speed of microvax II
Jari A Jaanto
run? if compared to pc
--
- jari jaanto - jaffa/static - a student of computer science -
- http://jaffa.stc.cx - jaa...@cc.helsinki.fi -
- commodore 64 forever! -
Erno Palonheimo
> what is the speed of microvax II? how many mhz does it's processor ka630
> run? if compared to pc
You could try using a web search next time. But anyway, the CPU speed
is 0.9 VUP. 1 VUP is the speed of VAX-11/780, and equals 1 MIPS. But,
MIPS is really a Meaningless Indicator of Processor Speed, as one VAX
instruction tends to do much more than one instruction of about any
RISC processor.
--
Erno Palonheimo -- Unix, OpenVMS, Linux -- No compromise computing
Personal: e...@iki.fi Work-related: erno.pa...@sysforte.fi
Paul Hughett
: Jari A Jaanto <jaa...@cc.helsinki.fi> writes:
: > what is the speed of microvax II? how many mhz does it's processor ka630
: > run? if compared to pc
: You could try using a web search next time. But anyway, the CPU speed
: is 0.9 VUP. 1 VUP is the speed of VAX-11/780, and equals 1 MIPS. But,
: MIPS is really a Meaningless Indicator of Processor Speed, as one VAX
: instruction tends to do much more than one instruction of about any
: RISC processor.
Ah, you must be a mathematician--your answer is absolutely precise and
completely useless. ;-)
More seriously, the problem is that there are many of us that don't
have any feel for how fast a VAX 780 runs, so telling us that a uVAX
II runs 90% as fast doesn't help any. What I would like to see is a
translation into some processor that I'm actually familiar with; I
would guess from the given MIPS that a 780 is roughly comparable to a
486-25 or so but I'm just talking through my hat. Do you know of any
comparisons like this that are actually backed up by real knowledge or
testing?
Paul Hughett
Terry Murphy
Paul Hughett <hug...@chaplin.med.upenn.edu> wrote:
>More seriously, the problem is that there are many of us that don't
>have any feel for how fast a VAX 780 runs, so telling us that a uVAX
>II runs 90% as fast doesn't help any. What I would like to see is a
>translation into some processor that I'm actually familiar with; I
>would guess from the given MIPS that a 780 is roughly comparable to a
>486-25 or so but I'm just talking through my hat. Do you know of any
>comparisons like this that are actually backed up by real knowledge or
>testing?
I seriously doubt that a MVII is as fast as a 486-25. Heck, I don't
think VAXstation 3100's, which are triple the MVII speed, are as fast
as a 486-25.
As a reference point, I remember my 386SX-16 got about 2.13 BogoMIPS
from Linux. Of course, the MVII also has a floating point unit, an
instruction buffer, and a better bus (I assume QBUS is faster than ISA ;-).
I would guess that the MVII is approximately as fast as a 386SX-16
with the 387 FP co-processor, but probably a bit slower.
-- Terry
Tim Llewellyn
Paul Hughett wrote:
>
> Ah, you must be a mathematician--your answer is absolutely precise and
> completely useless. ;-)
>
> More seriously, the problem is that there are many of us that don't
> have any feel for how fast a VAX 780 runs, so telling us that a uVAX
> II runs 90% as fast doesn't help any. What I would like to see is a
> translation into some processor that I'm actually familiar with; I
> would guess from the given MIPS that a 780 is roughly comparable to a
> 486-25 or so but I'm just talking through my hat. Do you know of any
> comparisons like this that are actually backed up by real knowledge or
> testing?
It ain't just the speed that matters, those MicroVAX II's just run and run and
run, and the memory and I/O architecture was far superior to anything you'd
find on a peecee of similar vintage.
It was fast enough at the time, as long as you didn't try to load decwindows.
Tim, who still remembers the joy of having a MicroVax II to use all to myself
(well, with just one other) for running simulations, compared to timesharing
with the whole dept on an 11/750.
Oh, and exactly what is that 486-25 you want to compare it with running?
Win98?
>
> Paul Hughett
--
Tim Llewellyn, OpenVMS System Manager, Remarcs Project
MedAS at the BBC, Whiteladies Road, Bristol, UK.
Email tim.ll...@bbc.co.uk. Home tim.ll...@cableinet.co.uk
I speak for myself only and my views in no way represent those of
MedAS or the BBC.
Peter da Silva
equivalent... there's nothing with that little CPU power and that much I/O
bandwidth. But a 68000 at 8 MHz was about 60-70% of an 11/780. A 68030 at 16
was definitely ahead of it.
I'd say that a 780 with one user would feel about the same as a late model
386 (16, 20, 33) with a good bus underneath it.
Your 486/25 would have been noticably faster than an 11/780.
--
In hoc signo hack, Peter da Silva <pe...@baileynm.com>
`-_-' Ar rug tú barróg ar do mhactíre inniu?
'U` "Twenty-two points, plus triple-word-score, plus fifty points for
using all my letters. Game's over. I'm outta here."
Frederik Meerwaldt
A VAX has a totally different Bus structure and Processor structure than a
PC.....
VAXs are real stuff... PC's are only for playing...
--
Regards,
Freddy
Jari A Jaanto wrote in message <7g6v6s$gg2$1...@oravannahka.Helsinki.FI>...
