directGMA vs SDI-link

Discussion created by rwistort on Aug 1, 2017
Latest reply on Aug 3, 2017 by dipak



I am trying to do object tracking in Realtime using an FPGA-based frame grabber and a gpu.


Bitflow, my frame-grabber manufacturer, supplies sample code demonstrating how to reduce latency by using DirectGMA to stream images directly from their frame grabber to an AMD GPU. They state that, for this to work, I'd need a FirePro gpu supporting DirectGMA, plus assorted drivers and sdks. So I bought a W4300, installed all the right software, and compiled and ran the sample code.

It failed while calling the extension loaded with the tag "clEnqueueMakeBuffersResidentAMD" with return code -5: CL_OUT_OF_RESOURCES


Before I bore you with a code snippet, let me ask:

Are directGMA and SDI-link two different things?

My W4300 supports directGMA, but not, I think, SDI-link.

Bitflow's sample code is known to work on a W8100, which supports both directGMA and SDI-link.

A (very) few sites on the web suggest that directGMA is for GPU-to-GPU transfers only, while SDI-link is for FPGA-to-GPU transfers, which is what I'm trying to do.


If this is the case, then I bought the wrong card!!


Can someone clear this up for me?


Many thanks.