Intel leads in x86 performance and comparing performance of System z CPU to x86 is not exactly trivial task as zArchitecture is "uber-CISC" with instructions like "give me SHA256 of this part of memory" that also typically runs on 4+GHz clocks (and the "CPU book" has TDP of several kW). But main part of the performance advantage (be it perceived or real) is incredible memory and IO bandwith combined with large caches and hardware offload for essentially anything that remotely looks like IO.
On the other hand, you can get rack-mount x86 server with several TB of RAM for fraction of price of mainframe which is exactly the hardware for new applications that would otherwise be best served by mainframe.
> On the other hand, you can get rack-mount x86 server with several TB of RAM for fraction of price of mainframe which is exactly the hardware for new applications that would otherwise be best served by mainframe.
I expect the largest one of these is still a fraction the size of the largest mainframe, at least for processor power (e.g. 224 cores[0] compared to 1700[1]). Maximum memory, though, is 32TB, which isn't exactly huge compared to 12TB [0] or 24TB [2].
From my point of view for typical modern bussiness workloads that necessiate large machines (ie. SQL RDBMS or something similar) only thing that you really care about is memory size. On the other hand many "legacy" mainframe workloads are significantly more CPU-bound because the software culture is simply different and brute force is often the way to go.
> From my point of view for typical modern bussiness workloads that necessiate large machines (ie. SQL RDBMS or something similar) only thing that you really care about is memory size.
That's my general impression, as well, but I have little enough (or inadequately broad) direct experience for that impression to be a strong one.
I've also seen, second-hand (i.e. benchmarks, so not quite real-world) that cache and memory latencies (including inter-CPU/NUMA) can significantly affect OLTP performance.
If that weren't true in practice, then the sheer bandwidth one can deliver from enough SSDs over PCIe rivaling a CPU's memory bandwidth would mean main memory size is much less relevant.
In theory you can hot-swap CPUs on x86 (with slightly hillarious fact that primary use of the Linux's CPU hotswap support is suspend to disk on laptops) but hardware that supports it and it really works is rare to non-existent.
