Anyone know of a comparative DOS speed tester software?

February 22, 2009 10:36 AM

I'm looking for a DOS-based CPU speed tester that I remember seeing years ago. It told you how your machine compared to various pre-measured setups. I don't remember how it presented the speed (MIPS, maybe?), but there was a large table of CPU types (386SX/25MHz, 486DX/50MHZ etc.) that it compared the measured result against. Anyone know of such a program?

Specifically, I'm looking to use the software inside DOSBox to get a ballpark figure of how many "emulation cycles" are needed on my host machine to correspond to various historical CPUs type. I would then write these settings down to create several DOSBox configs, one for each type of machine.

The software wasn't a CPU ID solution (it might have made a guess, I don't remember), but an app to specifically see comparisons to other hardware. CPU and other hardware identification would be rather nonsensical inside DOSBox.

The benchmark/speed test results are bound to be rough and inexact given timing discrepancies and other emulation quirks, and need to be fine-tuned inside the actual games, but a tool like this would still be nice to have.

I remember that the software (or one of many, I have probably tested more than one) had a real-time "ticker" and a sliding scale. I vaguely remember testing one of these inside DOSBox many years ago, and recall seeing the sliding scale move as I adjusted the cycles setting. Naturally, I've lost and forgotten the software since.

Doesn't matter if the software is free or not. I've googled up some things but not the one I remember seeing. Any help will be appreciated.

posted by lifeless to computers & internet (3 comments total)

Hmm, found some interesting stuff here shortly after posting:

http://www.eunet.bg/simtel.net/msdos/sysinfo.html

Still open to good suggestions, though :-)
posted by lifeless at 10:46 AM on February 22

It sounds kind of like Norton Utilities SI (System Info). Some of the later DOS versions had pretty baroque IBM line drawing character set displays. If I remember right, the rating would go down when I moved my mouse.
posted by rfs at 10:50 AM on February 22

There are several problems with this plan. Those really old DOS CPU benchmarks usually just consisted of measuring the time it takes to perform a set number of repetitions of a busy-loop. But for something like DOSbox (at least on the same architecture) this is not costly to emulate at all and it can run nearly at native speed. But what costs greatly to emulate are syscalls and virtualizing access to hardware, neither of which are performed in a simple busy-loop. So you're not really measuring what you need to be measuring.

The other problem is that a simple dumb spin-loop on modern hardware probably fits entirely into L1 cache, so its execution rate isn't indicative of actual performance with a real workload that involves cache misses -- those ancient 286s and 386s were much less sophisticated.

I'd say your benchmark should be actually playing those games under emulation, i.e. the real workload, not an extremely naive benchmark.
posted by Rhomboid at 8:29 AM on February 23