[Info-vax] VMS x86 performance ?

[Simon Clubley]

On 2020-10-30, geze... at rlgsc.com <gezelter at rlgsc.com> wrote:
> On Friday, October 30, 2020 at 1:38:25 AM UTC-4, IanD wrote:
>> It's been a while since this topic was first aired and the response at the time was "it is too early to tell" 
>> 
>> This was quite some time ago now 
>> 
>> I don't expect benchmarks as such but I would have thought we have a general idea as to how it performs compared to Itanium or Alpha by now? 
>> 
>> Ballpark figures? 
>> 
>> 10%, 20%, 50%, Rabbit on crack? Still too early to tell? 
>> 
>> *Yes, I'm talking about a VM environment since that will probably be the normal deployment environment
> IanD,
>
> Working with preliminary cross compilers?  Early field test software?
>

The compilers are only one part of this.

Ian is very, very correct. We should be getting some initial indications
by now, especially for I/O bound workloads, where the compilers are far
less important.

For example, tests from RMS could compare Alpha/Itanium performance
with x86-64 (and associated I/O hardware) performance.

Tests using RMS could also show if there are any noticable overheads
from mapping 4 VMS modes (KESU) into 2 hardware modes (KU).

Those same tests repeated on AMD hardware would show what additional
overheads are imposed by not having PCID available.

Initial I/O performance testing would show how much faster than Itanium
(or Alpha) VMS on x86-64 is (assuming that x86-64 VMS is indeed faster
than Alpha or Itanium VMS) for I/O based workloads and would make good
publicity within the VMS community for VSI.

So yes, it makes perfect sense in VSI-land that VSI wouldn't bother
doing such tests. :-(

> I would think that it is far too early to make any comments that would have
> any validity.
>

We are _6_ _years_ into the port! How much longer are people going to
have to wait ?

Simon.

-- 
Simon Clubley, clubley at remove_me.eisner.decus.org-Earth.UFP
Walking destinations on a map are further away than they appear.