cpa@monu6.cc.monash.edu.au (Christopher P Avram) (12/14/90)
I have seen a lot of talk on this group about the speed of
Windows 3.0 so here are some tests.
Question. Is 256 colour mode much slower?
Question. Is standard mode much faster than 386 enhanced mode?
To understand these tests find and run WINBench.EXE.
Its a PC Magazine program.
My conclusions are that for this particular card, enhanced
mode costs very little in video performance (perhaps it does
slow CPU bound processes this test didn't test that).
There is a performance cost for 256 colour mode on this video card.
The card is very much slower in area fill operations in 256 colour
mode. Screen re-paint is not always an area fill type operation.
The program used to do these tests is a windows 2 program that
seems to work in windows 3. I plan to publish tests for a TVGA
clone and a Paradise VGA Plus clone real soon now. The tests
are only comparative between modes not between video cards.
If you want to know which is the fastest video card, this series
of tests wont help.
--------------------------------------------------------
Windows 3.0 Video Speed Tests
Run by: Christopher P Avram,
Department of Computer Technology,
Monash University Caulfield Campus.
Run date: November 14, 1990
Using: PC Magazine DispInfo Version 1.2
by Charles Petzold 1986
Processor: 386 AT Bus clone; No co-processor;
8 Mbytes Memory; no cache; 25MHz
Video card: a Tseng Labs 3000 clone
Video Memory: 512 Kbytes
Video Bus: 8 bit
Shadow Ram enabled.
Video Drivers: From cica.cica.indiana.edu and Microsoft
Windows 386 Enhanced mode
Driver Line Rectangle Ellipse BitBlt StretchBlt Scroll Fill
VGA 640x480x16 3.9 11.0 53.8 5.5 79.6 6.6 44.5
Tseng 640x480x16 3.8 11.0 49.4 6.0 81.3 6.7 44.8
Tseng 640x480x256 5.5 43.4 104.9 6.6 92.3 7.4 287.7
Tseng 800x600x16 3.8 13.2 70.9 7.7 104.4 8.8 53.4
VGAfont 800x600x16 3.8 9.9 50.5 5.5 76.4 5.7 33.3*
Tseng 1024x768x16 3.9 17.0 70.9 7.1 108.8 9.3 80.9
Windows Standard mode
Driver Line Rectangle Ellipse BitBlt StretchBlt Scroll Fill
Tseng 640x480x256 5.0 42.3 86.2 6.0 86.8 7.3 282.0
Windows Real mode
Driver Line Rectangle Ellipse BitBlt StretchBlt Scroll Fill
Tseng 640x480x256 3.8 41.7 67.6 6.0 86.8 6.6 278.6
All data in seconds per 100 operations.
* NOTE: The test is performed in a window whose size is
set to 60 system font characters wide and 20 system font
characters high. The Tseng drivers install by default using
a large system font (8514), so area based operations operate on
a large screen area. The fonts available are for 640x480 or
1024x760 but 800x600 is between. When in 800x600 mode, I have
run the tests with both system fonts.
