3dfx "Napalm"

Say hello to FXT1™ Texture Compression...

Stop Yawning Damnit!

Which sounds totally unremarkable. After all, everybody and their dog (well, NVIDIA and S3 at least) already have texture compression support, and Microsoft have built it into DirectX as DXTC. In fact, even 3dfx will be supporting DXTC with the Napalm. Surely this is just another case of 3dfx playing catch up and trying to make it out to be a huge leap forward for the computer graphics industry?

Well, not quite. When asked about DXTC, the demonstrator said FXT1 "is very similar to that, but much better". In fact 3dfx are claiming that FXT1 is the "Next Generation" of texture compression, and if it works as well in silicon as it does on paper it should live up to that.

A compression ratio of 8:1 is being claimed, greatly reducing both the amount of memory taken up by the texture once it's on the card, and the amount of bandwidth needed to load the texture into the card through your AGP bus in the first place.

And although the compression system is "lossy" (ie the compressed version won't look exactly the same as the uncompressed version), 3dfx are claiming the loss in quality is so small you won't even notice it.

Here's The Science Bit

So, how does it work? Well, put simply the FXT1 system breaks each texture down into little 4x4 or 4x8 pixel blocks. It then tries to apply four different texture compression algorithms to each of the blocks, and picks the one that gives the best result for each individual block.

The card doesn't do all this encoding in real time, but the system does work fast enough for games to compress the textures on the fly during installation. This means that developers won't have to include a whole seperate batch of pre-compressed textures on the CD in addition to the standard and DXTC versions.

Once you've got the compressed textures into the card's memory it "knows" what to do with them automatically, and can process and decode them with no performance hit. Which is nice.

The bad news is that 3dfx weren't even demonstrating this technology at ECTS - they just told us how it worked and then showed a few before and after pictures. But assuming that FXT1 does what it says on the box, it should be more efficient than the current DXTC system.

Brave New World

Unfortunately that's not enough. 3dfx's big problem is that DXTC already has widespread support from both manufacturers and software developers. Why spend time adding a new texture compression system to your games when it will only work with a minority of the graphics cards out there?

The answer to this came as rather a shock. FXT1 will not only be cross-platform (working under Windows, MacOS, Linux and BeOS) and work with multiple APIs (Glide, OpenGL and DirectX), it will also be open source. All the source code and tools for FXT1 will be released for free! So, if you want to find out more technical details about FXT1 or download the source code for it, check this page on 3dfx's developer website.

Given that in the past 3dfx have kept a tight hold on their source code and even gone as far as threatening legal action against people who have attempted to replicate or back engineer their Glide API, this is a pretty big reversal. Whether it's a true change of heart or just sheer desperation remains to be seen, but either way it's an interesting decision.

Apparently this is just the beginning though. If you believe the hype, this is a "new 3dfx". I asked if this means that Glide will be made available as open source, and although 3dfx weren't ready to say that this was on the cards they certainly didn't rule it out either.

There's even talk of releasing information about their hardware for competitors to use in their own products, though personally I think it will be a cold day in hell before NVIDIA announce a new graphics chip based on 3dfx technology...

Free At Last!

The real question is whether making FXT1 available as Open Source will be enough to win over the game developers and graphics card manufacturers.

After all, if only 3dfx cards work with FXT1 and it doesn't offer a huge benefit over DXTC few developers will support it. But then if few developers support it there's not much incentive for competitors to swallow their pride and implement it in their own next generation graphics cards.

Epic are apparently adding FXT1 support to Unreal Tournament, but then they already support S3TC and DXTC so FXT1 support isn't going to give 3dfx users any big advantage at the end of the day. (Is that enough acronyms for you?)

FXT1 might be a better technology, but it's arriving over a year too late. Unless it proves to be vastly superior to DXTC and gains support from other graphics card manufacturers it might go the way of Betamax video recorders and the Sega Saturn...

T-Buffy

Of course, FXT1 isn't the only new technology that 3dfx are planning to introduce in their next graphics card. There's also the T-Buffer, which was announced several weeks ago.

The real advantage of the T-Buffer is that it allows "full scene anti-aliasing". This smooths out the picture that you seen on your screen, removing jagged lines along edges and other visual glitches.

For example, if you look at a wire fence in a game, the wire will begin to vanish as you move away from it because the wire is now less than one pixel thick on your screen. Full scene anti-aliasing gets around problems like this, improving the visual quality.

At the end of the demonstration they showed Need For Speed 3 running at 640x480 with full scene anti-aliasing. The effect was impressive - the game looked like it was running at 1024x768 or higher.

Unfortunately they couldn't switch off the anti-aliasing to show us what the game looked like without it, and they didn't try different resolutions either. Still, it looks very impressive in action and even at higher resolutions it should still improve image quality noticeably.

Burn Hollywood Burn

The T-Buffer also allows a number of other effects, including motion blur and depth of focus. These effects are far less useful though - it might be how your eyes work in real life, but it's certainly not the kind of thing you'd want to see in an action game! If I want everything to look blurry I'll take my glasses off...

In some circumstances though it can be useful. For example, Outcast uses depth of focus to great effect, focusing the "camera" on the key characters during conversations while their surroundings become slightly blurred. It's a subtle effect, but it helps draw your eye towards whatever the developer wants you to be watching.

The T-Buffer should also allow soft shadows and reflections, which look more life like than the sharp edged visuals of today's games. Strangely though 3dfx still only have a couple of pre-rendered shots to show us to demonstrate this.

Waiting For Godot

Perhaps more worryingly, 3dfx are still demonstrating the other T-Buffer effects on an array of eight overheated Voodoo II cards. In fact, it's the same setup that was used at E3, and having travelled around the world at least once it's starting to fall apart - several serious graphics glitches were visible in the demonstration we were shown that weren't there in the E3 demos.

The Napalm card was due out this autumn, but that obviously isn't going to happen now. In fact we'll be lucky to see it before next year in my opinion, as we're still being shown the same shoddy tech demo thrown together by 3dfx's engineers, and there's still no sign of real hardware.

3dfx aren't willing to talk about the card's performance either. They are promising to regain their crown as the Fill Rate King, but there's still no information about just how high that fill rate will be, what clock speed the card will run at, or how much memory it will support.

One thing they did let slip at ECTS is that the Napalm card will support bump mapping, although they wouldn't say exactly what type of bump mapping except that it will be different to the kind used by Matrox's G400 and supported by games like Expendable and Slave Zero.

Conclusion

3dfx's Napalm is the last of the great fill rate cards, pushing frame rates and resolutions to new extremes. And with the full scene anti-aliasing offered by the T-Buffer, image quality should improve noticeably as well.

The card also sees 3dfx finally catching up with the rest of the pack, introducing features like 32 bit colour, texture compression, and bump mapping that were missing from their older Voodoo cards.

But although the card will play fast and look good, it's certainly not a next generation card. For that we'll have to wait until some time next year when 3dfx should unveil a graphics card that combines the speed and visual quality of the Napalm and the T&L acceleration of cards like the Savage 2000 and GeForce 256.