Benchmark Programs for Older Systems



When I say older I mean pre-Pentium. Some will work on an XT or AT, but not all. While all programs should execute fine, the actual measurements taken will vary in usefulness, depending on the speed of the machine.

All these programs are best run under a clean boot in DOS. The only exception to this is when running tests on a drive controller, which may not perform at full speed without its driver. While many of these benchmarks are well-known and easily-obtainable, some are a little more obscure. A bunch of them (highlighted with a *) came from the BENCH_XT compilation, made available by user Ekb on Vogons and can be downloaded from here. I also used High Treason's benchmark pack (programs designated ^), which is here and includes some of the better-known tools to get you started. I have provided links for them individually where possible.

I have categorised each test into CPU & RAM, graphics and hard drive, though some programs cover all three categories. Where this is the case, I've put them in multiple categories and explained their functionality.


CPU & RAM

TOPBENCH 0.38^

Duration: <10s
The Oldskool PC Benchmark, written by Jim Leonard, is intended for live performance tweaking of a system. As the program runs, it measures the speed of the computer and provides a match to a similarly-performing system in its database. This is most useful for virtual systems such as DOSBox, so that you can adjust the emulator's performance to match a particular PC e.g. a 486DX2/66. It's also a great way to check that your real system is performing at a level expected of it. The score provided for the system is influenced by CPU, RAM and graphics performance with some 3D code tests. You can even send your results to Jim for inclusion in a future release.

Landmark System Speed Test 6.0^

Duration: <10s
I run this with the /B option to skip the splash screen at the beginning, which requires a keypress otherwise. This is from 1993 and tests pure computational speed i.e. CPU, FPU and CPU-to-video memory transfers. It attempts to displays a measurement of the actual CPU frequency, plus a live readout of the 'effective performance' of the CPU and FPU by comparing it to an AT system (i.e. a 286 equipped with a 287). I initially thought this program was useful for identifying the actual CPU frequency, but it thought my DX2 was running at 220.330MHz. Not very accurate, then. Because it's live, you can test the effect of a turbo button, for example. Scores from this version can't be compared to previous versions, however, as the metric was changed.

Norton System Information 8.0^

Duration: <10s
The 'CPU Speed' benchmark aspect of this program identifies the type and speed of the CPU and provides a measurement on its 'computing index' score. This is, unfortunately, not a particularly useful or intensive test of speed (at least according to Jim Leonard, whose amusing analysis can be read here), but could be useful if you're using the same motherboard with different CPUs.

MAXSPEED

Duration: <10s
The readme for this wasn't distributed with the version I have, but can be found here. Not that there's much to read - it simply describes the program as freeware and that it is 'extremely accurate' at measuring the actual speed of the CPU. Apparently this differs from the speed the CPU is rated for but I would say is more interesting than useful. It was written by MAXUM computers and can be found in the MAXSPEED folder on the linked shareware CD, or the BENCH_XT compilation.

Chips & Technologies MIPS

Duration: ~25s
I used to have a little program that actually gave you the millions of instructions per second score for a CPU and it was handy when testing the performance improvement when I overclocked my AMD 486DX4/100 to 120MHz back in the day. This is not that program, and it does not measure MIPS, despite the name. What it does do is more useful than most, as it measures a variety of performance indicators. The only downside of this is that there seems to be a bug whereby any CPU faster than a 50MHz 486 causes most of the results to be unreadable. There is a good description of the program here.

Memory Speed 3.1

Duration: <10s
These is a freeware program by Dietmar Meschede and is dated 1996. There is not much more info on it except that it performs write, read and move tests within RAM to provide a measurements in megabytes per second. This is done using chunks of data that vary in size from 1KB to 2MB resulting in 9 separate measurements. There are quite a few measurements taken, but I take only the first result for each operation. It also has a complementary Video Speed program, described below. It can be found within SPEED300.ZIP on the linked shareware CD, or the BENCH_XT compilation.

CABT*

Duration: <10s
Also known simply as Benchmark, this was written by Robert Stanton in 1992 and is intended specifically as a non-synthetic benchmark, as in it runs an actual program and measures how well that program runs. It measures the 'number crunching ability' of your computer, according to the utility's splash page, so only a change in the CPU will affect the score. This is another program that requires an FPU and the result is given in seconds. It also provides a 'performance index', comparing the result to some other well-known CPUs. I can't find a download link for this program, so it's best to get it from the BENCH_XT compilation mentioned earlier.

VIDSPEED 3.2*

Duration: <10s
As you can tell from the name, this tool is mostly used for measuring video performance, but it also tests RAM transfers. I don't know if it's related to the program of the same name with version 3.1, but they seem to differ significantly. Running the program without options provides a help screen but for this purpose I use the command VIDSPEED *. The program can be found in VIDSPD32.ZIP in the linked ISO, or the BENCH_XT compilation.

