Open hyperknot opened 7 years ago
In theory tangram-es can run using a software OpenGL implementation like Mesa, though performance will likely be greatly diminished. Integrated Intel GPUs generally provide acceptable performance for client-side rendering, but server workloads can be very different.
I wonder if performance would be that bad. I've just checked tileserver-gl and it simply runs mapbox-gl-native under xvfb. Tangram might be a bit more taxing on the GPU than Mapbox GL Native, but I wonder how much.
https://github.com/klokantech/tileserver-gl/blob/master/run.sh
We haven't seriously benchmarked a software only approach like Mesa, but @zerebubuth had it working in an earlier incarnation of a Tangram server renderer. I would certainly be interested to know how software and lower-end GPUs compare! We have primarily deployed Paparazzi internally on AWS GPU machines for heavier workloads.
I've started making a Docker image for paparazzi, based on baseimage docker, however it's not clear to me what are the system requirements for tangram-es?
Does it need a system with a serious OpenGL video card, like the Amazon GPU ones? Or it can run on a dedicated server with an Intel card, like budget servers from OVH? Or possibly it can even run in a simple Docker container without access to any host hardware?
For example, https://github.com/klokantech/tileserver-gl can run in a Docker container and rasterise MapBox GL maps, without access to a "real" GPU.