Trying Out MESA_NO_ERROR / KHR_no_error Support On Mesa 17.2-dev
Merged last month into Mesa Git and improved since then with follow-up commits has been KHR_no_error support for reducing the overhead of the OpenGL drivers by disabling certain error handling for OpenGL games/applications. This in turn can free up some CPU utilization and possibly lead to power-savings too.
If a game/application isn't error-free, however, this could lead to undefined behavior when disabling this error handling. Mesa's error handling with this new support can be disabled via the MESA_NO_ERROR=1 environment variable.
I tried out Mesa 17.2-dev Git as of this week with an out-of-the-box run and then a run under MESA_NO_ERROR=1. Tests were done on the Radeon RX 470 test box. During the testing process, the Phoronix Test Suite as monitoring the CPU utilization with the environment variable MONITOR=cpu.usage having been set.
Any game appears to be very minimal benefit or even non-existent, at least on this Intel Xeon E3 v5 + RX 470 box. At least though I didn't encounter any noticeable problems with MESA_NO_ERROR enabled.
The CPU usage on this Xeon rig didn't appear to be measurably impacted. So nothing really exciting, but if you want to look at all the data in detail, see this OpenBenchmarking.org result file.
If a game/application isn't error-free, however, this could lead to undefined behavior when disabling this error handling. Mesa's error handling with this new support can be disabled via the MESA_NO_ERROR=1 environment variable.
I tried out Mesa 17.2-dev Git as of this week with an out-of-the-box run and then a run under MESA_NO_ERROR=1. Tests were done on the Radeon RX 470 test box. During the testing process, the Phoronix Test Suite as monitoring the CPU utilization with the environment variable MONITOR=cpu.usage having been set.
Any game appears to be very minimal benefit or even non-existent, at least on this Intel Xeon E3 v5 + RX 470 box. At least though I didn't encounter any noticeable problems with MESA_NO_ERROR enabled.
The CPU usage on this Xeon rig didn't appear to be measurably impacted. So nothing really exciting, but if you want to look at all the data in detail, see this OpenBenchmarking.org result file.
22 Comments