>what is the speed of microvax II? how many mhz does it's processor ka630
>run? if compared to pc
>
Antonio Carlini
>what is the speed of microvax II? how many mhz does it's processor ka630
>run? if compared to pc
The MicroVAX II was rated at about 0.9 VUPs (i.e. about 0.9 the processing
power of a VAX-11/780).
It ran its 78032 at a cycle time of 200ns, so call it 5 million cycles per
second.
Compared to a PC or an XT it's obviously much, much better. Compared to a
Pentium, things are perhaps slightly less clear.
:-)
Antonio
Antonio Carlini Mail: car...@marvin.reo.dec.com
DECnet-Plus for OpenVMS Engineering
COMPAQ Reading, UK
Hoff Hoffman
In article <7g75k9$shd$1...@netnews.upenn.edu>, hug...@chaplin.med.upenn.edu (Paul Hughett) writes::: > what is the speed of microvax II? how many mhz does it's processor ka630
:: > run? if compared to pc
:More seriously, the problem is that there are many of us that don't
:have any feel for how fast a VAX 780 runs, so telling us that a uVAX
:II runs 90% as fast doesn't help any. What I would like to see is a
:translation into some processor that I'm actually familiar with; I
:would guess from the given MIPS that a 780 is roughly comparable to a
:486-25 or so but I'm just talking through my hat. Do you know of any
:comparisons like this that are actually backed up by real knowledge or
:testing?
Without intending to sound rude -- please do realize that these sorts
of performance-comparison questions are asked quite regularly, and
there is often some level of additional "educational" material required
of any responders -- additional information in terms of information
around general benchmarking and benchmarketing -- posting replies to
these sorts of (regular) questions.
Folks that have grown up with the Intel MHz comparisions may not
realize MHz- and MIPS-based comparisions are meaningless across
differing implementations within any particular computer architecture,
and are even less meaningful across different architectures. This
realization leaves you to consider performance comparisions using
"standard" benchmarks -- MHz numbers do not necessarily scale to
performance within families in an architecture, and regularly do
not scale across platforms.
(Asking for comparisions based on MHz can be a way to start these
benchmarking discussions (again), or sometimes simply flame wars.)
You will want to become familiar with the available benchmarks, and
particularly with various benchmark-related websites. As was stated
by a previous poster, "You could try using a web search next time".
Various benchmark-related resources are available. That said:
The VAX-11/780 (circa 1978) is the reference system for the SPECint92
and SPECfp92 benchmarks, with performance values of 1.0 for each.
The MicroVAX II (circa 1986) performance is roughly 0.9 of that of
the VAX-11/780 series system. (No, I don't happen to have actual
the SPECint92 and SPECfp92 values for the MicroVAX II series handy.)
A Compaq Deskpro 80486DX at 33 MHz (tested circa 1992) has SPECint92
and SPECfp92 ratings of 18.2 and 8.3, respectively.
-------------------------- pure personal opinion ---------------------------
Hoff (Stephen) Hoffman OpenVMS Engineering hoffman#xdelta.zko.dec.com
vha...@cc.helsinki.fi
Peter da Silva <pe...@baileynm.com> wrote:
>OK, to get a feel of a 780, well, there's nothing around today that's really
>equivalent... there's nothing with that little CPU power and that much I/O
>bandwidth. But a 68000 at 8 MHz was about 60-70% of an 11/780. A 68030 at 16
>was definitely ahead of it.
>
>I'd say that a 780 with one user would feel about the same as a late model
>386 (16, 20, 33) with a good bus underneath it.
>
>Your 486/25 would have been noticably faster than an 11/780.
A 11/780+VMS *felt* much slower than any pdp+RSX.. because RSX always
echoed characters immediately, whatever it was doing, and pdp sw was carefully
hand-optimised, while VAX/VMS terminal echo was sticky and the sw was written in
high level not-so-well optimised languages ;-)
Yeah.. the first experiences about VMS VAXEN were a great disappointment.
VesA
Peter da Silva
>In article <7g7co0$o...@web.nmti.com>,
>A 11/780+VMS *felt* much slower than any pdp+RSX.. because RSX always
>echoed characters immediately, whatever it was doing, and pdp sw was carefully
>hand-optimised, while VAX/VMS terminal echo was sticky and the sw was written in
>high level not-so-well optimised languages ;-)
I never used a 780 under VMS, nor had I used RSX when I was using the 11/70:
both were running what became BSD. Later I had an opportunity to compare
RSX-11 on an 11/44 and BSD on an LSI-11/23, and they seemed about as
responsive. The micro-pdp seemed to bog down more under load, though.
By the time I was using VMS, better systems were available.
Terry Kennedy
> As a reference point, I remember my 386SX-16 got about 2.13 BogoMIPS
> from Linux. Of course, the MVII also has a floating point unit, an
> instruction buffer, and a better bus (I assume QBUS is faster than ISA ;-).
You'd be surprised about the bus. A 386DX/25 was capable of running an
Adaptec 1542 SCSI controller at the 6.7Mbyte/sec adapter/host memory speed.
Some chipsets could even support 8Mbyte/sec, but that was uncommon.
From [fading] memory, the Q-bus is 3Mbyte/sec (again from memory, I think
that's for block mode). That's one of the things that made the DEC Q-bus
FDDI adapter so amusing 8-)
Terry Kennedy Operations Manager, Academic Computing
te...@spcvxa.spc.edu St. Peter's College, Jersey City, NJ USA
+1 201 915 9381 (voice) +1 201 435-3662 (FAX)
Paul DeMone
>
> what is the speed of microvax II? how many mhz does it's processor ka630
> run? if compared to pc
It runs at 10 MHz but that is a meaningless number for comparison
to x86 processors. The uVAX-II was rated about 90% of the performanc
of a VAX-11/780, a machine that by definition is rated at 1.00 SPECint92.
