we need a fast data transmission between a Xilinx Artix 7 (AC701 Eval Board) and Nividia Jetson TX2 Board.
We want to send 48 MByte data every 50 milliseconds from Artix 7 to Jetson TX2.
We try it with a IP-Core from Xillybus (http://xillybus.com).
The support from Xillybus told me the following:
“There is however a cache coherency issue with ARM-based platforms, which slows down I/O considerably, so I’m not all that sure about the NVidia platform.”
So my questions is:
Does anybody have experience with PCIe and Xillybus on TX2 or do you think this is possible with the TX2 Board?
“There is however a cache coherency issue with ARM-based platforms, which slows down I/O considerably, so I’m not all that sure about the NVidia platform.”
TX2 doesn’t have IO-Coherency enabled, so, it means that, APIs like dma_alloc_coherent() etc… return uncached memory, so, if the same memory is being used by both CPU and PCIe, there is a penalty because of the memory being uncached. But, that penalty won’t affect this particular use case.