RDMA with GPUDirect

Good Morning everyone,

I’ve been struggling on using GPUDirect to transfer GPU datas (images) (RTX Quadro 6000) through PCIe to a FPGA. I’ve been working on this subject for a month or so now. I’ve read a lot of threads on this forum. But it seems really unclear what we must setup to use GPUDirect ?

We had a hint to check GPU IO Mermory Mapping and this example GitHub - gpudirect/libgdsync: GPUDirect Async support for IB Verbs , specifically the memmgr.cpp file. But even with that I’m not sure about what i need to do.

It’s a very important subject for our company, we’ll be if needed glad to get in touch with nvidia support !

Thank you,

BrieucD