> According to the benchmarks in the article, many are also doing better at low-cardinality data. VQSort handles low cardinality quite well. Unfortunately only

> According to the benchmarks in the article, many are also doing better at low-cardinality data. VQSort handles low cardinality quite well. Unfortunately only a few benchmarks in the article (not including that one) were re-run after our performance bug fix. > it's very strange to compare to std::sort which does make some attempt to be adaptive. We chose std::sort mainly because it seems to be widely used. > But we all know why you're doing that. I suppose you'll keep your fingers in your ears regarding 32-bit radix sort huh, I'm curious what's the background or context for this remark? FYI I published a fast 32-bit radix sort in 2010: https://arxiv.org/abs/1008.2849 https://arxiv.org/abs/1008.2849 The trend has been towards less and less memory bandwidth per core, and we are often seeing servers bottlenecked on that.