@argv_minus_one@mastodon.sdf.org
Post #2429454
2026-04-30 04:43 UTC
@lispi314@udongein.xyz
That's a very extraordinary claim. It requires very extraordinary proof.
If there were a way to achieve performance on par with what we have now, yet do so without the security risk and extra circuitry required for speculative execution (and the security/stability risk of ambient-authority monolithic kernels), wouldn't we be using it already?
Replies (1)
-
@lispi314@udongein.xyz 2026-04-30 04:47
@argv_minus_one@mastodon.sdf.org No, because it performs slightly worse and isn't popular in the particular CPU architectures popular right now. Other architectures did use it in the past. extra circuitry required for speculative execution I am not familiar enough with electrical engineering and hardware design to comment on just how it's implemented physically. I consider that "slightly worse performance but actually correct behavior" is more important than "maximize meaningless benchmarks & bloat numbers as much as possible".