Closed GokulEpiphany closed 2 years ago
If possible, could you please spawn a remote instance with supported GPU and check for compatibility issues by doing a fresh install? This will greatly reduce the barrier to getting started..
Sorry to hear about that experience @GokulEpiphany .
At the moment, BenchBot expects there to be a hardware-accelerated screen to render to. There would be a bit of work to unwind this assumption on our end.
Something like VirtualGL is probably the most user-friendly way to achieve this.
Once a hardware-accelerated display exists, and it is set as the terminal's target with export DISPLAY=<address_of_display>
, then BenchBot should work as expected. A lot of BenchBot's development work was done from home during COVID lockdowns with this method.
If you're having trouble confirming a hardware-accelerated display is being used, I found tools like vkinfo
, glxgears
, and glxinfo
useful.
For now, working headless with benchbot..thanks for the suggestion.. closing this..
Is there a proven way to work with benchbot installed on a remote instance (AWS/azure)? I have tried everything I could think of, including things suggested previously. I have previously worked with Omniverse using websocket live streaming. Tried port forwarding, tightvnc, tigervnc. Install only works headless but fails when running benchbot_run with internal server error during starting of robot controller. Most errors seem to come from glx when installing with headless=False. Adding log file temp.log