Here are some comparisons to PC CPUs of various eras
CPU SPECint92
uVAX-II 0.9
33 MHz 386 8.3
33 MHz 486 18.3
100 MHz Pentium 81.8
The microVAX is outclassed by even the cheapest PC, even a few years
old.
All opinions strictly my own.
--
Paul W. DeMone The 801 experiment SPARCed an ARMs race of EPIC
Kanata, Ontario proportions to put more PRECISION and POWER into
dem...@mosaid.com architectures with MIPSed results but ALPHA's well
pde...@igs.net that ends well.
Terry Kennedy
> I'd say that a 780 with one user would feel about the same as a late model
> 386 (16, 20, 33) with a good bus underneath it.
We had an 11/785 here running our own "4.3-Carson City" BSD flavor (this
was a 4.4BSD-snapshot kernel with Reno memory management and assorted hacks
of string and baling wire to run on VAXen, which got desupported in 4.4-
final). It supported student Unix and had 6 RA81 disk drives (most of which
were snapshots of our development code, not user files).
One day we had "the great disk crash" and lost 4 of the 6 RA81's (gotta
love those flakey drives). Ben Cohen and I decided "the heck with this" and
grabbed an old Everex 386DX/25 slimline box, stuck a 3Com 3C503, 16MB, 387,
and an Adaptec 1542 in it, gave it a 2GB SCSI disk to go with its internal
121MB IDE drive, and installed BSDI's BSD/OS version 0.2.something (pre-
beta) on it. We restored the user files from the undamaged RA81's onto the
PC, recompiled everything we could find that was in VAX a.out format, and
let the users loose on it the next day.
The general comment was "hey! what did you do to the VAX - it's so fast!".
However, if you're comparing different operating systems, raw hardware
speed becomes meaningless. Some really silly comparisons:
1) A PDP-11/23+ running RSTS/E and my MOP loader (written in Basic-Plus 2)
beat a VAX 8978 (a cluster of the then-biggest VAXes) and the compiled
VMS DECnet/MOP code at loading multiple terminal servers.
2) A PDP-11/70 running 2BSD Unix beat a VAX 9000 running VMS at computing
UUCP path tables. This can be partially explained by the sorry state of
the VAX C stdio library. However, VMS isn't exactly an I/O screamer -
for a fun time, compare the "iozone" benchmark results from a VMS box
with either the same box running Unix or a PC-class box. VMS trades a
robust set of filesystem support against speed, and in many cases VMS
is the right choice.
Peter da Silva
and 80286s, and it beat both out... but by less than a factor of two.
Thor Lancelot Simon
>: > run? if compared to pc
>
>: is 0.9 VUP. 1 VUP is the speed of VAX-11/780, and equals 1 MIPS. But,
>: MIPS is really a Meaningless Indicator of Processor Speed, as one VAX
>: instruction tends to do much more than one instruction of about any
>: RISC processor.
>
>completely useless. ;-)
>
>have any feel for how fast a VAX 780 runs, so telling us that a uVAX
>II runs 90% as fast doesn't help any. What I would like to see is a
>translation into some processor that I'm actually familiar with; I
>would guess from the given MIPS that a 780 is roughly comparable to a
>486-25 or so but I'm just talking through my hat. Do you know of any
You sure are! A 486-25 is probably in the neighborhood of 15 MIPS
on real-world code.
--
Thor Lancelot Simon t...@rek.tjls.com
"And where do all these highways go, now that we are free?"
Thor Lancelot Simon
Peter da Silva <pe...@baileynm.com> wrote:
>OK, to get a feel of a 780, well, there's nothing around today that's really
>equivalent... there's nothing with that little CPU power and that much I/O
>bandwidth. But a 68000 at 8 MHz was about 60-70% of an 11/780. A 68030 at 16
>was definitely ahead of it.
I don't think that's so. There were 386-16's with EISA buses: more I/O
bandwidth than a 780, but not a whole lot more CPU... well, maybe five
times as much CPU, but two or three times the I/O bandwidth...
Jonathan Stone
|> In article <7g7co0$o...@web.nmti.com>,
|> A 11/780+VMS *felt* much slower than any pdp+RSX.. because RSX always
|> echoed characters immediately, whatever it was doing, and pdp sw was carefully
|> hand-optimised, while VAX/VMS terminal echo was sticky and the sw was written in
|> high level not-so-well optimised languages ;-)
|>
|> Yeah.. the first experiences about VMS VAXEN were a great disappointment.
That can't be completely right. Weren't many of the utilities in VMS
1.3 were acutally repackaged PDP-11 utilties running in compatibility
mode? You could tell by looking with suitable utilities.
>Yeah.. the first experiences about VMS VAXEN were a great disappointment.
I even heard quite a few shops bought 11/780s early on just to do MCR
TKBs, because they were faster at doing it than any available PDP-11...
