For the purposes of evaluating 3D performance we chose three real games: Quake II, Incoming and Forsaken along with 3D Winbench 98 from Ziff-Davis.

To measure 2D performance we used the Business Graphics Winmark 98 tests (part of the Winbench 98 benchmarking suite) from Ziff-Davis.

Our test machine was a Compucon Computers and Peripherals system with P2L97 motherboard, Intel 300MHz Pentium II processor, 128Mb SDRAM, 512Kb RAM cache, 6.4GB U/IDE hard disk, U/IDE controller, Sony Trinitron Multiscan GDM-200PS monitor, and L2300 AGP graphics accelerator with 8Mb of SGRAM (used with 3D only pass-through cards). Tests were conducted on Microsoft Windows 95 version 4.00.950 C (with the USB supplement installed and the updated VGARTD.VXD file) and MS-DOS 7.0, with Direct X version 5.0, FAT32 file system and no hardware disk cache.

Id Software's Quake II (version 3.14) was an obvious choice for a number of reasons. Firstly it is one of the most popular 3D titles currently available and a lot of new games are based around the Quake II engine. It also has the capability to utilise some fairly sophisticated 3D features if supported by the hardware. We ran the Timedemo test with high quality sound enabled. Obviously this is going to have some impact on maximum frame-rates, but who on earth plays games with no sound in the real world? Where possible, we used the newest version of the miniGL Quake II driver for each 3D card. We then ran the Timedemo using demo1 map in 640 by 480 and 800 by 600 resolutions with detail levels set to high, in full-screen mode with 16-bit colour (8-bit colour palette disabled). If you wish to duplicate our testing procedure simply do the following:

Start Quake II, change the relevant video settings (make sure you use the correct miniGL drivers for your card or chipset) then use the "~" key to bring up the console. Now type the following:

"timedemo 1" then press "Enter"

"map demo1.dm2" then press "Enter" once again.

Incoming is a sexy looking new Direct3D action game from Rage Software. It uses some advanced 3D effects and will really put a 3D card through its paces. We used the "Gameindex" feature which comes with the demo as our second real-world benchmark.

If you wish to duplicate our Incoming testing procedure simply do the following:

Once the demo installs there should be two Incoming shortcuts on your desktop. Go to the one that says "Gameindex.exe" and right click it with the mouse. Now select "Properties" and go to the "Shortcut" tab in the properties box. At the end of the "Target" input box insert the following: "C:\incoming\gameindex.exe –screenmode". Now double click the "Gamindex.exe" shortcut on the desktop. Before the timed demo starts you will be prompted to select which display driver you wish to use (here you should choose the one which corresponds with your 3D card) and then what resolution and colour depth you wish to run the timed demo in. For our purposes we used 640 by 480 and 800 by 600 in 16-bit colour (the only exception was the Righteous 3D, a first generation 3Dfx Voodoo Graphics based card which doesn't support 800 by 600, for this card we used a screen resolution of 640 by 480).

Forsaken is a hot new "Descent-style" 3D action title from Acclaim. The demo also features a frame-rate counting feature. Download the demo and start it using the standard shortcut from the "Start Menu" (the one which just says "Forsaken Demo", rather than "Forsaken Demo for 3Dfx", "Forsaken Demo for PowerVR" etc.). Now make sure all the detail settings are at maximum and go to the "Play Demo" option on the menu. Start the demo and press the "Escape" key to get a list of options. Change the screen resolution here if you like, but most importantly enable "display frame-rate". For our purposes we used 640 by 480 and 800 by 600 in 16-bit colour (the only exception was the Righteous 3D, a first generation 3Dfx Voodoo Graphics based card which doesn't support 800 by 600, for this card we used a screen resolution of 640 by 480).

While the Ziff-Davis suite of benchmarking utilities has been used by many magazines (yes us too) to measure the performance of various hardware components, there has recently been some debate in relation to the validity of 3D Winbench 98 results. We kept this in mind when testing the 3D cards for this feature and as our results indicate there are some pronounced discrepancies between 3D Winbench 98 results and real-world testing results.

While the nVIDIA RIVA 128 based products (Viper V330, STB Velocity 128) came up trumps after 3D Winbench 98 testing, this was certainly not reflected in the timed game-demo results. The same can be said of the Permedia 2 based product (Fire GL 1000 Pro). This seems to be due to the "quality tests" which run before the "performance tests". Passing these quality tests increases a card's overall score.

After running the whole gamut of tests there was absolutely no doubt in our minds that the 3Dfx Voodoo 2 based products were the fastest performers by a country mile. But this superiority just isn't reflected in the 3D Winbench 98 results.

While we feel that 3D Winbench 98 results don't really reflect real-world game performance, we still see this benchmark as a valuable means of assessing how well support for the Direct3D API has been implemented in the drivers of each card.

The following test settings were used for eleven of the tests: 800 by 600 pixels, 16-bit colour, Full Screen, Direct3D HAL, RGB Emulation, Front Buffer with a result, Execute buffers.

To test the 2D Windows 95 performance of the cards in this feature with 2D acceleration capabilities we used the Ziff-Davis "Business Graphics WinMark 98" suite of tests from "Winbench 98". This measures 2D performance when running a range of standard business applications software.