fast I/O on Raspberry Pi

[NiHaoMike]

https://www.raspberrypi.org/forums/viewtopic.php?p=1697532
So far, it only seems to work exceptionally well on the Pi Zero. It won't match a Beaglebone Black, but still impressive for how cheap a Pi Zero is.

[Doctorandus_P]

Why do you call 80MB/s fast for a (multi core?) processor that runs on a GHz or so?
And it has to be reverse-engineerd from missing datasheets also.

[brucehoult]

You can do something similar on an ATMega2560 (or the smaller versions) e.g. on Arduino Mega. 16 GPIOs are used for address and data (with the high 8 bits of the address multiplexed with the 8 data bits) to allow addressing 64 KB of total RAM. Not as quickly, obviously, due to the lower clock rate, but I think it might manage 8 MB/s.

[janoc]

Why do you call 80MB/s fast for a (multi core?) processor that runs on a GHz or so?
And it has to be reverse-engineerd from missing datasheets also.

I guess the difference between CPU clock and the external interfaces/peripherals has been totally lost on you. 

[Doctorandus_P]

Nope.

More then 12 CPU clocks for an I/O transfer is nothing exceptional.
This has nothing to do with absolute numbers.

If you want sustained high transfer rates, then look at the Cypress FX2 or FX3 for a nifty peripheral implementation.

[janoc]

Nope.

More then 12 CPU clocks for an I/O transfer is nothing exceptional.
This has nothing to do with absolute numbers.

If you want sustained high transfer rates, then look at the Cypress FX2 or FX3 for a nifty peripheral implementation.

The I/O transfer rate is not dictated by the CPU clock but by the peripheral clock itself and how fast (or not) are the connections to the bus where the memory is. Which given how the Broadcom SoC is built and that we are talking about GPIO (well, SMI) and not a dedicated high speed interface, it is pretty solid.

If someone wants a gigabyte/second speeds they won't use what essentially amounts to an old parallel port.