[Info-vax] DEC Multia (UDB) issues

Sat Aug 27 08:28:04 EDT 2011

On Aug 27, 10:16 am, Hans Vlems <hvl... at freenet.de> wrote:
> On Aug 26, 10:42 am, John Wallace <johnwalla... at yahoo.co.uk> wrote:
>
>
>
> > On Aug 26, 9:11 am, Hans Vlems <hvl... at freenet.de> wrote:
>
> > > On Aug 23, 1:26 pm, MG <marcog... at SPAMxs4all.nl> wrote:
>
> > > > On 23-8-2011 12:39, Hans Vlems wrote:
>
> > > > > Running VMS on a Multia is not that difficult, even I managed to make
> > > > > it work. My Multia came with just 40 MB main memory IIRC and though
> > > > > VMS will boot and run in that limited space it halted within half an
> > > > > hour.
>
> > > > Which version did you run, do you recall that?
>
> > > VMS V7.3.
>
> > > > I'm a bit surprised, to be honest, that VMS wouldn't be able to handle
> > > > that.  Is it as simple that Alpha, being 64-bit, would ordinarily
> > > > require more RAM?  Or were you performing complex calculations, heavily
> > > > using DECwindows and therein based programs, perhaps?
>
> > > The 40 MB came in four boards: two 12 MB boards and two 8 MB boards.
> > > My conclusion after a
> > > couple of sudden stops of VMS (and that doesn't happen too often,
> > > right) was that either 40 MB was
> > > not enough to run X-motif in or perhaps that one of the memory boards
> > > was slowly going south.
> > > Then I learnt that AlphaServer 1000A boards fit in a Multia and after
> > > the memory upgrade the problems were over.
> > > So the cause is still uncertain.
> > > BTW a Multia is slow compared to an AlphaServer 300 4/266 system which
> > > has about the same internals.
>
> > > > I'm asking because most VAXen with even less memory will run just fine,
> > > > I've noticed (like some of the IPv7.net nodes).
>
> > > Correct but the memory requirements of 32 bit VMS and 64 bit VMS are
> > > quite different. Furthermore Alpha/VMS has
> > > functionality in it that just is not present in VAX/VMS. And VAX/VMS
> > > started life on a VAX 11/780 where memory capacity
> > > could be as low as 256 kB (entry level 128 kB perhaps!).
>
> > > > > The memory was expanded to 256 MB with memory boards with a height that
> > > > > only just fitted with the box closed. Well, you read about its cooling/
> > > > > airflow issues so the memory was reduced to 4 x 32 MB boards. It shares
> > > > > memory boards with the AlphaServer 1000A 5/xxx models.
>
> > > > There's a mere 32 MB in mine.  I don't think I really want to use the
> > > > system for anything else than a glorified DECterm (especially since my
> > > > LK463 works fine on it) to connect to my I64 cluster.  32 MB should be
> > > > fine for that, correct?
>
> > > Very likely, my Multia ran the DCE desktop on the graphics console and
> > > at the same time I ran an X-session on
> > > a PC with Reflection installed. And Motif is possibly not the easiest
> > > windows environment...
> > > If you just run the old style DECwindows kit, like the one that runs
> > > on top of VAX/VMS then 32 MB will work.
> > > Though adding memory is cheap (eBay) and any upgrade from 32 MB will
> > > be noticable.
>
> > > > > This works well, though its built in 10 Mb/s ethernet adapter slows it
>
> > > > Purely out of curiosity, since you brought this up: Are 100BASE-TX
> > > > ethernet adapters, since there's one PCI slot, available/known to work
> > > > in it?
>
> > > I have never dared to try! The Multia has cooling issues and I didn't
> > > want to install too much additional equipment in it.
> > > I also have an AlphaServer 300 which outperforms the Multia, so once
> > > VMS was installed I lost interest in the Multia.
> > > My study in the attic is small, has no airco and is fairly warm from
> > > april to september. So not an environment to keep a
> > > Multia in for daily work (a Digital Server 5305 does just that now).
> > > AFAIK there are no restrictions on what you can put on the Multia's
> > > PCI bus. A DE500 interface ought to work and wouldn't
> > > raise the inside temperature too much. YMMV.
>
> > > >   - MG
>
> > > Hans
>
> > "a Multia is slow compared to an AlphaServer 300 4/266 system which
> > has about the same internals. "
>
> > A Multia is indeed slow, but there are no other mainstream Alpha boxes
> > with "about the same internals" because there are no other boxes based
> > on the 21066/21068 CPU chip family used in Multia. The AlphaBook is
> > close though (and there were VME boards but that's a different story).
>
> > It's the 21066/21068 that is the main performance constraint in the
> > Multia, not what's around it, although having enough memory is
> > obviously important too.
>
> > The 21066/21068 has pretty much everything integrated onto the CPU
> > chip itself, apart from the SIMMs and the graphics and  and a handful
> > of PCI-derived IO (the PCI bus comes direct off the 21066/21068).
> > There is no off-chip northbridge/southbridge equivalent as used on
> > 21064, x86, etc. The 21066/21068 is not quite an ARM-licencee-style
> > "system on chip" but it's a lot closer than an x86 or other Alpha
> > design would be.
>
> > Getting an x86-based Multia into the same form factor later in the
> > life of Multia was quite an achievement in its day, given all the x86
> > support electronics needed.- Hide quoted text -
>
> > - Show quoted text -
>
> The Multia does indeed run at a lower clock speed, 12% slower than the
> AS300.
> I'd forgotten about that. Another cause for a poorer performance is
> the 2,5" notebook
> harddisk.
> For its time the Multia was a marvel, lots of power in a very small
> box, not unlike the
> release of the VAXstation 2000. The latter has less heat issues and a
> more rugged
> construction quality, though.
> Hans

$ SET MODE /PEDANT

The clock speed difference tells only part of the performance story.
There are a number of other relevant differences between the
21066/21068 chip in the Multia and the 21064(A) chips and chipsets in
other Alphas of the era, including
. on chip cache size and effectiveness
. off chip cache size and effectiveness
. system bus width to main memory (64 bits on Multia, 64 or 128
depending on model with other Alphas)
. system bus performance in other respects (e.g. Multia uses CPU-
>memory bus cycles to do DRAM refresh)
. PCI bus performance

The marvel isn't just the Multia, it's based on everything that was
incorporated in the 21066/21068 chips (to reduce the overall system
cost). But the reduced system complexity and cost came at a price;
performance took a hit.

A quick dig around for price and performance numbers found these 1995
examples [1] (which I haven't verified but it sounds plausible):

Same clock speed, different processors, little config info:

     processor       SPECint92       SPECfp92
        21066-233       94              110
        21064a-233      157             183

Slightly different clock speeds, different processors, a bit of config
info:

Type     Speed   Cache   Processor       Price   SPECint/SPECfp
AXPpci33 233MHz  256kb   21066           $1,395    94 / 110

PC64     275MHz  2MB     21064a          $4,295    189 / 262

Summary: Clock speed alone is not a reliable predictor of performance.

[1] figures found in two separate posts in a
comp.os.linux.development.system thread titled "Linux on a DEC Alpha"
starting on Sep 5 1995, find it at
http://groups.google.com/group/comp.os.linux.development.system/browse_thread/thread/6d44522761e7757a/b1910954fb47307b
or wherever
but probably sourced originally from DEC data which wasn't so easy to
find.