I'd guess that this will output faster than the average reader can read, even while using only CPU inferencing on a modern-ish CPU.

The param count is small enough that even cheap (<$500) GPUs would work too.