Advertainment
Tech NewsInternetNewsTechnology

Intel Releases Ultra-Fast AVX-512 Sorting Library, Numpy Supports 10-17x Faster Sorting

Intel has released an open-source C++ header file library for SIMD-based high-speed sorting. It was initially focused on providing a super-fast AVX-512 quicksort implementation. As of today, this code has been merged into Numpy with a speedup of 10x to 17x.

Advertisement

Late last year, Intel quietly released x86-simd-sort from its GitHub account. This is a C++ header file library for high-performance SIMD sorting, but currently only focused on implementing AVX-512 quicksort.

The x86-simd-sort project is rarely mentioned but has achieved amazing results through the now widely used Numpy open-source project.

Newly merged into Numpy, PR 22315 is for vectorizing quicksort for 16-bit and 64-bit data types using AVX-512. On the Intel Tigerlake system, sorting of 16-bit int types is 17 times faster, sorting of 64-bit float types is nearly 10 times faster for random arrays, and 32-bit data types are 12-13 times faster. This Numpy modification was done by Intel engineer Raghuveer Devulapalli and utilizes code from x86-simd-sort.

The 10-17x speedup in sorting with AVX-512 is pretty amazing, especially considering the better AVX-512 efficiency on recent-generation Intel CPUs. With the latest Xeon Scalable processors, the AVX-512 is in rather a good shape without causing any significant CPU downclocking. Meanwhile, on AMD’s side, Zen 4 processors from the Ryzen 7000 series up to the 4th Gen EPYC server processors finally have AVX-512 support.

Advertisement
Advertisement

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button