Intel Publishes Sorting Library Powered by AVX-512, Offers 10-17x Speed Up

[R-T-B]

Didn't they just remove AVX512 from their consumer CPUs?

Yeah. That's sadly why it HASN'T seen wider adoption. It's segmentation hell.

Alder processors with the old Intel logo have full support, like Xeon.

Pretty sure the alder lake implementation was partial as well. Maybe more complete than AMD's, but still partial.

 

[Assimilator]

Yeah. That's sadly why it HASN'T seen wider adoption. It's segmentation hell.

IMO they removed it simply because it would've caused the nuclear reactors that are their CPUs to meltdown. Maybe we'll see a return of AVX-512 in the consumer space with their next node shrink.

 

[AnotherReader]

Let's hope that this code is vendor neutral. AVX-512 is a better vector ISA than SSE or AVX/AVX2, and for implementations like Zen 4, whose peak FLOPS for AVX and AVX-512 are identical, AVX-512 can be more power efficient due to requiring less power in decode and scheduling compared to its 256 bit alternative.

 

[mrnagant]

Didn't AMD refer to their AVX-512 as double pumping? Just using two 256-bit halves? Phoronix shows that AMDs AVX-512 is 60% faster than AVX2 while using similar power and running at basically identical clock speeds.