VAX speed vs. PC speed
Walter H Dick, III
a 486 at 12 MHZ or 8 MHZ or what? Thanks in advance!
Walter Dick
whd...@cs.rit.edu
Mighty Firebreather
>Does anyone know what a VUP would be equivalent to? Does one VUP equal
>a 486 at 12 MHZ or 8 MHZ or what? Thanks in advance!
>
VUPs are *very* roughly equivalent to MIPS or Specmarks. My own
unscientific benchmark suggests that my 486DX/33 delivers ten to fifteen
times the performance of a one VUP machine.
*************************************************************************
* Here, there be dragons! *
* dra...@nscvax.princeton.edu *
* *
* I'm job hunting. Any offers or leads will be appreciated. *
* Thanks! *
* Richard B. Gilbert *
*************************************************************************
Arne Vajhøj
> a 486 at 12 MHZ or 8 MHZ or what?
It is impossible to make a good general comparison of CPU power for two
different CPU architectures. You can get some benchmarks for more
specific purposes, that will enable you to very roughly to estimate how
fast a CPU is for certain types of applications. But the benchmarks must
be seleceted carefull to actually match your kind of applications and
be very aware about what you are measuring with the benchmark (CPU speed,
memory access time, disk access time, OS efficiency, compiler efficiency).
Enogh said.
Arne
Arne Vajhøj local DECNET: KO::ARNE
Computer Department PSI: PSI%238310013040::ARNE
Business School of Southern Denmark Internet: AR...@KO.HHS.DK
No, that's me over here!
architectures except for application-to-application comparisons.
With that caveat, I did a comparison between a scalar (i.e. non-
vectorising) compute-intensive code for both a VAXstation 3100/76
and a 486DX/33 under Unix and found the 'DX' speed to be within
3% of the VAX. This suggests a VUP rating of ~7.6 for the DX
**FOR THAT APPLICATION**. I have not done a VMS/DOS comparison.
Cheers
JB Hedley
AECL Research
Whiteshell Labs
Pinawa, Manitoba, CANADA
<hed...@wl.aecl.ca>
Walter H Dick, III
>> Does anyone know what a VUP would be equivalent to? Does one VUP equal
>> a 486 at 12 MHZ or 8 MHZ or what?
>
>It is impossible to make a good general comparison of CPU power for two
>different CPU architectures. You can get some benchmarks for more
>specific purposes, that will enable you to very roughly to estimate how
>fast a CPU is for certain types of applications. But the benchmarks must
>be seleceted carefull to actually match your kind of applications and
>be very aware about what you are measuring with the benchmark (CPU speed,
>memory access time, disk access time, OS efficiency, compiler efficiency).
>
>Enogh said.
>
Difficult yes. Impossible no. Some kind person was able to send me the
Specmark equivalents of various 486 PCs. I was only interested in RAW
CPU power. I realize that there are significant differences between bus,
disk, and memory speeds. The following information was given to me:
As you may already know, a VUP is a "Vax Unit of Performance", where
"Vax" means a Vax 11/780. The only way I know to compare Vax's to a
486 is via some SPEC benchmarks I have. Here are some relevant entries
from a table of SPEC benchmarks. Note that by definition, a Vax 11/780
is benchmarked at 1 SPECmark.
--------------------------------------------------------------------------------
System CPU Clk MHz Cache SPEC Info Source
Name Type Ext/Int Ext+I/D Mrk89 Date Obtained
------------- ----------- ------- -------- ----- ----- -------------------------
VAX 11/780 VAX 1.0 1.0 Reference System
i386/387 80386/7 33 64+0 4.3 1992 Intel
i486DX 80486 25 128+8 8.7 1990 Intel
i486DX 80486 33 0+8 11.1 1991 Intel
i486DX 80486 50 256+8 21.9 Oct92 comp.arch
i486DX2 80486 66 256+8 25.6 1992 Intel
So in round number, a 33Mz 386 is 4 times faster then an 11/780. Of
course, that's just raw computation, there may be things that an
11/780 can do better then a 386 (for example heat a room!).
Del Armstrong
Darrin Robinson
[misc stuff deleted...]
