Without being a cpu geek, a lot of the branch prediction details go over my head, however generally a good review. I liked the detail of performance on more complex workloads where IPC can get muddy when you need more instructions.

I feel these days however, for any comparison of performance, power envelope needs to be included (I realise this is dependent on the final chip)

ARM Cortex-X925 achieves indeed a very good IPC, but it has competitive performance only in general-purpose applications that cannot benefit from using array operations (i.e. the vector instructions and registers). The results shown in the parent article for the integer tests of SPEC CPU2017 are probably representative for Cortex-X925 when running this kind of applications.

While the parent article shows AMD Zen 5 having significantly better results in floating-point SPEC CPU2017, these benchmark results are still misleading, because in properly optimized for AVX-512 applications the difference between Zen 5 and Cortex-X925 would be much greater. I have no idea how SPEC has been compiled by the author of the article, but the floating-point results are not consistent with programs optimized for Zen 5.

One disadvantage of Cortex-X925 is having narrower vector instructions and registers, which requires more instructions for the same task and it is only partially compensated by the fact that Cortex-X925 can execute up to 6 128-bit instructions per clock cycle (vs. up to 4 vector instructions per clock cycle for Intel/AMD, but which are wider, 256-bit for Intel and up to 512-bit for Zen 5). This has been shown in the parent article.

The second disadvantage of Cortex-X925 is that it has an unbalanced microarchitecture for vector operations. For decades most CPUs with good vector performance had an equal throughput for fused multiply-add operations and for loads from the L1 cache memory. This is required to ensure that the execution units are fed all the time with operands in many applications.

However, Cortex-X925 can do at most 4 loads, while it can do 6 FMAs. Because of this lower load throughput Cortex-X925 can reach the maximum FMA throughput only much less frequently than the AMD or Intel CPUs. This is compounded by the fact that achieving better FMA to load ratios requires more storage space in the architectural vector registers, and Cortex-X925 is also disadvantaged for this, by having 4-time smaller vector registers than Zen 5.