It's unlikely that anything under a 14nm process node is used for the CPU, if that.
IBM has deep knowledge in high-availability systems, but it's unlikely that you could call these more advanced than a modern 5nm Ryzen server.
Performance between junior basic code and heavily optimized complex queries even on just Oracle can be 1000x easily, just ask any data warehouse guy (seen it few times myself even though I don't do warehouses). You can maybe add another 0 for AS-400 there in extreme cases.
Yes if you end up doing things wrong and expect that CPU to do some heavy math then yes this will be dusted quickly. Otherwise, not so much.
But this goes against most modern 'SV principles' so I don't expect much love from younger generations here. Business love that though, and secretly wish all IT could be as reliable and predictable as them although that ship has sailed long time ago. One of those 'they don't do them like good old times anymore'.
You might also find it interesting to consider that the CPU was likely made by Global Foundries (assuming that their arrangement with IBM is still in place).
I am still trying to get my coworkers excited about VMS on x86; our VAX users are completely disinterested.
I have my own skunkworks VMS 1.0 on SimH, from these instructions:
Naturally NGEN has the caveat of only being good enough for faster startup, and requiring strong binaries (aka signed dlls), which meant not everyone adopted it.
Windows 8.x adopted Bartok from Singularity, where applications would be precompiled on the store, and linked on device at instalation time.
Windows 10 moved full AOT compilation into the store when downloading into specific devices.
Note that full AOT on Android is only on versions 5 and 6, starting with version 7 onwards is a mix of interpretation, JIT and AOT, which is nonetheless quite cool.
Also there are several optimizations that are only possible after having PGO data, and that was lost on AOT at installation time.
We had massive queries that ran in seconds that HANA with millions blown into the consultants just couldn’t complete in time (nightly billing runs).
The solution in the end was that a lot of complexity and decades of cleverness in adjusting to specific patterns in customer behavior were thrown out and replaced by “the consultant will just adjust that every week by hand”.
Also - I was soooo much faster with a num pad and the terminal than clicking around the SAP GUI. Ultimately part of why I left there.
/no experience with SAP but I’ve never heard a successful story implementing it ever
OP isn't saying the SAP implementation wasn't successful so, just expensive (no surprise), and somethings didn't work as begore (no surprise neither).
It also has a ton of RAS features.[0]
ECC beyond SECDED, processor instruction retry, CRC on all fabrics part of which can fail and the system will degrade to half-bandwidth. Hot/Swap and/or Redudant everything to include the LCD panel on the OP Panel.
I'm actually curious what type of RAS stuff EPYC and XEON have, and hope somebody can link to the info.