact SIMD and compressed bitmap indexes has had. If in doubt, search on "daniel lemire" "simd".

rurban 6y ago
I doubt you will find much AVX512 code with Daniel. The costs are just too high to justify it. Downclocking because of overheating and insane waitcycles to get out of this mess which is AVX512, plus poor HW support (only on insecure Intel CPU's which need to be probed at runtime) makes it an esoteric extension. AVX256 is good enough for most purposes and is not that insane.
- sradman 6y ago
  Search "daniel lemire" "AVX-512". Vectorizing random number generators for greater speed: PCG and xorshift128+ (AVX-512 edition) [1]. AVX-512: when and how to use these new instructions [2]. The dangers of AVX-512 throttling: a 3% impact [3]. [1] https://lemire.me/blog/2018/06/07/vectorizing-random-number-generators-for-greater-speed-pcg-and-xorshift128-avx-512-edition/ https://lemire.me/blog/2018/06/07/vectorizing-random-number-... [2] https://lemire.me/blog/2018/09/07/avx-512-when-and-how-to-use-these-new-instructions/ https://lemire.me/blog/2018/09/07/avx-512-when-and-how-to-us... [3] https://lemire.me/blog/2018/08/15/the-dangers-of-avx-512-throttling-a-3-impact/ https://lemire.me/blog/2018/08/15/the-dangers-of-avx-512-thr...
- paulmd 6y ago
  > Downclocking because of overheating and insane waitcycles to get out of this mess which is AVX512, plus poor HW support (only on insecure Intel CPU's which need to be probed at runtime) makes it an esoteric extension. AVX256 is good enough for most purposes and is not that insane. > Lack of adoption of AVX-512 would be Intel's own issue. If they wanted it to succeed, they would try to put it on more chips than just the few lines you mentioned. As it currently stands, Intel is setting up AVX-512 to fail by only putting it in so few products. It will be fun to revisit these sorts of comments next year when AMD releases Zen4 based processors that support it. Like with raytracing, I suspect people's tune will change when the red man does it instead of the blue man or green man ;) (it may even be in Zen3, that "50% improved FP performance" sure is interesting if you figure that's about what you would get from ideal-case speedups of AVX-512... I suspect they are moving to dual 256b units and at that point you might as well make them gangable into a 512b unit. If they don't go there with Zen3 it is an obvious low-hanging fruit for Zen4.)

I'm a little shocked at how so few people outside of the column store industry are aware of the impact SIMD and compressed bitmap indexes has had. If in doubt, search on "daniel lemire" "simd".

rurban 6y ago

I doubt you will find much AVX512 code with Daniel. The costs are just too high to justify it. Downclocking because of overheating and insane waitcycles to get out of this mess which is AVX512, plus poor HW support (only on insecure Intel CPU's which need to be probed at runtime) makes it an esoteric extension. AVX256 is good enough for most purposes and is not that insane.

sradman 6y ago
Search "daniel lemire" "AVX-512". Vectorizing random number generators for greater speed: PCG and xorshift128+ (AVX-512 edition) [1]. AVX-512: when and how to use these new instructions [2]. The dangers of AVX-512 throttling: a 3% impact [3]. [1] https://lemire.me/blog/2018/06/07/vectorizing-random-number-generators-for-greater-speed-pcg-and-xorshift128-avx-512-edition/ https://lemire.me/blog/2018/06/07/vectorizing-random-number-... [2] https://lemire.me/blog/2018/09/07/avx-512-when-and-how-to-use-these-new-instructions/ https://lemire.me/blog/2018/09/07/avx-512-when-and-how-to-us... [3] https://lemire.me/blog/2018/08/15/the-dangers-of-avx-512-throttling-a-3-impact/ https://lemire.me/blog/2018/08/15/the-dangers-of-avx-512-thr...
paulmd 6y ago
> Downclocking because of overheating and insane waitcycles to get out of this mess which is AVX512, plus poor HW support (only on insecure Intel CPU's which need to be probed at runtime) makes it an esoteric extension. AVX256 is good enough for most purposes and is not that insane. > Lack of adoption of AVX-512 would be Intel's own issue. If they wanted it to succeed, they would try to put it on more chips than just the few lines you mentioned. As it currently stands, Intel is setting up AVX-512 to fail by only putting it in so few products. It will be fun to revisit these sorts of comments next year when AMD releases Zen4 based processors that support it. Like with raytracing, I suspect people's tune will change when the red man does it instead of the blue man or green man ;) (it may even be in Zen3, that "50% improved FP performance" sure is interesting if you figure that's about what you would get from ideal-case speedups of AVX-512... I suspect they are moving to dual 256b units and at that point you might as well make them gangable into a 512b unit. If they don't go there with Zen3 it is an obvious low-hanging fruit for Zen4.)