Skip to main content

Long read: The beauty and drama of video games and their clouds

"It's a little bit hard to work out without knowing the altitude of that dragon..."

If you click on a link and make a purchase we may receive a small commission. Read our editorial policy.

Tech Analysis: Forza 3 Demo

Digital Foundry takes a closer look.

Forza Motorsport 3 is due for release at the end of this month, and while we're reserving some time for a tech analysis on the full game, the recently released playable demo also offers up a number of decent performance metrics, some neat tech stuff, and a number of interesting decisions by the developers at Turn Ten.

A quick analysis of the in-game 60FPS mode throws up a couple of interesting graphical quirks. First up, you might notice a stippled "screen door" effect on all the surrounding vegetation: a similar effect is in play on SEGA Rally. This is known as "alpha to coverage" and it's a low bandwidth method of giving some kind of transparency. You can also see it in effect at the overpass near the end of the circuit.

The effect has almost certainly been included to help sustain 60FPS. Literally rendering fewer pixels helps contain overdraw, while the framebuffer format might also be important. On some of the formats supported by Xbox 360, transparent so-called alpha pixels take twice as long to render - not good when you're looking to maintain a smooth 60FPS.

Forza's reputation for 60FPS gameplay remains virtually unblemished in this analysis, with just one minor slip during the action. The other blips are reserved for switches between the game cameras; basically unnoticeable during gameplay.

Also interesting is the inclusion of what's known as a negative LOD (level of detail) bias. In the normal scheme of things, high resolution textures are reserved for areas of the game world close to the player, with lower resolution assets deployed further away. In the case of Forza 3, the road textures detail levels are much higher than they need to be for the available resolution, and this results in some superb definition at the expense of some shimmering as the detail is downscaled. Quite why Turn Ten has opted for this method over some kind of adaptive anisotropic filtering is a bit of a puzzle bearing in mind the relative paucity of the 360's texture cache – it may well simply be the case that the developer had the GPU time to spare.

HDR tone-mapping has also been included, although its usage is somewhat muted, presumably to stop over-the-top lighting effects affecting the gameplay.

Analysis of the game's replay mode suggests that Turn Ten has taken a leaf out of Polyphony Digital's book and has opted for increased detail and effects, sacrificing the 60FPS refresh rate in the process, on the track-side cameras at least.

A rock solid 30 frames in the replay sections, though a soft v-lock appears to be in play. On scene changes in particular you get the odd torn frame.

At this stage of the game, superior motion blur effects are added, anti-aliasing is upped to a smoother 4x MSAA and interestingly the game runs with a soft v-lock at 30FPS. In this particular replay you don't see the effect much at all, but torn frames can crop in here, particularly on scene-changing moments. On the final game it'll be interesting to see how the engine copes on really challenging scenes. Also intriguing is that the alpha to coverage effect looks different in the replays. Combined with the mask generated by the 4x MSAA, it has the bonus of adding additional colour shades.

However, as some have pointed out, it is very curious to note that the car models employed in the replay mode are not of the highest quality (judged by the standards of the models elsewhere in the demo). The angles shown of the cars at the beginning of the actual race employ the highest LOD (level of detail) models, whereas the replay shows significantly less detailed meshes for the vehicles. In all honesty, this is a bit of a mystery, maybe even a bug that will be corrected in the final build.

Check out the detailing around the headlight. The replay model (right) appears to be using lower detail 3D models when compared against the intro cameras on the actual gameplay (left). Very curious and it could be a bug in the demo code.

Onto the issue of crash damage then, and it's interesting to see how Turn Ten has implemented its model. The car has been divided up into different sections, and each has a pre-determined, incremental level of damage, with a number of removable parts. In the course of general racing, the effect is fairly realistic, but it's still somewhat removed from the reality of a high-speed crash during a real race. There's no deformation along the lines of Burnout Paradise, for example, and even the dynamically generated "battle damage" in the Xbox 360 version of Burnout Revenge hasn't been replicated.

Attempts to play Forza 3 as though it's Burnout show both the strengths and weaknesses of the damage model.

So, what's the big deal with crash damage any way? The bottom line is that it's a hugely important part of games that purport to be racing simulators. High speed racing is synonymous with danger, and risk must carry the threat of dire consequences... the days of ramming opponents with impunity and bouncing harmlessly off solid walls should be well behind us. The danger principle is one of the core philosophies behind Burnout, and it is interesting to note that the Criterion game's over-the-top, enhanced crashes are still vastly more realistic graphically than anything seen in Forza 3.

Interestingly though, Forza 3's model appears to have much in common with the latest work by Polyphony Digital on Gran Turismo 5. While our first impressions of the crash damage in GT5 at gamescom were not so positive, recent footage from the Tokyo Game Show suggests that Polyphony has made big, big improvements over the GC code. Removable body parts have been joined by localised damage. Whether it is dynamically generated, or pre-determined like Forza 3 is something we won't be able to confirm until there's better video or preferably some hands-on code to analyse.

It wouldn't surprise us at all if the damage is indeed pre-determined on both games. The Burnout titles have the luxury of using non-licensed cars; Criterion are the lords and masters of their creations and can do with them whatever they see fit. Both Forza and GT5 don't have that luxury, and it may well be the case that the developers need to get manufacturer sign-off on all the in-game representations of their vehicles. Detachable parts and pre-determined damage would be a convenient solution.

Forza Motorsport 3 is an impressive-looking game. Chances are that you've played the demo and you're as impressed with it as we are. Fingers crossed we can get a closer look at the final game pre-release...

Many thanks to Alex Goh for his input into this feature.