PM Benchmark Results for 1024x768 Device Drivers
Article Info | |
---|---|
Author | Mike Thompson |
Date | |
OS Version | |
Companion File | N/A |
Source | svgabn.zip |
I don't know how many people actually tried the 1024x768x256 TSENG device driver in the 10/92 service pack, but I did. Although it looked good, and it worked O.K. on my Diamond Speedstar board, I was pretty disappointed with its performance.
I have been using a 1024x768x16 ET4000 device driver for some time now, and have gotten used to its good performance. Unfortunately, it only partially works under OS/2 2.0. Certain operations don't work, like the fills for the tabs on the settings notebooks. So I am kind of stuck. Do I go back to 640x480 (plain old VGA), put up with a broken driver, or bear with a sluggish driver?
Well, I would like to use a faster 1024x768x256 color driver if possible. It doesn't have to be as fast as my 16 color driver to make me happy, just close.
To help the developers at IBM get a little more interested in this driver issue, I have performed a benchmark study using an old dusty copy of PMBench, written by GSS for PC Labs back when PC Magazine was still interested in OS/2 (November, 1989).
The results are very interesting. I have uploaded a file named SVGABNCH.ZIP to library #15 of the OS2DF2 forum. This zipped file contains a 123-G spreadsheet file, as well as an HPGL file which can be copied to an HP LJ III printer (with enough memory for full page graphics). Just use the COPY /B command to copy the .PRN file extracted from the zipped file. The result will be 8 pages of benchmark data to chomp your teeth on!
I encourage you to download the file and view it if you are contemplating 1024x768x256 on a TSENG board, but for those who are not interested in all of the details, here are the highlights.
- Good news
- Many operations are faster, while still more are just a little bit slower (20 to 30 percent). The power of 32-bit can really be seen!
- Memory to memory BITBLT operations are all faster!
- Arcs, polysplines, polyfillets are all faster!
- Bad news
Often used operations are significantly slower!
Aligned BITBLTs 200 percent (3x) slower. Inversions 400 percent (5x) slower. Horizontal lines 100 percent (2x) slower. Default pattern fill 600 percent (7x) slower. Overpaint Text Draws 400 percent (5x) slower. Window scrolling 300 percent (4x) slower. Window erase 600 percent (7x) slower.
If anyone at IBM is working on optimizations to this driver (and I desperately hope they are), this benchmark study should be very useful in pointing out areas where improvements should first be attempted. I would be very happy to be a beta tester for a new driver. I would also be glad to collect data for a new driver and add it to my study.
Just send E-Mail. I check at least twice a week.
Mike Thompson - Graduate Software
This file was also E-mailed directly to the following IBM representatives:
- 72360,3554 Kelvin Lawrence
- 72400,2740 Steve Woodward
- 71075,2551 Sam Casto