Introduction
The article first appeared in .
The German game developer Massive drew a great deal of attention to itself during the early stages of NVIDIA’s GeForce3 launch with its game, . NVIDIA had announced a developer contest of sorts, which called on developers to create demos and benchmarks to show off the GF3’s capabilities. Of course, the main requirement for these programs was that they should utilize the chip’s pixel and vertex shaders. Massive entered the fray with its own benchmark, , facing competition in the form of , and , among others.
The game engine of the final retail version of Aquanox differs from AquaMark’s in some respects, though. For example, there are no longer any options to change the settings for vertex and pixel shaders. When asked about this, Massive replied that the AquaMark engine is completely identical to the one used in Aquanox and uses DirectX 8 to its full extent. Considering the state of the Krass engine (yes, that’s what Massive’s 3D engine is called) at the time, I find this claim rather questionable.
Introduction, Continued
The following benchmarks show how current video cards perform in the actual retail version of the game (v1.17). There is a save game available at which I used as the base for a time demo run. As in Max Payne, the final scores have to be recorded manually by reading them from the screen at the end of the time demo. Since this method is obviously not as exact as getting a final score or frame rate reported, inaccuracies of +/- 0.2 frames are possible.
Pictures from the benchmark sequence. The fire causes the frame rate to sink dramatically. Screenshots were taken on a Ti500 at 800×600-32 with 2 x FSAA.
Test Setup
| Intel Hardware Socket 478 |
|
| Processor | Intel Pentium 4/1800 MHz (400 MHz QDR FSB) |
| Motherboard | ASUS P4T-E (I850) Revision: 1.00 |
| Memory | 256MB 400MHz RDRAM |
| AMD Hardware Socket 462 |
|
| Processor | AMD Athlon 1333 MHz 266 MHz DDR |
| Motherboard | MSI K7 Master-S (MS-6341) AMD 761 |
| Memory | 256MB DDR-SDRAM, PC2100, CL2.0 |
| Graphic Cards | |
| NVIDIA GeForce2 Ultra | Chip-Clock: 250 MHz Memory: 64 MB DDR-SDRAM Memory-Clock: 230 MHz NVIDIA Reference Driver v23.11 |
| NVIDIA Geforce2 Ti | Chip-Clock: 250 MHz Memory: 64 MB DDR-SDRAM Memory-Clock: 200 MHz NVIDIA Reference Driver v23.11 |
| NVIDIA Geforce2 MX400 | Chip-Clock: 200 MHz Memory: 64 MB SD-SDRAM Memory-Clock: 166 MHz NVIDIA Reference Driver v23.11 |
| NVIDIA GeForce2 Pro | Chip-Clock: 200 MHz Memory: 64 MB DDR-SDRAM Memory-Clock: 200 MHz NVIDIA Reference Driver v23.11 |
| PowerVR Kyro II | Chip-Clock: 175 MHz Memory: 64 MB SD-SDRAM Memory-Clock: 175 MHz Driver v1.00.09.0031 |
| NVIDIA GeForce3 | Chip-Clock: 200 MHz Memory: 64 MB DDR-SDRAM Memory-Clock: 230 MHz NVIDIA Reference Driver v23.11 |
| NVIDIA GeForce3 Ti200 | Chip-Clock: 175 MHz Memory: 64 MB DDR-SDRAM Memory-Clock: 200 MHz NVIDIA Reference Driver v23.11 |
| NVIDIA GeForce3 Ti500 | Chip-Clock: 240 MHz Memory: 64 MB DDR-SDRAM Memory-Clock: 250 MHz NVIDIA Reference Driver v23.11 |
| ATI RADEON 7200 | Chip-Clock: 183 MHz Memory: 64 MB DDR-SDRAM Memory-Clock: 366 MHz ATI-Driver: V7189 (Win98) ATI-Driver: V3276 (WinXP) |
| ATI RADEON 7500 | Chip-Clock: 290 MHz Memory: 64 MB DDR-SDRAM Memory-Clock: 460 MHz ATI-Driver: V7189 (Win98) ATI-Driver: V3276 (WinXP) |
| ATI RADEON 8500 | Chip-Clock: 275 MHz Memory: 64 MB DDR-SDRAM Memory-Clock: 550 MHz ATI-Driver: V7206 (Win98) ATI-Driver: V3286 (WinXP) |
| Drivers & Software | |
| AGP (AMD) | Miniport Driver 4.80 (Win98SE) |
| DirectX Version | 8.1 |
| OS | Windows 98 SE, Version 4.10.2222 A Windows XP Pro(V2002) |
| Aquanox | Version v1.17, Timedemo Magelan |
Currently, beta drivers which fix these problems are available on the web. But be warned: these are not official drivers and, as such, are not supported by ATi.