Jonathan Stone
|> In article <7g7co0$o...@web.nmti.com>,
|> >OK, to get a feel of a 780, well, there's nothing around today that's really
|> >equivalent... there's nothing with that little CPU power and that much I/O
|> >bandwidth. But a 68000 at 8 MHz was about 60-70% of an 11/780. A 68030 at 16
|> >was definitely ahead of it.
|>
|> I don't think that's so. There were 386-16's with EISA buses: more I/O
|> bandwidth than a 780, but not a whole lot more CPU... well, maybe five
|> times as much CPU, but two or three times the I/O bandwidth...
Wouldn't SBI bandwidth a better comparison? How far could one
(over)load an SBI anyway? three UBAs and an MBA?
Tim Llewellyn
vha...@cc.helsinki.fi wrote:
>
>
> A 11/780+VMS *felt* much slower than any pdp+RSX.. because RSX always
> echoed characters immediately, whatever it was doing, and pdp sw was carefully
> hand-optimised, while VAX/VMS terminal echo was sticky and the sw was written in
> high level not-so-well optimised languages ;-)
>
That was the "interrupt per keypress" terminal driver/IO controller mainly, wasn't
it.
DZ11 or something like that.
Kevin Oberman
> In article <7g7sdh$s...@alca.helsinki.fi>, vha...@cc.helsinki.fi
> writes:
> |> In article <7g7co0$o...@web.nmti.com>,
>
> |> A 11/780+VMS *felt* much slower than any pdp+RSX.. because RSX
> |>always echoed characters immediately, whatever it was doing, and pdp
> |>sw was carefully hand-optimised, while VAX/VMS terminal echo was
> |>sticky and the sw was written in high level not-so-well optimised
> |>languages ;-)
> |>
> |> Yeah.. the first experiences about VMS VAXEN were a great
> |>disappointment.
>
> That can't be completely right. Weren't many of the utilities in VMS
> 1.3 were acutally repackaged PDP-11 utilties running in compatibility
> mode? You could tell by looking with suitable utilities.
Yes. Under VMS V1.x most utilities were compatibility mode. Read that
as VERY slow.) Even VMS V2.x had quite a bit of compatibility mode
stuff. It was not until V3 that all utilities were either re-written
in VAX code or replaced by new utilities that ran as native.
--
R. Kevin Oberman, Network Engineer
Energy Sciences Network (ESnet)
Ernest Orlando Lawrence Berkeley National Laboratory (Berkeley Lab)
E-mail: obe...@es.net Phone: +1 510 486-8634
Tim Shoppa
> Yes. Under VMS V1.x most utilities were compatibility mode. Read that
> as VERY slow.) Even VMS V2.x had quite a bit of compatibility mode
> stuff. It was not until V3 that all utilities were either re-written
> in VAX code or replaced by new utilities that ran as native.
Some may think that it is stretching the definition of "utility"
(though I certainly don't): TECO was compatibility mode up until
late V4 or early V5.
Tim.
Paul Hughett
: Oh, and exactly what is that 486-25 you want to compare it with running?
: Win98?
Linux. It's a lightly used file server used for archiving medical reports.
But you could argue that Win98 vs VMS is a fairer comparison.
Paul Hughett
Carl Perkins
}>run? if compared to pc
}
}power of a VAX-11/780).
}
}It ran its 78032 at a cycle time of 200ns, so call it 5 million cycles per
}second.
}
}Compared to a PC or an XT it's obviously much, much better. Compared to a
}Pentium, things are perhaps slightly less clear.
}
}:-)
}
}Antonio
}
}Antonio Carlini Mail: car...@marvin.reo.dec.com
}DECnet-Plus for OpenVMS Engineering
}COMPAQ Reading, UK
According to the MicroVAX II 630QE Technical Information manual, the KA630
had a clock rate of 20 megaHertz. The memory cycle time was 400 nanoseconds.
Was the processor not running at the "clock rate"?
--- Carl
Adriaan Joubert
>
> But you could argue that Win98 vs VMS is a fairer comparison.
>
Not fair.
VMS worked then and works now. Win 98 wasn't around then and doesn't
work now either -- and definitely not in the type of networked
environment in which VMS worked just great (was I impressed logging onto
a cluster of 5 vaxes around 1987 for the first time, and you always
ended up on the machine with the lowest load). We gave up trying to use
Windows 95/8 machines on a mainly UNIX network last year, converted them
all to NT and that doesn't work either.
And I didn't work on a MicroVax, but I did work on a 3100 and, yes, it
was a real dog. But of course I made the mistake of running decwindows.
Adriaan
Thor Lancelot Simon
Jonathan Stone <jona...@DSG.Stanford.EDU> wrote:
>In article <7g8tvr$74l$1...@panix7.panix.com>, t...@panix.com (Thor Lancelot Simon) writes:
>|> In article <7g7co0$o...@web.nmti.com>,
>|> >OK, to get a feel of a 780, well, there's nothing around today that's really
>|> >equivalent... there's nothing with that little CPU power and that much I/O
>|> >bandwidth. But a 68000 at 8 MHz was about 60-70% of an 11/780. A 68030 at 16
>|> >was definitely ahead of it.
>|>
>|> I don't think that's so. There were 386-16's with EISA buses: more I/O
>|> bandwidth than a 780, but not a whole lot more CPU... well, maybe five
>|> times as much CPU, but two or three times the I/O bandwidth...
>
>Wouldn't SBI bandwidth a better comparison? How far could one
>(over)load an SBI anyway? three UBAs and an MBA?
