[Engineering Feature]
Games Flourish In A Parallel Universe
Multicore processors accelerate games if developers can take advantage of the features and live with the limitations.
BALANCING ACT Clinton Keith, chief technical officer for Vivendi Software's High Moon Studios, notes that the difficulty in targeting games to different platforms is in matching the application with the system hardware. High Moon Studios developed the popular Darkwatch game that runs on both the PlayStation and Xbox platforms.
Keith indicated that bottleneck identification is critical to enhancing system performance. Once identified, the bottlenecks can be addressed. This means moving where and how computation is performed, depending on the architecture.
For example, CPU/GPU communication is constrained on both the PlayStation 3 and Xbox 360. Memory and DMA bandwidth are often more of an issue than raw computational performance. That's something developers should consider when designing the system because it may mean that those extra cycles can be used for other chores.
Game developers have taken shortcuts and made approximations due to the game platforms' lower-than-necessary performance to provide a computationally accurate real-time gaming environment. It's only recently that more complex simulation could be performed.
Backgrounds started as fixed images, grew to sprite objects, and then became objects with more density and complexity. A tree may appear lifelike from one angle, but not from another. Shadows may not be realistic, and so on. Incorporate more algorithms and processing power, and the difference between the apparent and real instances becomes smaller.
Platforms like the Xbox 360 and PlayStation 3 open up the possibility of modeling objects like trees. It would use a rule-based system running on an SPE that might otherwise be idle. Other possibilities include moving computation currently found on some GPUs to the CPU (or SPE). Taking advantage of more computation power may only be part of the issue, though.
Loading, memory, and bandwidth also come into play, so moving code and data to different parts of the system may open up opportunities. It may be possible to reduce the amount of information that flows between the CPU complex and GPU, improving overall system efficiency. For example, a CPU can "chew down" polygons into triangles handled more efficiently by the GPU. This, in turn, may free up cycles for other tasks.
Many tasks like the tree simulations are becoming possible with this new hardware. High on the list is improved accuracy, particularly when it comes to simulating the physics of a game. However, gaming platforms can't take advantage of new hardware like Ageia's PhysX chip (see "BFG Technologies And Ageia Make Physics Fun,").
But they can apply their multicore performance to the task, making them significantly better than earlier gaming platforms and most PC-based solutions. Chips like Ageia's will still have the edge, just like GPUs have the edge for graphics, so they may show up in future gaming platforms.
Another key area affected by the new hardware is improved scripting language execution and artificial-intelligence support. The latter would allow flocking algorithms to be used, thus enhancing the realism.
Even more possibilities open up as multiplayer games are added to the mix. Using single systems to coordinate the other systems involved limits overall system utilization. Balancing ease of programming, system response, and system utilization is a complex task that's made more difficult with the need to address multithreaded applications running on these platforms.
Keith notes that the potential of the Xbox 360 and PlayStation 3 is just at the onset. It will take a copule of years, plus tools like CodePlay's Sieve, to really see what they can do. Developers at companies such as High Moon Studios are already exploring the possibilities when using up to 32 hardware threads. In the meantime, expect some stunning and aggressively intelligent games to emerge on these platforms.
Reprints
