I am trying to implement UMsg on fpga-bdx-opae servers on Intel HARP with OPAE sdk. I have been successful in simulating the design with ASE. When I run it on hardware, I get errors.
1. I tried using the API for getting the UMAS pointer: fpgaGetUmsgPtr() but it returns a FPGA_NO_MEMORY error.
2. Output for fpgaGetNumUmsg() gives me that there are 0 UMsgs supported.
Is there something that I might be missing since the document I’m referring to clearly mentions that “UMsg is only supported in the Integrated FPGA Platform”, which “fpga-bdx-opae” satisfies.
May I know which processor are you using in your system? The UMsg can only be supported if you are running the Intel PAC card with Intel XEON Processor.
May I know which PAC card are you using?
After further checking, the UMsg is only supported in special FPGA device which is teh Intel XEON-FPGA Hybrid device.
You may refer to https://www.anandtech.com/show/12773/intel-shows-xeon-scalable-gold-6138p-with-integrated-fpga-shipp... for the announcement of the Xeon Gold 6138P with Arria 10 FPGA.
Thank you very much for the answer! However, it is still not clear how we can access a server with such a device to run our experiments on HARP. https://wiki.intel-research.net/FPGA.html states that fpga-bdx-opae cluster is built with integrated XEON-FPGA Hybrid devices you mentioned. However, we observe uMsg is not working on that machine and AVX-512 is not supported either (gcc explicitly tells that).
Is fpga-bdx-opae a right server to work with XEON-FPGA Hybrids? Seems like is has the right FPGA, but the processor differs from the one you mentioned. Can you kindly give us a little more information on how we can access a server that is powered by Xeon Gold 6138P so we can play around with uMsg and AVX-512? I think, this is more vLab/HARP infrastructure specific question though.