Because the performance IS dependent of the workload. Almost, for 21 century hardware.
El jue, 17 abr 2025, 18:33, Seymour J Metz <[email protected]> escribió: > None of which gives you a measure of performance independent of the > specific workload. > > -- > Shmuel (Seymour J.) Metz > http://mason.gmu.edu/~smetz3 > עַם יִשְׂרָאֵל חַי > נֵ֣צַח יִשְׂרָאֵ֔ל לֹ֥א יְשַׁקֵּ֖ר > > > > ________________________________________ > From: IBM Mainframe Discussion List on behalf of salva > Sent: Thursday, April 17, 2025 9:45 AM > To: [email protected] > Subject: Re: z17 > > > External Message: Use Caution > > > "In practice, there was never a way to measure the performance of a > computer, only its performance on specific benchmarks." > > Since z10 CPU MF (measurement facility) provide counters for instruction > executions, cycles, time, cache misses, tlb... > > El jue, 17 abr 2025, 14:18, Seymour J Metz <[email protected]> escribió: > > > How rapidly can a z17 dispatch instructions? That's a significant factor > > in estimating instructions. > > > > Long ago in a galaxy far away, IBM used to include timing formulae in > > functional specifications. The was a clear trend that when you looked at > a > > larger or newer machine, the formulae got more complicated. I shudder to > > think of how complicated timing formulae for the z17 would. > > > > In practice, there was never a way to measure the performance of a > > computer, only its performance on specific benchmarks. > > > > -- > > Shmuel (Seymour J.) Metz > > http://mason.gmu.edu/~smetz3 > > עַם יִשְׂרָאֵל חַי > > נֵ֣צַח יִשְׂרָאֵ֔ל לֹ֥א יְשַׁקֵּ֖ר > > > > > > > > ________________________________________ > > From: IBM Mainframe Discussion List on behalf of Abe Kornelis > > Sent: Thursday, April 17, 2025 1:24 AM > > To: [email protected] > > Subject: Re: z17 > > > > > > External Message: Use Caution > > > > > > All, > > > > I have been surprised nobody has mentioned the parallel execution. > > From the announcement up to [decimal] ten instructions can be in > > progress simultaneously, > > and up to four instructions can be fetched and decoded in parallel. > > > > Doing ten instructions simultaneously would imply an effective speed of > > 55GHz. > > > > Obviously, that limit of ten will rarely be reached, but only one at a > > time is > > probably rather rare as well these days. > > > > I'd have liked to see some metrics from IBM testing, if only to please > > the taste buds :-) > > > > Kind regards & Happy programming, > > Abe Kornelis > > ========== > > > > > > On 17/04/2025 06:27, Timothy Sipples wrote: > > > Steve Beaver wrote: > > >> We all know the z17 has been announced. > > >> What I am disappointed in is the CP's have not gone faster than 5.5 > Ghz. > > >> I know the z17 is an evolution, but why have they not gotten faster? > > > The main processors’ clock speed has in fact increased from 5.2 GHz in > > the IBM z16 to 5.5 GHz in the IBM z17. > > > > > > Please note that these systems are designed to operate continuously for > > their entire lives at these extremely high clock speeds provided your > data > > center doesn't drift outside typical industry standard environmental > > limits. (Outside environmental limits the system will defend itself if/as > > necessary, first by slowing down then by shutting down. And the system > will > > try to warn operators that's what's happening — and notify IBM if you > have > > Remote Support Facility a.k.a. "Call Home" enabled.) It's not some "turbo > > mode" clock speed that you only get for 5 seconds after you start using > the > > processors. > > > > > > As others have astutely pointed out, the clock speed is only one aspect > > of processor performance and throughput. Many other design factors are at > > least as important. However, if you'd like to applaud the amazing > designers > > and engineers who figured out how to increase the lofty clock speed to be > > even loftier, I'm sure they'd appreciate it. > > > > > > ————— > > > Timothy Sipples > > > Senior Architect > > > Digital Assets, Industry Solutions, and Cybersecurity > > > IBM Z/LinuxONE, Asia-Pacific > > > [email protected] > > > > > > > > > > > > ---------------------------------------------------------------------- > > > For IBM-MAIN subscribe / signoff / archive access instructions, > > > send email to [email protected] with the message: INFO IBM-MAIN > > > > ---------------------------------------------------------------------- > > For IBM-MAIN subscribe / signoff / archive access instructions, > > send email to [email protected] with the message: INFO IBM-MAIN > > > > > > > > > > ---------------------------------------------------------------------- > > For IBM-MAIN subscribe / signoff / archive access instructions, > > send email to [email protected] with the message: INFO IBM-MAIN > > > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to [email protected] with the message: INFO IBM-MAIN > > > > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to [email protected] with the message: INFO IBM-MAIN > ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
