I created a proof-of-concept 2D noise generator, that uses a vertex queue and propagates neighbor-to-neighbor, to generate gradient noise over a specified area. Source code and more detailed explanation here: https://github.com/KdotJPG/Noise-VertexQueue-AreaGen

I've had ideas related to this for a while. I posted generally about it here. The first idea that came to my mind, was to simply iterate over skewed lattice coordinates. It would have probably required the least implementation effort, but I don't think it would have been very good for generating slices of higher dimensional noise. At least, not unless you rotated and re-parameterized the domain, or used Old Perlin. There is also a company that is trying reserve it for themselves now. I want to make sure everybody has access to speed-optimized gradient noise for whole-area generation, so I spent some time to come up with this instead.

Feel free to try this out. Use it in whatever you want. Also let me know about any bugs you find, or features you think should be added. I have it under the MIT license for now, but I might switch it to a public domain dedication after refining it a bit more.

As far as performance goes, I'm seeing improvements in the ballpark of 250% for the 2D noise, when applied to the right scenarios. You can probably improve the performance further, if you reduce the kernel size to stop at the cell boundary like classic Simplex, instead of the nearest vertex. Relatedly, you might notice that it is called "SuperSimplex noise". Stay tuned for more about that in the future!