Post #2429458

2026-04-30 04:52 UTC

@lispi314@udongein.xyz Server people don't care what CPU architecture is popular. They'll compile their code for whatever hardware is best. So if the performance of such a machine were only slightly worse, then I really can't imagine why people wouldn't be using it by now. They've already accepted performance penalties for Meltdown mitigation, and those penalties were not slight.

Replies (1)

@lispi314@udongein.xyz 2026-04-30 04:55

@argv_minus_one@mastodon.sdf.org Server people don't care what CPU architecture is popular. They'll compile their code for whatever hardware is best. What's best (performing) is typically the thing that's botching and cutting corners (right up until it turns out those corners matter and it's not an option to ignore them). I find that unappealing. They've already accepted performance penalties for Meltdown mitigation, and those penalties were not slight. CPU manufacturers need to actually feel like implementing and commercializing it. Presumably it's all legacy spaghetti and they'd prefer to just patch it over than rip it all out and try to make something remotely equivalent & adjust it to work with their design. Some of their users would probably also get whiny about any perceived regression, even if it's to accomplish proper correctness. (Intel seems more interested in "AI" bullshit and making CPUs defective from factory.) A lot of "server people" disable the mitigations, by the way. (I expect these would be the whiny ones.)

Open ##2429459