>
>--------------------------------------------------------------------------------
>System CPU Clk MHz Cache SPEC Info Source
>Name Type Ext/Int Ext+I/D Mrk89 Date Obtained
>------------- ----------- ------- -------- ----- ----- -------------------------
>VAX 11/780 VAX 1.0 1.0 Reference System
>i386/387 80386/7 33 64+0 4.3 1992 Intel
>i486DX 80486 25 128+8 8.7 1990 Intel
>i486DX 80486 33 0+8 11.1 1991 Intel
>i486DX 80486 50 256+8 21.9 Oct92 comp.arch
>i486DX2 80486 66 256+8 25.6 1992 Intel
>
>So in round number, a 33Mz 386 is 4 times faster then an 11/780. Of
>course, that's just raw computation, there may be things that an
>11/780 can do better then a 386 (for example heat a room!).
>
>
> Del Armstrong
I'm sorry, but I have to say something! :-> I want to see a "four times
faster than a VAX 11/780" 33Mhz 386 beat a VAX 11/780 with 157 users on it
running an I/O intensive application...
What others are trying to say is that the *application* is VERY important
in determining if the system/platform/OS/compiler that you use is the best
for YOUR situation. I'm starting to really HATE the "numbers" game with
SPECmarks and the like...
The VAX 11/780 I ran had 157 users on one night all running a telemarketing
application with each users WSDEF and QUO set to 1200 pages. The system had
32MB of memory and a farm of RA81/RA82 (and HSC50s over CI). It had five
paging/swapfiles each with 60MB in them (120,000 blocks). The system was
tuned by hand and RMS tweaked to squeeze out some more. I don't even
want to imagine a i386/33 attempting this feat! The numbers tell you raw
computational power, but what happens when you start multi-tasking or your
particular application doesn't use a particular RISC instruction or does
a lot of Floating Point arithmetic, Integer arithmetic, etc.?
BTW, after that summer with 157 users on a VAX 11/780 and 40 on a 11/750
(yes, it was rough, believe me! It took users 30-40 minutes to log in!)
I was granted a VAX 8810... we put all the users on that one and hired
a 100 more people and had 300+ users on that each night and ran 70% idle
(except during logins!))
Darrin
, Darrin E. Robinson (DER31) Hamnet N1LLV 146.700-, 146.880- MHz
/| Systems Programmer Internet dar...@MIT.EDU
\| Dist. Computing & Network Services robi...@Planetary.Brown.EDU
|\ M.I.T. Information Systems ICBMnet 41 29 24 N 71 18 48 W (NPT)
|/ 1 Amherst St. - Rm E40-338 SPANet PGGIPL::ROBINSON (7132)
' Cambridge, MA 02139, USA AT&Tnet (617) 253-0131
Luke Brennan
[snip]
> BTW, after that summer with 157 users on a VAX 11/780 and 40 on a 11/750
heh heh.. we ran a 11/750 with 15Mb and 48 users! (yes 48!)
It wasn't too slow - even on logins
Luke
g...@brt.deakin.edu.au
robi...@stout.geo.brown.edu (Darrin Robinson) writes:
>In article <1994Jan6.2...@cs.rit.edu> whd...@cs.rit.edu (Walter H Dick, III) writes:
>[misc stuff deleted...]
>>
>>--------------------------------------------------------------------------------
>>System CPU Clk MHz Cache SPEC Info Source
>>Name Type Ext/Int Ext+I/D Mrk89 Date Obtained
>>------------- ----------- ------- -------- ----- ----- -------------------------
>>VAX 11/780 VAX 1.0 1.0 Reference System
>>i386/387 80386/7 33 64+0 4.3 1992 Intel
>>i486DX 80486 25 128+8 8.7 1990 Intel
>>i486DX 80486 33 0+8 11.1 1991 Intel
>>i486DX 80486 50 256+8 21.9 Oct92 comp.arch
>>i486DX2 80486 66 256+8 25.6 1992 Intel
>>
>>So in round number, a 33Mz 386 is 4 times faster then an 11/780. Of
>>course, that's just raw computation, there may be things that an
>>11/780 can do better then a 386 (for example heat a room!).
>>
>>
>> Del Armstrong
>
>I'm sorry, but I have to say something! :-> I want to see a "four times
>faster than a VAX 11/780" 33Mhz 386 beat a VAX 11/780 with 157 users on it
>running an I/O intensive application...
But our old Sequent S27 ( two 386/387 16MHz processors) will beat the
daylights out of a VAX 11/780 on any type of application.
Just goes to show that the Intel processors in a well designed system
(Not a PC compatible) will perform up to their SPEC ratings.
The Sequent has a fast, wide internal bus, good cache design and good I/O.
