Here’s the second, and probably last, part of my analysis. In the first part I took a look at synthetic benchmark scores, but let’s face it, that’s not much of a real scenario. Sure it’s nice to get a single result and even to take a look at filters quality, but in the end, people really want to know how games actually run on different cards. So let me introduce our two contestants – Quake 2 and Forsaken.
Quake 2 is the grand-daddy of all OpenGL benchmarks. There was GLQuake and a few other games before it, but those usually relied on MiniGL drivers and other unsightly measures. With Quake 2 you got colored lighting, multitexturing, and everything else you could ask for. Not to mention, it’s lighter in comparison to newer games, so even less powerful cards can run it somewhat. As long as they have OpenGL support.
Forsaken was one of Acclaim’s most average games. A sort of Descent clone, it wasn’t horrible like that BMX adult game, nor great like ShadowMan or Turok. It was utterly unremarkable, so much that nowadays only benchmarkers like me remember about it. Only the graphics were remarkable enough, especially all those lights. Again, it’s a fairly lightweight game that even most old cards can run somewhat.
So let’s see our results, and a few comments.
Unlike 3DMark, the numbers here speak for themselves. But there are still a few things to note.
– As you can easily notice, the GTS is tremendously bound by the low processor speed. I have been thinking about upgrading to something better, but I’m afraid it would help the older cards a bit too much. And we don’t want that, do we? Even so, Nvidia’s old flagship easily crushes every other card, especially at higher resolutions. And it should, since it came out at least one year later.
– The Savage 3D is really only playable at 640×480, and that’s with the latest beta driver: the stable release was far slower, and produced black screen flashes.
– The Voodoo 3 drops quite a bit at 1024×768. I’m afraid I was using regular OpenGL there, because for some reason, 3dfx GL made the game crash at that resolution.
– The G200/G250 have a big problem here: they run well for the first couple minutes, but soon, as you keep repeating tests, their framerates drop dramatically. These tests are based on the first benchmark each, to get the best result across the board, though it might be inaccurate for real gameplay scenarios. It might be throttling, but it doesn’t seem to happen in Direct3D, so I’m more convinced it’s about buggy OpenGL support. After all, if it were due to high temperatures, the G200 should resist a lot longer (it’s the same chip as the G250 but far lower-clocked).
– The G550 is obviously Vsynced at 60fps. I couldn’t disable it, no matter what I tried. So it’s not much use. The TNT2 M64 was apparently Vsynced at 75fps, too.
– Luckily, all of my cards with OpenGL support had a tleast 8MB of video memory, so every test could run at up to 1024×768. Direct3D wasn’t so easy, as you’ll see.
A few more comments, because things here are somewhat more complicated.
– All tests are run with 5650 color space. All except for the Trio 3D, which is stuck with 5550, and churns out some pretty unsightly color banding.
– Of course, 4MB graphics card won’t go above 800×600. The old Virge can’t even go above 640×480, and that’s with a whole lot of blank textures and other issues – I could only get a decent picture at 512×384, and I didn’t want to include that resolution here since the Virge was the only card that required it. Instead, you get a less accurate score. Sorry.
– I said I was using beta drivers for the Savage 3D, which helped with Quake 2. Well, they also break Forsaken, sort of. It runs, and quite fast too, but ost of the time you’ll get black screen flashes. For the sake of consistency, I didn’t change drivers. I can tell you however, that using older drivers does make it better, but at the expense of speed. As I couldn’t even disable Vsync on the other drivers, I chose not to show them here.
– The TNT2 M64 result is weird because my screen seemingly had different refresh rates at different resolutions. The game was capped anyway, and I couldn’t find a way to disable it, so there isn’t much point in changing things.
– Again, the GTS is obviously CPU-bound. But in this case, as we have also seen in 3DMark, it’s enough to let the G400 and Voodoo 3 surpass it (the Savage 3D doesn’t really count because it had rendering errors), at least on lower resolutions. What could have caused this issue? I have had similar problems with other Direct3D games, for example Blood 2, and yes, I know that game doesn’t count for much. But still, I don’t have a clue.
– The G100 gets above the Mystique, and does so with bilinear filtering enabled. If clocks are indeed lower, then it must have been an improved architecture rather than a rehash with a few features tacked on. Speaking of the Mystique, it’s surprisingly playable at 640×480. Of course you’ll have to deal with alpha stippling and no filtering. The Nintendo 64 version would be better, really. If you can deal with 240p and low-res textures.
– The G550 is synced, while the other Matrox cards aren’t. Damn drivers. Anyway, with no multitexturing in Forsaken, there’s no reason why the G550 should be any faster than the G400. In fact, it would probably be slower due to its DDR memory.
– The Rage 128 results are especially bad. I think it’s capped, but so was the Rage LT Pro, and we still get only slightly better framerates. Oh, and the less said about the Rage IIc, the better.
What’s next? Not much, I think. But just yesterday I found a cheap SiS 305, so I’ll be testing that soon.
In truth, I would have liked to run a few more benchmarks on these cards. Some choices wouldn’t be so good (MDK2 is too heavy for older cards, Expendable takes forever to clear a run, etc). But others would be okay, for example Turok. Maybe one day I’ll look into expanding these results.