Well, let's see. A 16-MHz EISA bus moves four bytes every cycle, so it's
got a peak bandwidth of 64MB/sec. Even if you cut that in half, you get
32MB/sec assuming very inefficient bus utilization.
I don't have my VAX architecture book here any more, but ISTR a Unibus
maxes out at around 5MB/sec and a Massbus at 10 or 12. So three Unibuses
is 15MB/sec, plus a 12MB/sec Massbus gives you 27MB/sec.
The real issue with the PC architecture isn't bus bandwidth -- because
the only thing that's ever *really* sucked that dry is video, and a box
being used for timesharing doesn't do any of that, to speak of -- it's
the totally insane interrupt architecture. But even so, I _distinctly_
recall a 486/66 with a decent SCSI disk and VLB controller feeling quite
a bit zippier than the 11/750 I still at that time occasionally booted
to keep myself grounded -- even with four users logged in doing stuff
on each, on the serial ports.
Remember, in those days nobody ran X -- not on timesharing machines! --
data sizes were much smaller (how big is your mailbox now? How big was
it then?), and even though you might have five people compiling at once,
since pcc couldn't do -O and -g at once (and pcc never optimized nearly
as hard as a modern GCC does) few would burn much CPU doing so
(particularly since the programs they built were much smaller!);
basically, the machines just weren't pushed as hard.
I know. I ran a large news server (about a GB of disk, which was large
at that time!) on that 11/750 well into the alt.binaries.pictures.foo
era. We'd rebuilt *everything* on the machine with gcc and a better
stdio library and it still dragged. A machine that had felt perfectly
fine just a year before was reduced to uselessness by the sheer volume
of data it now had to handle.
We replaced that .7 SPECint92 machine with a what, 20 SPECint92 Sun
4/330, which had maybe 2X the disk bandwidth and about 4X the RAM
(but was a RISC, so code was 2-3X as big...), and that held us about
a year and a half -- we hadn't been out of bandwidth, we'd been out
of CPU. So much for the myth of the perfectly-balanced VAX. At
that point, you couldn't take a full feed by UUCP anyway, so we ditched
Usenet news at that site completely.
VAXen were _always_ monumentally underpowered CPU-wise. It's
instructive to look at the debacle that ensued when DEC tried to turn
a small multiprocessor VAX into a small multiprocessor MIPS box - even
two CPUs more than saturated the backplane, so the machine generaly
ran faster with only one! The only thing that had kept the VAX feeling
"well balanced" was its puny CPU.
Boxes with much faster CPU's, particularly micros, get a lot of abuse
for being "unbalanced", but then again the bottlenecks are only revealed
because people run jobs on them they could never have run on a VAX in
the first place, because it didn't have the CPU to handle it!
VAXen were slow, in almost every respect. We just didn't know that because
their very slowness prevented us from pushing them hard enough to see where
they fell down in any way other than running out of CPU cycles.
Hoff Hoffman
In article <sreeml3...@ptavv.es.net>, Kevin Oberman <obe...@es.net> writes:
:Yes. Under VMS V1.x most utilities were compatibility mode. Read that
:as VERY slow.) Even VMS V2.x had quite a bit of compatibility mode
:stuff. It was not until V3 that all utilities were either re-written
:in VAX code or replaced by new utilities that ran as native.
That should read "It was not until V4.0..."
V3.0 included a number of compatibility-mode applications.
SYE was one example.
Jonathan Stone
|> > A 11/780+VMS *felt* much slower than any pdp+RSX.. because RSX always
|> > echoed characters immediately, whatever it was doing, and pdp sw was carefully
|> > hand-optimised, while VAX/VMS terminal echo was sticky and the sw was written in
|> > high level not-so-well optimised languages ;-)
|> >
|>
|> That was the "interrupt per keypress" terminal driver/IO controller mainly, wasn't
|> it.
|> DZ11 or something like that.
I wondered about that. The UBA780 interupt mechanism makes DZ-11s a
really bad choice for 780s. (When I had unibus macines to play with
the DZ11s had all been moved to 750s and the 780 used Able VMZs
instead, since the UBA&50 could vector the interrupts via the SCB
instead of reading the vector in software).
Peter da Silva
A VAX 11/750 was about equivalent to an Amiga 2000, except without the
graphics.
Jonathan Stone
|> In article <7g90sk$pl9$2...@nntp.Stanford.EDU>,
|> Jonathan Stone <jona...@DSG.Stanford.EDU> wrote:
|> VAXen were slow, in almost every respect. We just didn't know that because
|> their very slowness prevented us from pushing them hard enough to see where
|> they fell down in any way other than running out of CPU cycles.
I did enough tuning to frequently notice where machines were running
out of memory in ways that looked like they were IO bound, due to all
the paging traffic...
The other way to look at it is that before RISC, nearly all CPUs were
slow. (vaxes were acutally fast compared to similar IBm offerings of
the time, like the 4341). And VAXes (with admittedly slow CPUs) made
better timesharing machines than one might think, because they have
I/O bandwidth to burn, so they could swap jobs (which never fitted in
main memory) in and out very fast, relative to the CPU.
But first time I evaled a mips M/120 it seemed to scream -- until we
threw five compiles on it and the poor thing thrashed its PC-architecture
disks. In comparison, vaxes had relatively fast I/O.
