Does OS/2 Benefit from New Hardware?

Computer hardware is undergoing rapid development and its capabilities increase every month. Video cards are becoming faster, CPUs continue to increase in speed, disk drives are larger and faster than ever, and system memory has become inexpensive enough that desktop systems usually have 512 Mb or more of memory installed. OS/2 users, however, often believe that they will not see real improvements in OS/2 performance with newer hardware due to a lack of specific driver updates for OS/2. Can this really be true? Sounds like an opportunity for some testing!

The Hardware

Last month, I described here a new Opteron 240 system that was built to run OS/2 and has hardware that is typical of what is currently being sold. It would be interesting to see if benchmark testing would show any improvement in the performance of OS/2 when running on the new Opteron new hardware compared with another Athlon XP 1700+ system containing typical two-year-old hardware. What makes a comparison of these two systems especially interesting is that both CPU's have an internal clock speed of 1.4 GHz so any differences in performance will be due to factors other than the clock speed of the CPU. Here is a summary of the two systems we'll use for our testing:

	System 1	System 2
Processor	Opteron 240	Athlon XP 1700+
Motherboard	ASUS SK8N	ASUS A7M266
Memory	512MB dual-channel DDR 333 Mhz	256MB DDR 266 Mhz
Video Chipset	Nvidia FX5200 with 128MB	Matrox G450 with 32 MB
Disk Drive	West. Dig. WD1200JB 120 GB ATA/100	West. Dig. WD40BB 40 GB ATA/100
IDE Controller	Nvidia Nforce3	VIA® VT82C686B

To begin our tests, we need to find a benchmarking program that will run on OS/2. The venerable Sysbench 0.9.4g is widely available and provides a pretty good range of tests which evaluate several specific performance areas. Another handy OS/2 video benchmarking application is the Clear and Simple Software VidPerf benchmark which you can download and try on your system as a comparison. Finally, we'll time some real-world file transfers between volumes as a check on our benchmark results. So, let's begin!

Video performance is central the look-and-feel that we experience on our systems as users so let's start there. System 1 has an Asus 9520 video card with an Nvidia FX5200 chipset, 128mb of memory, and support for AGP 8x while System 2 has a Matrox G450 video card which supports AGP 1x, 2x, and 4x and has 32 mb of onboard video memory. Based on these hardware features, we would certainly expect the video performance of System 1 to be much better than System 2, so what does Sysbench tell us?

Sysbench Graphics (million pixels/second)	System 1	Sytstem 2
BitBlt S-> S copy	531	187
BitBlt M-> S copy	336	19
Filled Rectangle	2219	842
Pattern Fill	2216	842
Vertical Lines	46	13
Horizontal Lines	1138	106
Diagonal Lines	22	23

Sysbench runs a series of graphics tests that show a clear winner here. System 1 is significantly faster in graphics performance on every test. Let's see what the Vidperf benchmark shows, though. Running Vidperf three times and averaging the results gives a result of 2.38 seconds.

Above is the image from the third test. Running Vidperf on System 2 gives a result of 7.63 seconds. It looks as though Vidperf supports the results that we saw on Sysbench and the video performance on System 1 smokes System 2 by a wide margin. System 1 is running the Scitech SNAP driver v2.2.2 while System 2 is running the Matrox drivers for OS/2 v2.58.144. Some of the credit for the video performance of System 1 must be given to Scitech for its SNAP driver package for OS/2. Scitech claims to be offering AGP8X support for OS/2 and, based on the results above, it looks like they are delivering on that. Here is the configuration screen for the SNAP drivers:

So, let's move on to file input/output operations. Both System 1 and System 2 are using ATA/100 IDE disk drives manufactured by Western Digital and both are running with the DANIS506.ADD IDE driver replace for IBM's IDE driver. What will we see in speed difference between the systems? Running Sysbench gives the following results.

Sysbench File and Disk Results	System 1	System 2
File I/O-marks, kilobytes/second	43113	21794
Avg. data access time, milliseconds	13.3	13.8
Cache/bus transfer rate, megabytes/second	71.8	83.0
Track 0 transfer rate forward, megabytes/second	47.5	46.5
Middle track rate forwards, megabytes/second	42.6	40.1
Last track rate backwards, megabytes/second	24.7	23.3
Average transfer rate, megabytes/second	38.3	36.6
Disk use CPU load, percent	3.1	5.2
64k sequential, cached, write, kilobytes/second	59887	39517
64k sequential, cached, read, kilobytes/second	62521	17686
64k random, cached, write, kilobytes/second	36154	25253
64k random, cached, read, kilobytes/second	86728	12131

As a check on these results, let's do a real-world test in which we copy 16 files, with a combined size of 47 megabytes, from one HPFS volume to another HPFS volume on the same disk drive. System 1 accomplishes this task in a speedy 5.15 seconds while System 2 requires 5.90 seconds to copy the same files, a 15 percent greater time. Based on the Sysbench results, System 2 is much slower on reads and somewhat slower on writes.

The general conclusion from this limited testing is that OS/2 appears to benefit much more than I would have expected from the hardware found in new systems, for graphics, file transfers, and disk operations.