GPUDirect rdma

Hello everyone,

I have successfully developed code to capture UDP data packets sent by an FPGA using raw queue pair code. I can receive these packets using a ConnectX-4 Ethernet adapter card and store them in system memory.

My next goal is to directly send this data to GPU memory (Tesla K80) using GPUDirect RDMA. I have reviewed the following document and also learned that the same task can be accomplished using GPUDirect Async.

I have read the documents listed below, but I couldn’t find any complete sample code to guide me on how to implement GPUDirect RDMA/GPUDirect async for my project:

  1. GPUDirect RDMA Overview
  2. GPUDirect RDMA Product Page
  3. GPUDirect RDMA Documentation

Could anyone provide references or sample code that could help me achieve my goal?

Hello and thank you for writing to us

Considering the technical depth of your project, involving direct data transfer between an FPGA, using a ConnectX-4 Ethernet adapter card, and a Tesla K80 GPU, and the challenges you’ve encountered in finding comprehensive examples or sample code to guide your implementation, it’s advisable to seek direct support in our NVIDIA Enterprise Support Portal.
Our support team can engage our engineering team and help with your setup.

Thanks,
Ilan.