08-22-2019 03:17 AM
I am trying to add u200 card as a service using this plugin. The card is detected properly by the plugin and the plugin is sending the request to kubelet.
In kubelet journactl logs, we find that gRPC client state is in the READY state, but the response is not at all received to kubelet and we were unable to set up the FPGA card as a separate service.
Here are the environment details:
OS - ubuntu 16.04
Kernel - 4.15.0-45-generic
kubernetes - 1.15.3
docker - 18.06.2-ce
XRT - tried with both 2019.1 and 2018.3
Kindly, help us troubleshoot this issue.
Thanks in advance,
09-05-2019 09:15 AM
Sorry for the delay. I'm still going through my proof of concept to further assist.
However, I see that your kernel version isn't something XRT typically works with.
Can you double check that XRT by itself is able to find your acceleration card?
09-06-2019 02:50 AM - edited 09-06-2019 02:53 AM
We have installed the XRT driver and all the "scan and validate" commands are working fine. In the pod, it was instructed to run "./validate.exe verify.xclbin" from plugin instructions. But we are facing error while running it saying that it mismatches the DSA timestamp. In the pod, it is showing 18.04 in LSB release and 16.04 in uname. Is this proper or any changes needed?
Can you please let me know what are the Kernel Versions supported?
09-06-2019 07:31 AM
My only input for now is that if you are seeing a message saying there is a timestamp mismatch, the XCLBIN you are trying to load onto your board was not built against the DSA/Shell that is currently loaded on it.
The DSA timestamp is gathered from the command
$sudo xbutil flash scan
the xclbin timestamp can be found in the output of
$xclbinutil -i <xclbin .xclbin> --info
Can you run those commands and paste their output here?
09-06-2019 08:05 AM
Thanks for the assistance.
We are able to detect it in the Host machine and the pod too now. Can you please let us know the better supported OS(16.04 or 18.04) and the Kernel too for XRT 2019.1 and XOCL 2018.3?