32-bit ST/V outruns 32-bit ST80 on Intel 
Author Message
 32-bit ST/V outruns 32-bit ST80 on Intel

+++++++++++++++++++++++++++++++++++++++++
+ Use monospaced font to view this file +
+++++++++++++++++++++++++++++++++++++++++

Summary: ST/V-PM is 1.5 to 3 times faster than ST80-Win on two benchmarks.
Althougth not substantial enough to draw strong conclusions, they suggest that
for an Intel machine Digitalk's 32-bit OS/2 version is the performance leader.
------------

Last month Marten Feldtmann and Mark Nouwen posted two benchmarks for ST/V-PM
and ST/V-Windows. Here are the same benchmarks for VisualWorks-Windows 1.0.
This is ParcPlace's superset of Smalltalk-80 that includes an interface
builder and other extras.

The benchmarks were a while loop and a recursive fibonacci number generator.
Both ended up generating numbers which were large integers on 16-bit machines
(100000 while repetitions; 1346269 as '30 fibonacci'). This caused Digitalk's
16-bit Smalltalk versions to perform badly.

However, Digitalk's 32-bit version whomps ParcPlace's. Possibly one reason is
that ParcPlace runs fastest with 'clean' blocks and slowest with 'copying' and
'full' blocks. Neither test used clean blocks.  Perhaps Digitalk doesn't make
this distinction and doesn't suffer from the performance overhead associated
with copying and full blocks. I'm just speculating, since I'm not familiar
with Digitalk's technology. I doubt this could fully account for their
advantage.

Ideally all tests should be done on the same machine. Unfortunately, they were
done on three different machines, but at least all were 486/33s. It will be
interesting to benchmark ParcPlace's OS/2 implementation when it is released.

Hardware 486/33 (Feldtmann, Samuelson 16MB; Nouwen 8MB)

                                       word       while loop      fibonacci
                                       length     milliseconds    milliseconds
tester     vendor     version          (bits)*    (avg result)    (avg result)

Nouwen     Digitalk   ST/V-Win 2.0     16         3570            32960
Feldtmann  Digitalk   ST/V-PM 1.4      16         2530             9503
Nouwen     Digitalk   ST/V-PM 1.4      16         2530             9470
Feldtmann  Digitalk   ST/V-PM 2.0      32          125             4673
Nouwen     Digitalk   ST/V-PM 2.0      32          120             4650
Samuelson  ParcPlace  VW-Win 1.0       32          353             6909

* These are my assumptions about the underlying word length in the virtual
machines. Although VW-Win runs on a 16-bit operating system (Win 3.1), it is a
32-bit implementation because it is compiled with a 32-bit DOS extender.

1-----------------------------------------------------------------------------

Newsgroups: comp.lang.smalltalk
Subject: Speed of Digitalk ST-V/PM 2.0
Date: Wed, 02 Dec 92 16:32:48 GMT
Organization: Private OS/2 System

 I made two little simple tests to check the speed of Digitalk ST/V-PM:

test machine: 486/33 with 16MB RAM

1.) program

| anIndex |
Time millisecondsToRun: [
  anIndex := 100000.
    [anIndex > 0] whileTrue:[ anIndex := anIndex - 1]]  

results ST/V-PM 1.4:    2530 2470 2560 2560
results ST/V-PM 2.0:     130  120  130  120

2.) program

fib (in class integer)

  self > 1
    ifTrue:[ ^((self - 1) fib + (self - 2) fib ) ]
    ifFalse:[ ^1]

calling: "30 fib"

results ST/V-PM 1.4:    9590 9570 9380 9470
results ST/V-PM 2.0:    4650 4690 4750 4600

 That's all .... :-)

 MF

--
Marten Feldtmann,     Kieler Str. 29,       2300 Kiel 14,    Germany

2-----------------------------------------------------------------------------

Newsgroups: comp.lang.smalltalk
Subject: Speed of Digitalk ST-V/Windows 2.0
Date: 3 Dec 92 10:27:46 GMT
Organization: Mediasystemen, Netherlands

As respond to the article of Marten Feldtmann here are the result
of the same pieces of code running Smalltalk-V/Windows 2.0 on
a 486 33 Mhz with 8 Mb.

(I am a little ashamed but here they come:)

PM 1.4   2530
PM 2.0   120
WIN 2.0  3570

Second test (fib)

PM 1.4     9470
PM 2.0     4650
Win 2.0    32960   (No really 5 numbers, no typing error)

Hopefully Windows NT will speed thinks up other wise I will have to buy
another 8 Mb and start running OS/2 (which I will also need to run
Windows NT so whats the question ??).

         /                                  Mark Nouwen                        
         |       Take a byte              Mediasystemen              
     /--------\    into it.                  Holland                  
    |  AdLine  |                        sun4nl!media01!mno          

     \___++___/                        Fax.:  +31 23-315210

3-----------------------------------------------------------------------------
Bruce Samuelson's tests on a 486DX/33 16MB, VW 1.0, using Win 3.1 and DOS 5.0.

| anIndex |
Time millisecondsToRun: [
  anIndex := 100000.
    [anIndex > 0] whileTrue:[ anIndex := anIndex - 1]]

353  354  362  343

fib (temporarily compiled in class Integer)

  self > 1
    ifTrue:[ ^((self - 1) fib + (self - 2) fib ) ]
    ifFalse:[ ^1]

Time millisecondsToRun: [30 fib]

6906 6910 6914 6905
--
**********************************************************
* Bruce Samuelson       Department of Linguistics        *

**********************************************************



Thu, 06 Jul 1995 08:53:38 GMT  
 
 [ 1 post ] 

 Relevant Pages 

1. Division 32-Bit/32-Bit with 16-Bit Register

2. Building 32 bit DLL from 32 bit EXE

3. 32 Bit ASM from 32 Bit Windows COBOL

4. 32 Bit ASM from 32 Bit Windows COBOL

5. Calling 32-bit FORTRAN dlls from 32-bit VB

6. top 32 bits of 64-bit product of two 32-bit integers

7. Absoft ships 32-bit and 64-bit IMSL Fortran libraries for MacOS/Intel

8. 32 bit ST communicating with 16 bit VB

9. Graphics Error 16-bit vs 32-bit

10. 16 bit VS 32 bit

11. 16 BIT VS. 32 BIT

12. 16 bit Vs 32 bit difference?

 

 
Powered by phpBB® Forum Software