August 8 - 14, 1998
Archive

This is the latest i740-driven board to reach my desk. This time around it's a Hercules board we're looking at. They call it the Terminator 2x/i. This board is equipped with 8MB of SDRAM. This is a full-blown AGP2X board too. This allows you to store the textures in system RAM, leaving the 8MB open as frame buffer memory, giving you the ability to support some spiffy high resolutions.

Further down in the review:

Pros:

• Full blown AGP 2X support.
• It's a very nice solution for those with under a $100 to spare.
• Excellent image quality in D3D and Open GL games
  
  Cons:
  
• No Open GL support in Win95 out of the box (yet).
• Must be running a Pentium II with an AGP slot, running Win95 OSR2 with the USB update or Win98
• No software bundle.

3. I've been reading a lot of RUMORS about a V2 rev. 3 chipset,
supposedly spec'd at 125 Mhz. While I see no reason to doubt this is possible, I have yet to get any confirmation. Could you clarify
please?

> We have developed a new board design (the rev-3 board) that makes better utilization of memory, and does run substantially faster if populated with faster memory. We . are working with OEM's to help them bring this new board to market, but the chipset is unchanged.

Read the whole interview with lots of other juicy stuff from Silicon Investor.

1. The SDK is now broken into 2 components:
   • g3sdk.exe (Glide 3 sdk)
   • g23dgs.exe (Glide 2 and 3 diagnostics)
2. No installation program. This is a self-extracting archive. All the files in the archive, with the exception of "readme.txt" and "manifest.txt" have a path starting in the root "3Dfx\", so this is safe to unzip anywhere. See the readme within the archive for more details.
3. No drivers included. The SDK depends upon the operating system, not the cards. The Glide 3x drivers are available in a separate kit. This is the SDK for Windows. Glide 3x does not and will not support DOS.
4. Restructured directories. This kit has a simplified directory structure as compared with the previous kit. It will also coexist as a sibling to the Glide 2x kit.
5. No EXE's included. A few of these are in the diagnostics kit, sufficient to test if Glide and the card are working properly. The rest of the Glide tests should be easy to build from the source included in the SDK.
6. MSVC 5 directly supported with projects and workspaces. Also makefiles MSVC using NMAKE, a Microsoft make utility.

The RivaTNT, V3300, and the Voodoo2 will probably be the 3 top performing chip sets by the end of the year. Each has a different design philosophy, and the architectures that have resulted, each have certain strengths and weaknesses. This comparison will focus mainly on performance, but I'll touch on some quality features too.

There's too much good information to post little bits of it here, so go there and check it out.

3D Rendering Features

• Single Cycle Trilinear Filtering
• S3^® Texture Compression
• True Color Rendering
• Void and Cluster Dithering for 16-bit Modes
• Specular Lighting and Diffuse Shading
• Alpha Blending Modes
• MPEG-2 Video Textures
• Edge Anti-Aliasing
• Vertex and Table Fog
• 16- or 24-bit Z-buffering
• Sprite Anti-Aliasing, Reflection and Environment Mapping, Texture Morphing, Shadows, Procedural Textures, and Atmospheric Effects

Talking about that performance, it's really good... The 3D part outperforms the G200 chipset from Matrox on most 3D benchmarks, though in 2D the G200 is superior. OpenGL performance is really good too. There wasn't a MCD OPENGL32.DLL file supplied with this card, but it ran Quake II quite OK with the standard OPENGL32.DLL file located in the WINDOWS/SYSTEM directory. However, this file is well over 700 Kb and not by a long shot all the functions specified in it are used by Quake II... As a result of that it carries a lot of unneeded load and the performance isn't optimal... I happened to have the OPENGL32.DLL wrapper from Matrox's G200 (only 130 Kb) around, and that one happened to work flawlessly with the S3 Savage3D chipset, image quality was great and so was the performance.. I know that end users won't get the wrapper from Matrox, but this is meant to show what we can expect when S3's own MCD (or ICD) will become available

The X-24's performance is clearly at 100% of the level of the standard SLI Diamond Monster3D II setup, with each score between the two rigs coming in at almost identical levels.

All modes of the X-24 performed perfectly, and the Quantum3D drivers allowed a good deal of flexibility in the setup/implementation of the various Voodoo2 native options.

One item we'd like to mention here is the X-24's ability to stabily run while overclocked. As part of each Voodoo2 review, we try to test the maximum MHz that the product will run properly at, and record it for our notes. When something out of the ordinary occurs, we mention it, and with the X-24 we noticed that even with no measures taken to cool the large card, it still pounded out the Cyrellis 3D benchmark suite at 106MHz without hesitation. This is about three to five MHz above what the average Voodoo2 has turned in on the same tests without cooling, so it's worth noting. Most likely the quality of components that Quantum3D has selected for the X-24 is directly correlated to the increase in sustainable MHz levels.

Otherwise, the X-24 was a great performer, living up to the expectations that anyone would have when looking for an SLI-Voodoo2 solution.

Let's get straight to the point. If you absolutely must play 3D-accelerated games at 1,024 by 768 resolution, then you might consider spending $599 for this graphics card. Or if you have an extremely fast system (Pentium II 400-MHz class), then you'd at least be able to get some performance benefit over a standard Voodoo 2 card with the Obsidian2 X-24. Otherwise, this card is overkill.

San Jose, CA, August 10, 1998 – 3Dfx Interactive™, Inc. (NASDAQ: TDFX) announced that its newly introduced Voodoo Banshee™ chip for 2D and 3D graphics has earned the top score of 1290 in Ziff Davis' 3D Winbench 98® according to tests run by Mercury Research, a leading PC market research company, eclipsing a score from the next closest product by 100 points. The tests were run using an alpha version of the software drivers and a beta version of the chip. Results on final shipping drivers and the final version of the chip are expected to be higher.