Dear AMD Community,
I have the intention to use a GPU with DirectGMA capability and pair it with an FPGA (Artix 7 or Kintex 7). After browsing for similar questions and topics, I found out some suggesstions regarding the FirePro series. However, I did not manage to find anything newer since 2018... Because of this, may someone enlighten me about the following points:
1) Which models do support DirectGMA nowadays? As far as I know, Radeon Pro W5500 is one of those... but are there any non-pro Radeon models that also feature it as well?
2) In the context of the ROCm framework, where can I find some examples and tutorials regarding PCIe communication between the GPU and an FPGA - regarding the driver, BAR configurations and an example code for a basic transfer?
3) In addition to point2, is it mandatory to use a PCIe switch for the DirectGMA between a GPU and the FPGA?
Regards,
Varban
Solved! Go to Solution.
DirectGMA is supported throughout all FirePro and Radeon Pro GPUs. This is a Pro feature and not included in our consumer cards.
https://www.amd.com/en/system/files?file=documents/radeon-w5700-datasheet.pdf
Check under Key Features
We have plenty of info on:
https://github.com/RadeonOpenCompute/ROCm
DirectGMA is supported throughout all FirePro and Radeon Pro GPUs. This is a Pro feature and not included in our consumer cards.
https://www.amd.com/en/system/files?file=documents/radeon-w5700-datasheet.pdf
Check under Key Features
We have plenty of info on:
https://github.com/RadeonOpenCompute/ROCm
Hi @fsadough ,
Thank you for your response! I was happy to find such a detailed documentation of ROCm.
The only thing that I am still wondering about is the PCIe topology that will need for the DirectGMA transfers, but I will open a new thread about it.