Thor Lancelot Simon
Jonathan Stone <jona...@DSG.Stanford.EDU> wrote:
>In article <7gaa16$b8f$1...@panix7.panix.com>, t...@panix.com (Thor Lancelot Simon) writes:
>|> In article <7g90sk$pl9$2...@nntp.Stanford.EDU>,
>|> Jonathan Stone <jona...@DSG.Stanford.EDU> wrote:
>[snip]
>
>|> VAXen were slow, in almost every respect. We just didn't know that because
>|> their very slowness prevented us from pushing them hard enough to see where
>|> they fell down in any way other than running out of CPU cycles.
>
>I did enough tuning to frequently notice where machines were running
>out of memory in ways that looked like they were IO bound, due to all
>the paging traffic...
That is, because they ran out of memory. That was the other thing about
VAXen, you couldn't put enough RAM in them to get them *really* busy. At
least, not in a 750, and I never had a 780.
>The other way to look at it is that before RISC, nearly all CPUs were
>slow. (vaxes were acutally fast compared to similar IBm offerings of
>the time, like the 4341). And VAXes (with admittedly slow CPUs) made
>better timesharing machines than one might think, because they have
>I/O bandwidth to burn, so they could swap jobs (which never fitted in
>main memory) in and out very fast, relative to the CPU.
I'm not sure I grasp this. They made better timesharing machines than
one might think because the CPU was slow enough that relatively speaking,
they could push and pull pages from the disk before that bottleneck was
too obvious? I guess that might actually have an effect on interactive
performance, or the perception thereof... hm. I never though about it
that way.
>But first time I evaled a mips M/120 it seemed to scream -- until we
>threw five compiles on it and the poor thing thrashed its PC-architecture
>disks. In comparison, vaxes had relatively fast I/O.
Well, to be fair, if you'd put as many individual spindles on the M120 as
you'd had to provide the same capacity on the VAX, it probably would have
performed just as well, right? Or did it have deliberately crippled PIO
disk controllers like a pmax? One thing people seem to mean when they
say that VAXen had "tons of I/O bandwidth" (which they didn't, you can
put as much data on an ISA bus as a Unibus, and QBus vaxen quite notably
have _way too little_ I/O bandwidth, about 1/2 an ISA bus!) is that they
had enough individual spindles on them to have a lot of _concurrency_
in the I/O subsystem, which allowed you to hand the disks many
simultaneous job-streams of I/O at once and not get a noticeable decrease
in performance. Newer machines had brand-new disks with higher capacities
per unit, and less of them, and nobody stopped to think that that meant
that they were sacrificing concurrency...a mistake which is being made
again today with cheap fast IDE disks in the 6GB range minimum, where
it's very tempting to buy as much space as you need, but not as much
concurrency as you need. People keep learning the same lessons over and
over again...
...other systems, even "unbalanced" PCs, with I/O subsystems configured
right for their workload -- even a single SCSI adapter and three or four
old SCSI disks, instead of one big disk -- always outperformed the old
"perfectly balanced" VAXen and their ilk. The thing is, they were often
built by people who didn't understand how to balance machines themselves,
and without the limitations of the VAX hardware (you couldn't *buy* big
disks, so you *had* to buy a lot of little ones, etc.) they tended to
screw it up.
Even my pmax (pmin, really) was okay for multiuser interactive use when I
put four SCSI disks on it, instead of the two it shipped with.
Zane H. Healy
> A VAX 11/750 was about equivalent to an Amiga 2000, except without the
> graphics.
Sounds more like a A1200, unless the A2000 has a 68020 accelerator in it.
Zane
--
One of these day's I've got to quit being lazy and come up with a signature.
<<Yawn>>
Peter da Silva
Zane H. Healy <hea...@shell2.aracnet.com> wrote:
>Peter da Silva <pe...@baileynm.com> wrote:
>> I've got it.
>> A VAX 11/750 was about equivalent to an Amiga 2000, except without the
>> graphics.
>Sounds more like a A1200, unless the A2000 has a 68020 accelerator in it.
A 68020 is much faster than an 11/780. A 68000 was about 70% of a 780, and
so was a 750.
Yes, I know a 68000 can't do VM. But in terms of performance they're close.
I'd suggest a UNIX-PC, but I don't know how fast its bus was, and fewer people
will have experience with it.
Thor Lancelot Simon
Zane H. Healy <hea...@shell2.aracnet.com> wrote:
>Peter da Silva <pe...@baileynm.com> wrote:
>> I've got it.
>
>> A VAX 11/750 was about equivalent to an Amiga 2000, except without the
>> graphics.
>
>Sounds more like a A1200, unless the A2000 has a 68020 accelerator in it.
Not really. A 68020 has more oomph than an 11/750, by a good margin --
it's probably got about the horsepower of the processor in the earlier
MV3100s, or maybe an 8650. Workstations based on the 68010 plus MMU
('451 or roll-your-own) were basically '750 equivalent or even a bit
better in straight CPU performance, so aside from the lack of memory
management a 68000-based Amiga would be, too.
Paul DeMone
>
> car...@marvin.reo.dec.com (Antonio Carlini) writes...
> }In article <7g6v6s$gg2$1...@oravannahka.Helsinki.FI>, Jari A Jaanto <jaa...@cc.helsinki.fi> wrote:
> }>what is the speed of microvax II? how many mhz does it's processor ka630
> }>run? if compared to pc
> }
> }The MicroVAX II was rated at about 0.9 VUPs (i.e. about 0.9 the processing
> }power of a VAX-11/780).
