During EuroGraphics 2010, I assist to a really amazing presentation about the state of the art in procedural noise by Sylvain Lefebvre and Ares Lagae.
Since Perlin Noise, the researches have continued and lead to a better understand of noises in general. Perlin Noise may remain a descent solution for some cases where speed is critical but new solutions provide better noise for some other scenarios with less side effects. What I really appreciated is the representation of the different methods throught their power spectrums which allows the understand the nature of the noises. As a result of the lastest researches, some scientists manage to extract the noise of some pictures to generate new pictures... This is still really early for practical usage but definitely interesting and impressive as I could not have tall which one was the photograph. With sparse gabor convolution, a noise texture can be apply on a mesh without aliasing effects and without multiple sampling per pixel.
On a different aspect, I was present for a key note by Erik Hagersten about 'Super-linear Slowdown with Multicores'. Yes, having multi core CPUs can reduce efficiency... at least at equivalent transistor count. He defined the issue behind multicore with three words: bandwidth, locality and parallelism.
What about noises? Noises are procedural which implies almost no data set, few locality issues beside reuse of generated results and easily paralelisable...
What this actually means? I guess, noise is a problem that fit into the current multicore environment. Let's through out of nowhere an idea: What about a procedural noise unit in OpenGL 5 hardware? I'm not thinking about something like the GLSL noise functions (still not implemented anyway...), no, I am thinking about something where programmers would have a proper control on the generated noise. I guess, it could be really cool but this is definetly an early thought. I'm not sure if the researches are mature enough but I am definetely looking forward for more results on the on going researches.