Beware that some of the Geekbench number are the result of them suddenly gaining support for streaming SVE and SME just when Apple implements it.
I'm not doubting the number represent real peak throughput on M4. There is just a taste to the timing lining up so well. Also they don't take advantage of fully SVE2 capable ARM cores to compare how much a full SVE2 implementation would help especially at accelerating more algorithms than those that neatly map to streaming SVE and SME.
The single core performance gains of M4 variants over their predecessors are distorted, because the streaming SVE and SME are apparently implemented by combining what used to be them AMX units of four cores.
I'm not doubting the number represent real peak throughput on M4. There is just a taste to the timing lining up so well. Also they don't take advantage of fully SVE2 capable ARM cores to compare how much a full SVE2 implementation would help especially at accelerating more algorithms than those that neatly map to streaming SVE and SME.
The single core performance gains of M4 variants over their predecessors are distorted, because the streaming SVE and SME are apparently implemented by combining what used to be them AMX units of four cores.