- cross-posted to:
- [email protected]
- cross-posted to:
- [email protected]
Good grief.
Intel’s pushing AVX10 as the next-best-thing when AMD is the one with the best AVX512 implementation in practice (Zen4 support). Intel needs to get its P-cores and E-cores supporting AVX512 before anyone believes in this next thrust IMO.
I’ll be going through this myself eventually. I do think that Intel has excellent engineers, as proven by the pext and pdep instructions. And AVX512 itself even (very well designed instruction set).
But the Intel business and marketers are braindead, they don’t have any idea of how to make products or take advantage of these ISAs / advantages before AMD. Hopefully Intel’s business teams can do a better job with AVX10 than they did with AVX512.
EDIT: Looks like the biggest change is that AVX10 is going to take advantage of AVX512’s flag-registers to perform cmov / conditional moves / branchless if-statements for high performance. Not seeing much else here to wow me except confirmation that P-cores + E-cores promise to support AVX10 eventually. But this is already a mistake: 1st generation of P-cores/E-cores fail to support AVX512, let alone AVX10. It will be years before new computers are deployed and AVX512 (or AVX10) is usable.
But in theory, this seems like a decent step forward. Just gotta trust Intel to keep things going I guess?