CACHECHK 4.0

Duration: ~50s
Written by Ray Van Tassle in 1996, this program is apparently postcard-ware. I don't know if Ray still lives at the address in the readme, but I feel like I need to send a postcard either way now because I've run this so many times. It performs memory access timing tests and is useful for all things cache-related. It will tell you how many caches you have, how large they are, and how fast they are, so it's useful for diagnosing issues. It's also particularly useful for situations where you are changing the cache in some way e.g. upgrading the CPU to one with more L1 cache, or upgrading the L2 cache on the motherboard, or changing the timing settings in the BIOS.

System Speed Test 4.78^

Duration: ~3m45s (with 512KB of L2 cache)
Aka Speedsys, this is possibly the most-recognisable, best known and most popular benchmark tool because it does so damn much and it's freeware. It was written by Vladimir Afanasiev and the list of features in the readme is very long indeed. I use this program for the CPU speed index and the cache and RAM transfer speed measurements (and the other tests, described below). There is an archive of previous versions here and the old homepage is here.

Navrátil Software System Information 0.60.45*

Duration: <10s
This is mostly useful as a system detection tool, but also has some rudimentary benchmark capabilities, providing a CPU and FPU score in dhrystones and whetstones. It also charts this score in comparison to other common CPUs and can provide CPUID info, amongst lots of other info.

CPU Identification Utility 1.23

Duration: <10s
CPUCHK was written by Jan Steunebrink and has had a somewhat long development arc; this version was released in February 2013, and version 1.0 released in December 1997. Most of the information this tool can provide comes from a CPU's CPUID register. Sadly a number of the CPUs I tested either do not support this feature, or CPUCHK was unable to read it. Programs like AIDA or NSSI can be tried as an alternative and there are online databases that can be used to cross-reference chip markings and photos. 1.26 is the latest version.

Graphics


Landmark 6.0^

Duration: <10s
The video performance aspect of this program measures RAM to VRAM transfers in characters per millisecond, so it's a text-based test rather than graphics-based, but provides an accurate measurement of bus transfers to the video card.

3D Benchmark VGA 1.0^

Duration: ~20s
Aka Superscape, this is a public domain utility created by New Dimension International and released in September 1990. It is based on their commercial SUPERSCAPE 'alternate reality' environment. An interactive version can be downloaded or run here, while the readme can be found here. It measures the FPS a system is capable of by rendering a 3D scene, and is influenced by the CPU type & speed, the cache, the video card bus width and the 'efficiency' of that card.

PC Player Benchmark 1.0^

Duration: ~30s
This utility is by Markus Oberrauter of German magazine PC Player (not to be confused with the English magazine of the same name), which was released for free on their cover CDs in late 1996. It works in a similar way to Superscape in that it renders a 3D scene and presents an FPS score at the end. It runs at a resolution of 640x480 by default but is usually run with the /VGAMODE switch for slower systems. It can be found in the CBENCH folder of the December 1997 Cover CD linked above, or the BENCH_XT compilation.

Doom 1.9^

Duration: ~90s
The benchmark mode of Doom is run with the command DOOM -TIMEDEMO DEMO3 (other demos are available), which renders a demo that is 2134 frames long. A framerate figure is reached by multiplying the frames by 35, then dividing the result by the realtics score. This score is heavily influenced by the speed of the CPU, RAM and graphics card.

Chris's 3D Bench

Duration: ~30s
Sadly this program's original web page has not been archived so it's difficult to find out more about it, but the first reference I can find to it is from 1998. It runs a 3D, environment-mapped routine (not dissimilar to the DirectX cube test) to give a result in FPS. It is affected by CPU, FPU memory and video speed and requires an FPU to run. This suggests the program is from a time post-486, when all desktop CPUs had an integrated floating point unit. There are separate executables for VGA and SVGA modes, though I only use the former.

Video Speed 3.1

Duration: <10s
This program, which comes with the Memory Speed utility (above), runs tests that measure the speed of transfers from main memory to video memory over 8-, 16- and 32-bit bus widths, producing scores in MB/s and FPS, the latter being the only score I recorded. As with Memory Speed, it can be found within SPEED300.ZIP on the linked shareware CD, or the BENCH_XT compilation.

VGAFPS 1.46*

Duration: <10s
I can find no info on this program at all, and it doesn't come with a readme, but it appears to transfer data from CPU registers into video memory, providing a measurement in KB/s and an equivalent 'frame rate'. Run the command VGAFPS to output the result to the screen, though you might need to run it multiple times to get a proper result as it can be inconsistent. Can't find an individual download anywhere so find it in the BENCH_XT compilation.

