For awhile now I've been noticing that when I test Doom at different CPU speeds, there seems to be a wall of diminishing returns. If you look at Phil's Ultimate VGA Benchmark Database Project, you will see that except for a few of the very fastest processors, Doom seems to hit a wall at about 150 fps. So I decided to test this on some systems I have, ranging from a 200Mhz Pentium I to a 2.53Ghz Pentium IV, using MS-DOS 6.22. Obviously, each of these systems have different chipsets and bus implementations, so the results aren't ideal, but good enough for my purposes. So let's see what happens when we compare Doom to Quake I 320x200 with various graphics cards on a Pentium I 200, Pentium II 400, Pentium III 933, and Pentium IV 2.53Ghz.

First, let's look at Doom with PCI graphics cards.

The TNT2 M64 seems to scale well all the way up to the P3-933, but the rest of the graphics cards start petering out once they get to the P2-400.

Well, maybe it has something to do with the bus. I happen to have a TNT2 M64 in AGP, so let's see how the PCI bus compares to AGP.

It doesn't seem to have anything to do with the bus that the graphics card is on.

Let's see how Quake scales with PCI cards.

That's more like it! I also noticed that the TNT2 M64 loses its dominance in Quake, especially as you increase CPU speeds. Also, the two S3 cards run out of steam compared to the competition in the P3 and P4 systems, showing where the CPU bottleneck starts to relax its hold.

Let's compare PCI vs AGP in Quake.

Similar scaling, maybe slight better scaling with the AGP version.

Next post: Some AGP cards. A couple of things first. We can only look at the P3 and P4 systems as I don't have any AGP motherboards in slower speeds. Also, some of my AGP cards are keyed for 3.3v slots, so they don't have results in the P4 system.