Is there any reasonable assumption how much fps I could expect when driving 2048 WS2812B on 16 data lines and rendering one noise layer? (On a Teensy 3.1)
I think it’s 4-6ms per frame writing out. The bigger, more expensive, ugly would be the noise functions (which I need to do a round of arm optimization on, they’ve been far more optimized for avr)
Does that mean that Fastled supports per pin DMA now on the Teensy 3.1? I’ve been away from the community for a bit and may have missed this. Usually I would expect 2048 WS2812b to be about 16fps, or 61ns a frame.
If you want 8-way DMA based output, you can use the OctoWS2811 controller which wraps OctoWS2811 which would use DMA (though, it turned out fast led 8-way output was faster, in terms of CPU usage than OctoWS2811 on its own, just because of more efficient bit rotation/assignment code. FastLED w/OctoWS2811 is a nice boost)
However, you can also do 16-way parallel output that doesn’t use DMA (at the moment). 2048 leds spread out across 16 lines is 128 leds per line, which is about 4-5ms to write them all out.
Awesome 
I have been really enjoying the Teensy 3.1, and will totally admit to writing poor code because the overhead is so much better compared to a standard arduino.
That’s ok, I make up for that in FastLED because I’m always trying to do more than I’m told the hardware at hand can do. I have some more optimizing for “STUPIDLOADS” of leds on the teensy 3.1 before I take the next step and start playing with overclocking it