VESATEST

Duration: <10s
A program written by Vogons user RayeR aka Martin Rehak and requires an FPU. It identifies the VESA VBE version of the graphics card and the resolutions available, plus the amount of video memory (VRAM) available. It then measures transfers between the LFB (linear frame buffer) and VRAM, providing an FPS score, plus a measurement of the transfer speed in MB/s. It can also measure the current refresh rate.

VMax256 1.11*

Duration: ~20s
This program from 1996 appears to be by a Russian only known as Nomagic. It tests the throughput of different video modes in 8-, 16- and 32-bits. You can measure from 320x200 up to the max resolution of your card, but I just stuck to basic VGA. The results are provided for each bit width and are in FPS (in brackets). It can also test the RAM speed but I discarded this result. I use the command VMAX256 /x400 /x480 /x600 /x768 /x1024 /c200


VIDSPEED 3.2*

Duration: ~15s
Measures read and write speeds for video RAM and normal RAM in KB/s at 4-, 16- and 256-colour modes (CGA, EGA and VGA). Running the program without options provides a help screen but I usually use the command VIDSPEED A J L which tests CGA, EGA and VGA. Other modes are available. The program can be found in VIDSPD32.ZIP in the linked ISO, or the BENCH_XT compilation.

3DSPACE 1.2*

Duration: ~40s
There's no documentation for this but it renders a texture-mapped, 3D environment not unlike Wolfenstein 3D, but more brown. At the end it provides a percentage score of how 'ideal' your computer is, where 100% is the 'most ideal'. A table of comparable results is provided at the end of the benchmark. This is another one I can't find a direct link for, so find it in the BENCH_XT compilation.

Morph 3D*

Duration: ~30s
Written by Courtney S. Sharp in November 1993, this tiny program written in assembler renders some nice 3D shapes with rotation and zoom effects and provides a FPS score at the end. It runs indefinitely so I just quit it after 30s, though this an arbitrary amount. Get it from the BENCH_XT compilation.

Wolfenstein 3D*

Duration: ~20s
Yes, although the benchmark feature of Doom is widely known, Wolf 3D (specifically the WOLF_286.EXE version linked above, or the BENCH_XT compilation) also has such a feature if you run it with the TIMEDEMO option. The results given are much the same as with Doom (ticks and frames), except it calculates the FPS for you.

System Speed Test 4.78^

Duration: <10s
Speedsys can provides a VESA memory transfer measurement in KB/s. If that's the only measurement you want, you can just cancel further tests as soon as the program loads.

Hard Disk


Norton System Information 8.0

Duration: <10s
Aka Sysinfo, this program has a few useful benchmarks and can measure drive performance. It's worth mentioning that, if you have a caching controller or something SCSI, you will get a message saying that the results will not be accurate. This is because the controller blocks the 'drive seek' test, so it comes out as zero usually. It does measure the data transfer rate in KB/s either way. It also provides a 'Disk Index' metric, but I don't know how useful that is yet.

QDI Mark

Duration: <10s
This tool was included with drivers for QDI hard disk controllers and measures data transfer rate (KB/s) along with the average and track-to-track seek times.

Core Disk Performance Test 2.7*

Duration: <10s
This is an older tool from 1986 and provides hard disk information plus the usual performance measurements. It also provides a 'performance index' for fun. Obtainable from the BENCH_XT compilation.

Disk Meter 1.3*

Duration: <10s
By Procom Technology from 1989, this gives more extensive performance info. Rather than a basic transfer rate, it can measure random, sequential and buffered measurements in KB/s, plus average, maximum and track-to-track seek values in milliseconds. My testing has shown this program to be unreliable on faster disk subsystems, as the variance for the buffered transfer rate measurement can vary by 50% or more.

4 Speed 1.0*

Duration: <25s
This rather fancy program, by Dustbowl Designs from 1993, draws a live graph on screen as it measures the transfer rate of the disk. This makes it much easier to show the performance quirks of cached disks, for example, as an average will only give one figure, while the graph shows the point at which performance drops off. The graph can be saved as a PCX for later analysis and be represented as a line- or scatter-graph. Find it in the 4SPD100.ZIP file on the linked ISO, or the BENCH_XT compilation.

System Speed Test 4.78

Duration: ~15m (2GB drive)
Speedsys has the most extensive set of hard drive tests, with a short or long test suite available. I always use the 'short' test, which takes bloody ages anyway so I'm disinclined to do the long one. It does provides geometry info, access time, plus min/max/avg values for verify speed, linear read speed and buffered read speed. For some drives it can even show the RPM. It also provides a 'speed index' for the storage subsystem. There does appear to be an issue that I've noticed with the linear read speed test (the yellow line), which can appear to get near the end but never complete. It's safe to press escape to end the test and still keep the readings up to that point.

History


1/1/23: First version