Benchmarks — Level: Magellan
The Pentium 4 benchmarks were run both in Windows 98SE (red) and Windows XP Professional (green). Since no real differences are expected, the Athlon tests (blue) were run only in Windows 98.
800×600-32: Thanks to their higher clock speed, the GeForce3 cards grab the lead. The RADEON 8500 is close on the heels of the Ti500, though. The performance differences of the ATi cards in Win98SE and WinXP are interesting. The GeForce2 cards show no such phenomenon, offering nearly identical performance in both operating systems, across the board. The RADEON 7500 follows at a respectable distance, trailed by the RADEON 7200 (the graphics card formerly known as RADEON DDR 64MB ViVo). The GF2 MX barely makes the magical 25fps barrier, but the Kyro II is already on its knees, crawling along at 20fps. We can safely say that the game is not playable on either of these two cards, even at this resolution.
Benchmarks — Level: Magellan, Continued
1024×768-32: As we increase the resolution, we see the Ti500 now tied for the lead with the RADEON 8500. Slowly but surely, memory bandwidth, not raw processing power, is becoming the deciding factor in this race. Consequently, the field now shows a clearer ranking, with the GeForce2 Ultra outrunning the Ti and the Pro. The R7500 can also close the gap to the GF2 group, threatening the slower models. Unsurprisingly, the MX400 and Kyro II continue to drag behind — the game is completely unplayable. They are not alone though, as the R7200 now shares their fate. It, too, can no longer pass the 25fps mark.
Benchmarks — Level: Magellan, Continued
1280×1024-32: Cranking the resolution up even higher leaves us with only three cards still able to handle this game: the GF3, the Ti500 and the RADEON 8500. Even on the Ti200 the frame counter dips below the 30fps mark, barely remaining above the scores of the GF2 Ultra. This is a direct result of the Ti200’s low memory clock speed. On the MX400 and the Kyro II, the game is effectively reduced to a slide show.
Benchmarks — Level: Magellan, Continued
FSAA 1024×768-32: The cards take another performance hit when anti aliasing is enabled. Even the likes of a Ti500 or a RADEON 8500 are barely able to offer playable frame rates at this resolution.
Conclusion
While current video cards have no trouble with older games like Quake 3, running them at record speeds of around 150fps even at high resolutions, the new crop of DirectX 8 games can bring even these performance monsters to their knees. The CPU is really only of secondary importance, as the scores of the Athlon 1333 system show.
The scores shown here are average frame rates. Performance may be even lower in the game, depending on the complexity of the scene, the number of enemies on screen, etc. Activating FSAA is the final blow, bringing the frame counter lower still. Yet the trend is pointing towards ever more complex 3D engines (due in large part to the Xbox). If you’re planning on buying a GF2 or R7500 now, you may find yourself lowering the detail level or the resolution in the near future, just to make upcoming games playable. Even a GF3 Ti200 reaches its limits rather quickly. Nonetheless, don’t panic just yet — games that make extensive use of pixel or vertex shaders are still a bit further off on the horizon. In today’s games, you’d need a magnifying glass to even find any of these new effects (as in Comanche 4).