Greg O'Sullivan
(g...@brt.deakin.edu.au)
Arne Vajhøj
> >> a 486 at 12 MHZ or 8 MHZ or what?
> >
> >It is impossible to make a good general comparison of CPU power for two
> >different CPU architectures. You can get some benchmarks for more
> >specific purposes, that will enable you to very roughly to estimate how
> >fast a CPU is for certain types of applications. But the benchmarks must
> >be seleceted carefull to actually match your kind of applications and
> >be very aware about what you are measuring with the benchmark (CPU speed,
> >memory access time, disk access time, OS efficiency, compiler efficiency).
>
> Specmark equivalents of various 486 PCs. I was only interested in RAW
> CPU power. I realize that there are significant differences between bus,
> disk, and memory speeds. The following information was given to me:
I admit that the Specmark benchmark has a good reputation, but it is
still not a good general comparison (a good general comparison will
enable you to forecast the speed difference between two machines for an
application given that you know enough about the application f.ex.
instructio mix). Specmark is based on what the writers has considered
a typical instruction mix. But they may have considered wrong. And
there are a lot of applications that are not typical.
Bob Comarow
>>>So in round number, a 33Mz 386 is 4 times faster then an 11/780. Of
>>>course, that's just raw computation, there may be things that an
>>>11/780 can do better then a 386 (for example heat a room!).
>>>
>>>
>>> Del Armstrong
>>
>>I'm sorry, but I have to say something! :-> I want to see a "four times
>>faster than a VAX 11/780" 33Mhz 386 beat a VAX 11/780 with 157 users on it
>>running an I/O intensive application...
157 users on a 780?
>
> But our old Sequent S27 ( two 386/387 16MHz processors) will beat the
> daylights out of a VAX 11/780 on any type of application.
>
>
> Just goes to show that the Intel processors in a well designed system
> (Not a PC compatible) will perform up to their SPEC ratings.
> The Sequent has a fast, wide internal bus, good cache design and good I/O.
>
>
> Greg O'Sullivan
> (g...@brt.deakin.edu.au)
You mean, like in running an applciation written on any size workstation
or larger size VAX without recompiling?
Bob
Bill Bochnik (Info Systems)
>>> Does anyone know what a VUP would be equivalent to? Does one VUP equal
>>> a 486 at 12 MHZ or 8 MHZ or what?
>>
>As you may already know, a VUP is a "Vax Unit of Performance", where
>"Vax" means a Vax 11/780. The only way I know to compare Vax's to a
>486 is via some SPEC benchmarks I have. Here are some relevant entries
>from a table of SPEC benchmarks. Note that by definition, a Vax 11/780
>is benchmarked at 1 SPECmark.
>
>--------------------------------------------------------------------------------
>System CPU Clk MHz Cache SPEC Info Source
>Name Type Ext/Int Ext+I/D Mrk89 Date Obtained
>------------- ----------- ------- -------- ----- ----- -------------------------
>VAX 11/780 VAX 1.0 1.0 Reference System
>i386/387 80386/7 33 64+0 4.3 1992 Intel
>i486DX 80486 25 128+8 8.7 1990 Intel
>i486DX 80486 33 0+8 11.1 1991 Intel
>i486DX 80486 50 256+8 21.9 Oct92 comp.arch
>i486DX2 80486 66 256+8 25.6 1992 Intel
>
>So in round number, a 33Mz 386 is 4 times faster then an 11/780. Of
>course, that's just raw computation, there may be things that an
>11/780 can do better then a 386 (for example heat a room!).
>
If you are going to post "silly" stats, how about comparing some of DEC's
larger offerings, such a 6610 or an Axp ?
--
Bill Bochnik | It's hard to be a James Bond in an Abbott
Systems Analyst | and Costello world.
Ciba-Geigy Corporation | Signed and sealed, they deliver oblivion.
philabs!crpmks!billb | It's worse than that, he's dead Jim.
Walter H Dick, III
will give you an approximate range. Let's not keep on mentioning that all
applications perform differently depending on their characteristics. I think
most people should already realize this. I/O bandwidth of a particular machine
is probably the most important characteristic after CPU speed. But something
like a Specmark will give someone an idea. In addition, if the application
is mostly CPU and little I/O, it will give you all that you need. Otherwise,
you can look to other benchmarks which also test I/O ability. I already know
that the PC architecture does not lend itself to high I/O throughput.
Walter Dick
whd...@cs.rit.edu