| HoloDust said: To sum EG's article, from things that a typical game engine cycle consists of: - Game physics (update models) - Triangle setup and optimization - Tessellation - Texturing - Shading - Various render passes - Lighting calculations - Post effects - Immediate AI - Ambient (world) AI - Immediate physics (shots, collisions) - Ambient physics things that seem suited for running on cloud: - ambient background tasks - some forms of lighting (though it seems only for static environments) - background AI So, according to EG, not that much really can be co-processed via cloud, though it obviously can benefit open-world games with background AI/physics. |
They also said that background AI is not all that useful, as in not all that noticeable to the user. Radiant AI experiments with Oblivion didn't amount to much. The AI might do all sorts of clever things on the other side of the world, but what's the point if you'll never know about it.
I guess the best use of it is a dynamic growing world. Although not too dynamic since sending altering geometery over will quickly exhaust the bandwidth. Initially I was thinking it would be cool for a From dust style skyrim world, or Motorstorm with actual persistant terrain deformation mud physics, or Forza 5 with rain pooling and fluid physics. All too much data, especially with next gen's mesh sizes.
Can you animate large crowds on the server, dead rising, dynasty warriors style. Just to animate 2000 npc's at 30 fps you already need 7.3 mbps if you can compress the data down to a mere 16 bytes per npc.
So what can you do. Update the lighting of a race track (static environment) for a specific time of day. Is that really such hard work, GT5 can do it real time. Downloading the updated track, pre-calculated shadows and all, probably takes longer then preparing it locally. All next gen engines are going for dynamic global illumination. Moving objects affecting the lighting, smoke, swirling fog, trees swaying, clouds casting shadows, dynamic fire, dynamic lights, we're finally leaving pre-baked light maps behind.
What's left. Procedural level generation. Track editors where the server can render a highly detailed custom track. It might take a while to download the result but the server can have access to vast amounts of constantly updated data to create unique content. Certainly cool but not boosting the local rendering power.
Anyway I'm very curious what/if they show anything using cloud processing at E3.