> }
> }It ran its 78032 at a cycle time of 200ns, so call it 5 million cycles per
> }second.
> According to the MicroVAX II 630QE Technical Information manual, the KA630
> had a clock rate of 20 megaHertz. The memory cycle time was 400 nanoseconds.
>
> Was the processor not running at the "clock rate"?
>
> --- Carl
To summarize the MicroVAX 32 device
- 3.0 um NMOS, 2 metal layers, 430 Angstrom gate oxide thicknes
- 125,000 transistor sites on 8.7 x 8.6 mm die packaged in 68 LCC,
- worst case power dissipation is 3 Watts
- implements 175 of 304 native VAX instructions directly, the
rest are emulated by macrocode (the 70 FP instructions are
not emulated if an external FPU is present).
To summarize the clocking of the MicroVAX 32 device:
- takes a 2x external clock input - 20 MHz, 50 ns
- internal 10 MHz clock, divided into four overlapping clock phases, PHI1,
PHI2, PHI3, and PHI4
- each microcycle takes 2 internal clock periods or 200 ns
- the main data path or "E box" can read two registers, perform
an ALU or shift operation and write the results back in one
200 ns microcycle.
- external bus cycle time is 400 ns
Intruction Timing examples:
- MOVL reg,reg 400 ns
- ADDL2 reg,reg 400 ns
- MOVL mem,reg 800 ns
- ADDL2 mem,reg 800 ns
- MOVL reg,mem 600 ns
- ADDL2 reg,mem 1200 ns
- conditional branch no taken 200 ns
- conditional branch taken 800 ns
The microVAX's closest equivalent in process, performance, power, pin count
and support for virtual memory would be a 10 MHz 68010 (found in the Sun-2
family). Any 68020/i386 or above clearly outperforms the microVAX.
Can will we put this thread out of its misery now?
Sheila and the GeeSer
Many of the EISA bus PC machines could sustain 16MB/s across the bus but
most of the drives and the software could not support that.
The Qbus is slower, at about 3MB/s.
If you want to get into comparisons, really trying to evaluate the
architectures, you have to be very careful to determine what you want to
know so you don't step into the bogo traps.
bob
Terry Kennedy wrote:
>
> Terry Murphy <tsmu...@students.uiuc.edu> writes:
> > As a reference point, I remember my 386SX-16 got about 2.13 BogoMIPS
> > from Linux. Of course, the MVII also has a floating point unit, an
> > instruction buffer, and a better bus (I assume QBUS is faster than ISA ;-).
>
> You'd be surprised about the bus. A 386DX/25 was capable of running an
> Adaptec 1542 SCSI controller at the 6.7Mbyte/sec adapter/host memory speed.
> Some chipsets could even support 8Mbyte/sec, but that was uncommon.
>
> From [fading] memory, the Q-bus is 3Mbyte/sec (again from memory, I think
> that's for block mode). That's one of the things that made the DEC Q-bus
> FDDI adapter so amusing 8-)
>
> Terry Kennedy Operations Manager, Academic Computing
> te...@spcvxa.spc.edu St. Peter's College, Jersey City, NJ USA
> +1 201 915 9381 (voice) +1 201 435-3662 (FAX)
--
real address is shsrms at erols dot com
The Herbal Gypsy and the Tinker.
Tobias Göller
>
> I've got it.
>
> A VAX 11/750 was about equivalent to an Amiga 2000, except without the
> graphics.
I don't think this was a fault. Graphics take up a lot of CPU;
the amiga 2000 had a MC68020 cpu.... and that wasn't that slow.
I often dream of my "first" computer, the sinclair QL. When I
switched to the i286 two years later (because the QL went
dead.... :-(( ) I wasn't terribly impressed by it's speed. Many
things run better and faster on the QL!
> --
> In hoc signo hack, Peter da Silva <pe...@baileynm.com>
> `-_-' Ar rug tú barróg ar do mhactíre inniu?
> 'U` "Twenty-two points, plus triple-word-score, plus fifty points for
> using all my letters. Game's over. I'm outta here."
--
--
COM.BOX WINET
Tobias Goeller
Tobias Göller
>
> In article <7gahr5$an8$1...@spitting-spider.aracnet.com>,
> Zane H. Healy <hea...@shell2.aracnet.com> wrote:
> >> I've got it.
>
> >> A VAX 11/750 was about equivalent to an Amiga 2000, except without the
> >> graphics.
>
>
> so was a 750.
Sure, but as the 11/780 didn't come with a GUI, the remaining
power was far enough....
> Yes, I know a 68000 can't do VM. But in terms of performance they're close.
>
> I'd suggest a UNIX-PC, but I don't know how fast its bus was, and fewer people
> will have experience with it.
>
Terry Kennedy
> I often dream of my "first" computer, the sinclair QL. When I
> switched to the i286 two years later (because the QL went
> dead.... :-(( ) I wasn't terribly impressed by it's speed. Many
> things run better and faster on the QL!
Along that same line, I was involved with some aspects of the design of
the Spectravideo 318/325 systems. These were the ancestors of the "MSX"
systems.
The 318/325 used the same Microsoft BASIC as BASICA on the IBM PC. Yet
the 318/325 used a Z80 at 3.58MHz, while the PC used an 8088 at 4.77 MHz.
The 318/325 blew the PC away (at least by 2 to 1) on all of the BASIC
benchmark tests anybody did.