-------------------------------------------------
Chris Avram
Department of Computer Technology
Faculty of Computing and Information Technology
Monash University Caulfield Campus
PO Box 197 Phone + 61 3 573 2196
Caulfield East Vic 3145 Fax + 61 3 573 2745
AUSTRALIA email cpa@monu6.cc.monash.edu.ausystem@infopls.UUCP (SYSOP) (12/17/90)
cpa@monu6.cc.monash.edu.au (Christopher P Avram) writes: > I have seen a lot of talk on this group about the speed of > Windows 3.0 so here are some tests. > > Question. Is 256 colour mode much slower? > Question. Is standard mode much faster than 386 enhanced mode? > > To understand these tests find and run WINBench.EXE. > Its a PC Magazine program. > > My conclusions are that for this particular card, enhanced > mode costs very little in video performance (perhaps it does > slow CPU bound processes this test didn't test that). > There is a performance cost for 256 colour mode on this video card. > The card is very much slower in area fill operations in 256 colour > mode. Screen re-paint is not always an area fill type operation. > > The program used to do these tests is a windows 2 program that > seems to work in windows 3. I plan to publish tests for a TVGA > clone and a Paradise VGA Plus clone real soon now. The tests > are only comparative between modes not between video cards. > If you want to know which is the fastest video card, this series > of tests wont help. > > Chris Avram I compared Chris's numbers with what I've determined from testing on my own system (386/25 w/ 64K cache, ATI VGAWonder in 16-bit mode, 4M RAM.) If you use the stock 640x480x16 VGA driver as a basis, and compute ratios, my numbers come very close to his. The biggest difference I saw is that for 640x480x256, my fill was nearly twice as fast. (3x slower vs 6x slower). I changed drivers by simply changing the driver name in the SYSTEM.INI file. I made no other changes. If anyone else does similar testing, you should try to do a few things to keep things comparable. Winbench and Program Manager should be the ONLY programs loaded. Program Manager should be minimized. Move the mouse OUT of the Winbench window, and use the keyboard. (Having the mouse in the window, or moving it while testing, can have a MAJOR impact on the numbers.) I also left the window in the upper left corner of the screen. I don't know if moving it will have any effect. --------------- Andrew Rossmann andyross@infopls.UUCP or ..!uunet!ddsw1!infopls!system Infoplus Support BBS +1 708 537 0247, 1200/2400, 24 hours
rob@pcad.UUCP (Ralph Brown) (12/19/90)
I've been looking through the samples in the DDK and running some timing test of my own and from what I can see, the drivers are the key to Windows speed. In particular the EGA/VGA driver code that comes with the DDK seem very general but at the cost of speed. Has anyone seen any drivers that are written to optimize speed of drawing (perhaps at the cost of generality)? Ralph
mitsolid@acf5.NYU.EDU (Thanasis Mitsolides) (12/21/90)
> Windows 386 Enhanced mode > > Driver Line Rectangle Ellipse BitBlt StretchBlt Scroll Fill > Tseng 1024x768x16 3.9 17.0 70.9 7.1 108.8 9.3 80.9 > > All data in seconds per 100 operations. I run program bench30.exe (version 1.20) from PC Magazine (chyde.uwasa.fi:pc/mswindows/win_bnch.zip) on a 386SX/16 with a paradise 16 bit card and 512K of RAM. I use 8514/A fonts so the test area is as large as in the above test. It seems to me the above test results are not in "seconds per 100 operations". My results in the same units are on the average 10 times smaller numbers (except of scroll where they are 3 times larger). For that reason I report my results in miliseconds (Msec) per operation. (this is the last number z in the report (ie x operations y seconds z Msec each)) This is a more absolute value. Driver Line Rectangle Ellipse BitBlt StretchBlt Scroll Fill Paradise 1024x768x16 7.7 18.7 104.4 29.7 164.0 29.8 58.1 Thanasis ------------------------------------------------------------------------------- Internet: mitsolid@cs.nyu.edu (mitsolid%cs.nyu.edu@relay.cs.net) UUCP : ...!uunet!cmcl2!cs!mitsolid ------------------------------------------------------------------------------- -- ------------------------------------------------------------------------------- Internet: mitsolid@cs.nyu.edu (mitsolid%cs.nyu.edu@relay.cs.net) UUCP : ...!uunet!cmcl2!cs!mitsolid -------------------------------------------------------------------------------
cpa@monu6.cc.monash.edu.au (Christopher P Avram) (12/21/90)
In my posting of Date: Fri, 14 Dec 90 13:40:19 GMT on this subject, I quoted times said to be in seconds per 100 operations. I was wrong. They are in Msecs per each operation. My thanks to Thanasis mitsolid@SPUNKY.CS.NYU.EDU for pointing out my error. Chris Avram.