[A04eArchitect]

Actually, the bottleneck wasn't the I/O, it was the context switching. If anyone wants the specific memory map addresses I used for the ARM v8-A bypass, let me know