| zarx said: GabeN was right back in 2005 multi-threaded programing is still an issue all these years later. Splitting code up so that it executes across multiple threads is really really hard to do with many tasks. Some things just have to be done serially. It's a pity that we still haven't replaced silicone yet. The GHz wall it imposes has pretty much forced the CPU industry to go for more cores to maintain any meaningful performance increases. And while that is good for many tasks it's just not suitable for everything. I can't wait for graphene or nanotube processors with 10x higher clocks. |
This. While multiuser, multitasking and DSP (including graphics) mainstream applications and supercomputing ones started using available parallelism quite early, single user mainstream SW, including games, often hogged single core's power, but almost always lagged behind what HW offered in terms of parallel computing power, and this affects games the most, as they need most of the computing power available on a machine, and outside of easily parallelisable tasks as graphics, the performance of a single core can still be a bottleneck for intrinsicly serial tasks. Better libraries and dev tools can help, developers learning to identify and reduce intrinsicly serial parts to the bare minimum can help even more. For some AI parts, using fuzzy logic and devising AI that can work on partial data, to reduce mutual dependencies, could help too, while mission and quest files could specify when conditions must be met more strictly for the correct advancement of plots, missions and quests.
