Angels on the head of a pin, guys. Running x86 at 100% runs into various architectural bottlenecks to which IBM Z (generally) doesn't fall prey; we know that. Whether those are necessary or OS-caused doesn't matter: they are there.
"A z17 offers n* the performance of an Intel 12345" is also semi-meaningless, unless we're comparing the same workload. Until you RUN a Google search on IBM Z, you don't know whether it will do better, worse, or the same. A given search is a relatively small operation, so 100,000 searches MIGHT do better on n,nnn separate x86 servers Just Because; or might benefit from shared memory etc. on IBM Z. Again, without testing, we don't know. This is Performance 101, eh? -----Original Message----- From: IBM Mainframe Assembler List <ASSEMBLER-LIST@LISTSERV.UGA.EDU> On Behalf Of Jon Perryman Sent: Sunday, August 24, 2025 3:14 PM To: ASSEMBLER-LIST@LISTSERV.UGA.EDU Subject: Re: Telum and SpyreWAS: Vector instruction performance On Sun, 24 Aug 2025 16:38:51 +0100, Martin Ward <mar...@gkc.org.uk> wrote: >So if Google switched to using z17's they would be running at 100% and >therefore using the full 35kW per machine. Got it, thanks. Ok, you don't understand power curves nor understand facts. I asked Gemini questions about power consumption which says running at 100% saves 37% kW instead of running at 75%. Telum power curve is not available. Without changing hardware, Google would save around 10-15% kw plus the 16% for idle consumption of the removed servers but only mainframers understand how to run at 100%. Can you tell us for the 100th time, was it 35 kW. Got it, thanks Dr. Martin. > >> z17 is a Ferrari, not a moped >For some applications, 10 mopeds beat one Ferrari :-) Your point? Google uses servers and z17 significantly outperforms those servers but again, you show no facts to prove either way. Got it, thanks.