Antonio Carlini
>According to the MicroVAX II 630QE Technical Information manual, the KA630
>had a clock rate of 20 megaHertz. The memory cycle time was 400 nanoseconds.
IIRC the 78032 had four phases in each cycle so 20MHz could well correspond to
the 5MHz I derived from a 200ns cycle time. My notes claim that the 200ns
cycle time came from DTJ Vol 1 Num 7 but I don't have that to hand. The
nearest I have to a 630QE tech manual is the 630QY,QZ one which is silent
about almost anything remotely technical :-)
The next time I'm passing a spare KA630 I'll look to check, but I'm reasonably
certain that the clock crystal is 40MHz so I presume it gets divided down to
form the chip clock(s) required. I don't know what was quoted as the "module
clock speed". I do know that the VAX 4000 Model 705A Tech Info manual quotes a
cycle time of 12ns and a CPU clock of 444.44MHz for the VAX 4000-705A; this is
one of the reasons I prefer cycle times to MHz ratings :-)
My recollection about the bus cycle time is the same as your manual i.e.
400ns. This is (IIRC) the reason that the KA630 had no cache: it could not get
off chip any faster than 400ns/cycle which was how fast memory ran anyway: the
whole 16MB is a cache!
John F Carr
Tobias =?iso-8859-1?Q?G=F6ller?= <t.go...@combox.de> wrote:
>Sure, but as the 11/780 didn't come with a GUI, the remaining
>power was far enough....
MIT used to have several VAX 11/750s with about 20 VT 220/240
terminals and a pair of VS100 graphics displays each. Very slow.
I think they were running X9 at the time. One of the labs I
worked in had a 512x512 color display hooked up to its 11/750,
although by that time it had been obsoleted by VS II/GPX
workstations.
--
John Carr (j...@mit.edu)
Alan Adams
A 780 with 8 DZ-11s doesn't work very well.
When the first terminal servers appeared we put some on our ethernet
to connect to the 780. The improvement in 780 performance was
staggering.
Prior to that we had locked the terminal lines to 2400 baud so that all
the programmers running EDT didn't stop the system. (The system could
easily run 50% interrupt with 10 people running EDT at 9600).
(EDT was the less efficient editor, replaced by TPU which handled
screen updating MUCH more efficiently.)
With the terminal servers (1 interrupt per 80mSec for up to 8 users)
we upped the speed to 9600 and improved performance some more, because
more
characters could be sent to the screen per interrupt.
--
Alan Adams
Roy Omond
> A 780 with 8 DZ-11s doesn't work very well.
>
> When the first terminal servers appeared we put some on our ethernet
> to connect to the 780. The improvement in 780 performance was
> staggering.
> Prior to that we had locked the terminal lines to 2400 baud so that all
> the programmers running EDT didn't stop the system. (The system could
> easily run 50% interrupt with 10 people running EDT at 9600).
Who remembers the "performance" on a 780 with DZ11s when two
(or even one !) users were playing Space Invaders :-) ?
The DMF32 was a huge improvement; it allowed us to have 4 or 5
Space Invaders simultaneously :-)
Roy Omond
Blue Bubble Ltd.
Brian JARAI Chase
Peter da Silva <pe...@baileynm.com> wrote:
>OK, to get a feel of a 780, well, there's nothing around today that's really
>equivalent... there's nothing with that little CPU power and that much I/O
>bandwidth. But a 68000 at 8 MHz was about 60-70% of an 11/780. A 68030 at 16
>was definitely ahead of it.
>
>I'd say that a 780 with one user would feel about the same as a late model
>386 (16, 20, 33) with a good bus underneath it.
Well, running NetBSD... My MicroVAX II get's about 1300 dhry/s vs
60000 dhry/s for a Sparc 2 or I think about 10000 dhry/s on my 386DX/40.
-brian.
--
---
Brian "JARAI" Chase | http://world.std.com/~bdc/ | VAXZilla LIVES!!!
Tobias Göller
>
> Tobias G?ller <t.go...@combox.de> writes:
> > I often dream of my "first" computer, the sinclair QL. When I
> > switched to the i286 two years later (because the QL went
> > dead.... :-(( ) I wasn't terribly impressed by it's speed. Many
> > things run better and faster on the QL!
>
> Along that same line, I was involved with some aspects of the design of
> the Spectravideo 318/325 systems. These were the ancestors of the "MSX"
> systems.
>
> The 318/325 used the same Microsoft BASIC as BASICA on the IBM PC. Yet
> the 318/325 used a Z80 at 3.58MHz, while the PC used an 8088 at 4.77 MHz.
> The 318/325 blew the PC away (at least by 2 to 1) on all of the BASIC
> benchmark tests anybody did.
.... and I don't think, it's an all-intel problem. It's just
surprising, how fast a PC could be - if it's not running a
MS-Operating-system....
Compilers coming from Microsoft (if I think of the C-Compiler...)
weren't bad; better: they were good. But the OS-Stuff was NEVER
something m$ did with competence.....
> Terry Kennedy Operations Manager, Academic Computing
> te...@spcvxa.spc.edu St. Peter's College, Jersey City, NJ USA
> +1 201 915 9381 (voice) +1 201 435-3662 (FAX)
--
COM.BOX WINET \|/
Tobias Goeller ({o.o})
\./
t.go...@combox.de U
Bob Marcan
code.