[HECnet] My newest toy

[Brian Schenkenberger, VAXman-]

Mark Benson <md.benson at gmail.com> writes: 

On 13 Jan 2013, at 16:04, Brian Schenkenberger, VAXman- wrote: 

...but I don't BUY systems. ;) 

You KNOW what I mean ;) 

I have.   There's quite an air volume but I would hessitate to call it
= a > toaster oven.=20 

Okay try this comparison. Stick your hand behind a normal modern Intel =
PC. Now stick it behind a rx2600/zx6000. See what I mean. It's hot in =
comparison. It might be the flow is slower and hotter in a zx6000 but =
you can see where all that electrical energy is ending up :) 

No "intel" PeeCees here as I do not do WEENDOZE.   The only intel CPU h/w,
other than the Itaniums, are in the MacBookPros.



I'm not saying Itanium is inefficient as it has a LOT of raw CPU =
horsepower per Watt but I anm skeptical as to wether it's more efficient
= than a late Alpha on a per-MHz comparison. Both CPUs get very hot in =
operation. Both have a LOT of raw power for that draw and per MHz. It's
= not like comparing a Pentium 4 and an Alpha, for example. 

Don't get me wrong.   I loved Alpha and I believe that if it was permitted
to live beyond the late term abortion that HP gave it that it would be on
par with or have exceeded the Itaniums.   Of course, that's fodder for one
of those time-travel into the past altered timeline theories where Brian
and Peter wind up with a chalkboard in their living room.



I'm not in Florida. ;) 

No, I know, You're in NY/NJ, I know what 'Sandy' was, 2 of my best =
friends live in Hoboken. I was using Florida as a comparison as a 'it's
= hot there most of the year' place. If you'd prefer I use Texas or =
Louisiana I can do ;) 

NJ shore area.



With regards to performance, the VUPS.COM benchmark (which is probably =
wildly inaccurate on fast non-VAX hardware) considers my zx6000 (472 =
VUPs) to be 3 times as fast as my 500Mhz PWS500au (161 VUPs). Even so, =
that's probably one CPU only on the zx6000 so it's probably 6x faster if
= you multi-thread. Like I said, though, it's not that accurate on a
fast = VAX, I doubt it's accurate at all on non-VAX CPUs ;) 

What's in your VUPS.COM?   The one I'm aware of is pure DCL and it's called
VUPOMETER.COM.   

Running this VUPOMETER will not stress multiple CPUs as DCL is supervisor
mode single-threaded.   Also, DCL was written in Macro-32 on VAX.   On Alpha
and Itanium, it's also Macro-32.   DCL's data structures were generally not
very well aligned for modern memory access systems like Alpha and Itanium.
Alignment faults abound.   Alignment faults on Alphas could be tolerated but
on Itaniums it is a completely different story.   I'd wager that your VUPS
determinations aren't accounting for alignment faults in DCL.   In addition,
there are a number of things that were done in DCL macro that require some
more extensive handling on Alpha and Itanium.   From the silly nonsense like
passing the CLI service block in R9 for CHMS handling to weird unwinds[*].
I believe that a new VUP meter -- one that's not pure DCL -- is needed to 
better guage relative VUP performance.

[*] After development of the DCL debugger, I could elaborate for hours on
so much of the strangeness in DCL that makes it a less than ideal candidate
for   being any sort of deterministic VUP indicator.

-- 
VAXman- A Bored Certified VMS Kernel Mode Hacker       VAXman(at)TMESIS(dot)ORG

Well I speak to machines with the voice of humanity.