Some time ago, I got very bored.
Boredom is a very powerful thing with me. Becuase when I have nothing else to do, I get strange ideas, often times ones that only appeal to myself. Such is the nature of neurodivergence, I suppose. But in this particular instance - apparently the 30th of April, 2022, going by the date stamp of this image - boredom became Science, in the Adam Savage sense. ("The difference between screwing around, and Science, is writing it down." Or however that went.)
The nature of this boredom led me to run Quake benchmarks on two different computers at once, and attempting to gauge how many times the fast machine could complete demo1
in the time it took the slow machine to do the same. The two computers are in no way equivalent: One is my Toshiba Libretto 70CT, alias Scarface One via its previous owner. (I don't like to change the names of computers if their owners have given them any - it's like a maritime superstition but with computers.) It is a 120 MHz Pentium MMX; I was unaware that Pentium MMXes came that slow. The other, faster one is a 2009, 27-inch iMac, alias Aegir. It has 12 years worth of advances in it. The ultimate figure I came to, entirely by vibes, was 11 demos per demo. In other words, Aegir can finish demo1, eleven times, in the time it takes Scarface One to do the same, despite the vast differences in screen resolution, hardware acceleration, and underlying OS.
So I did what any good scientist did and wrote it all down. And then I ran more tests and wrote those down to. Which gave birth to the silliest spreadsheet I've got access to. And you can contribute to it, too.
You may observe my data on The Spreadsheet here.
Ultimately, my goal with this spreadsheet is to find the most hilarious edge-cases in the process of benchmarking the original Quake, and eventually figure out how to compile this data into a CPUBoss-like interactive comparison chart. I'm not very good at that kind of scripting, even though I'm fairly sure Google Sheets is capable of it. But more important than the scripting is the data, and boy have I got a lot of data. Everything from ultra-powerful Ryzen gaming rigs to an emulated 386 with a co-processor, from a Steam Deck to an AliExpress special emulation handheld, from a classic G3 Macintosh to an Amiga 500 whose brains have been replaced with a Raspberry Pi. And I feel like I'm far from done with this, either.
Want to submit your system to this project?
Cool! Well, first of all, install whatever Quake port your system is most comfortable with. I generally target near-vanilla ports (QuakeSpasm appears most frequently; QuakeSpasm Spiked, Ironwail, TyrQuake, or even old builds of TomazQuake would work just as well!). I am not being scientific with graphics settings At All; for LCDs, I've been going with the monitor's native resolution, while for DOS-compatibles, I've been sticking with 320x200 as default. Pick whatever resolution gives the thing the best fighting chance - or go as absurd as possible and try to bog your machine down as badly as you can.
I am always looking for more bizarre and absurd hardware to put through the Quake test. Please get in touch especially if you have something absurdly slow. The PCem 386/387 is theoretically as slow as it is possible to get, but I consider that a tool-assisted effort - I'd love to see it happen on a real 386!
Since I first unveiled the sheet to Cohost a few years back, it gained few nice new additions, thanks to a local friend of mine. We picked at each other's brains a bit until we came up with a nicer way to organize some of the data, and a way to factor all of that into a newer, abstracted Performance Index (inspired by Forza Motorsport).
The Performance Index, in short, multiplies the screen resolution (X pixels * Y pixels) and the bit depth (divided by 8, so 8-bit is 1, 16-bit is 2, etc.), then divides by the average run time, and then cuts the value in half if it is using a hardware renderer - then normalizes the whole thing by square-rooting the whole thing. I don't have any particular reason for building the formula that way, just that the numbers look a bit nicer and less absurd further up the scale. (And then I drop the decimal places for sake of it looking cleaner.)
Now, sure, this is still not exactly a professional benchmark. 3DMark and Cinebench, I am not. But now I have a slightly nicer metric to compare modern systems against older ones, because 60 FPS in low-res software mode is a very different beast than 60 FPS in 4K OpenGL.
If you want to see your machine in here, instructions are in row 1 of the sheet; I just need 4 runs of timedemo demo1
and the full hardware specs of your machine. I'm especially looking for any non-x86 machines if anybody has any wacky old ones to submit, I know there's Quake for SGI, and I'd really love to get, say, a real Acorn RISC PC or a Transmetal Crusoe-based machine in here. Get in touch with me via whatever method of contact suits you and I'll happily work with you to get your machine chronicled in the hall of heroes. ...Or something.