01-19-2020, 11:23 AM
(This post was last modified: 01-19-2020, 11:31 AM by Brian Beuken.)
I had to build it myself using these instructions, which also work for ubuntu systems, (*though ubuntu can apt-get it normally) you just need to change the final settings to make it GLES2
so i'm using GLMark2-es2 not GLMark2 which is going to be unpredictacble on GLES systems, even those that claim to do full fat OpenGL.
https://www.pcsuggest.com/install-glmark2-debian/
I'm just using standard Tinkerboard OS, which I think is based on Debian. I should probably update it again to see if the downgrade has been fixed. I always try to avoid exotic OS's, and work with the bog standard OS supplied.
The main thing is that many boards have GLMark2-es2 locked to frame rate, which means that when you run it will never be more than 50 or 60fps depending on the refresh rate, so the --off-screen option gives you a truer measure of performance.
Emulated systems usually report a standard mesa gles3.0 render system and nearly always show the 1st 2 tests as the same values, so there's no buffering in action. As such you're looking at CPU power rather than GPU power. the Raspberry3's default to emulation becuase GLMark2 can't instansiate a dispman render surface so its not a very accurate example, I guess if I was feeling crazy I could try to add that to the source code but it's a lot of effort.
so i'm using GLMark2-es2 not GLMark2 which is going to be unpredictacble on GLES systems, even those that claim to do full fat OpenGL.
https://www.pcsuggest.com/install-glmark2-debian/
I'm just using standard Tinkerboard OS, which I think is based on Debian. I should probably update it again to see if the downgrade has been fixed. I always try to avoid exotic OS's, and work with the bog standard OS supplied.
The main thing is that many boards have GLMark2-es2 locked to frame rate, which means that when you run it will never be more than 50 or 60fps depending on the refresh rate, so the --off-screen option gives you a truer measure of performance.
Emulated systems usually report a standard mesa gles3.0 render system and nearly always show the 1st 2 tests as the same values, so there's no buffering in action. As such you're looking at CPU power rather than GPU power. the Raspberry3's default to emulation becuase GLMark2 can't instansiate a dispman render surface so its not a very accurate example, I guess if I was feeling crazy I could try to add that to the source code but it's a lot of effort.
Brian Beuken
Lecturer in Game Programming at Breda University of Applied Sciences.
Author of The Fundamentals of C/C++ Game Programming: Using Target-based Development on SBC's
Lecturer in Game Programming at Breda University of Applied Sciences.
Author of The Fundamentals of C/C++ Game Programming: Using Target-based Development on SBC's