Trying Out Mesa's KHR_no_error With An Intel Pentium + Radeon RX 580
With a lot of work going in recently to Mesa's KHR_no_error implementation for being able to optionally disable some error checking/handling within the OpenGL stack for potentially some CPU savings, I did some fresh tests of this feature (also known as MESA_NO_ERROR) when having the Kabylake Pentium CPU installed for the earlier Mesa GL threading tests.
KHR_no_error work that's been ongoing for Mesa is about turning off some OpenGL error checking in exchange for some CPU power/work savings but the potential for undefined behavior if an OpenGL error is to occur. This support is most easily forced right now by setting the MESA_NO_ERROR=1 environment variable.
I ran some tests of Mesa 17.2-dev as of this week on the Radeon RX 580 while the Intel Pentium G4600 was the processor.
Similar to past KHR_no_error tests, the gains appear to be incredibly small to non-existent depending upon the title, even when using this Pentium CPU.
The CPU usage wasn't measurably different on this modern dual-core Pentium + HT CPU.
So overall not really exciting, even with a sub-$100 Intel CPU. But with most of the big ticket performance items for Mesa OpenGL having been addressed especially on the RadeonSI side, the developers are left on tweaking the remaining finer items for squeezing out every last frame.
KHR_no_error work that's been ongoing for Mesa is about turning off some OpenGL error checking in exchange for some CPU power/work savings but the potential for undefined behavior if an OpenGL error is to occur. This support is most easily forced right now by setting the MESA_NO_ERROR=1 environment variable.
I ran some tests of Mesa 17.2-dev as of this week on the Radeon RX 580 while the Intel Pentium G4600 was the processor.
Similar to past KHR_no_error tests, the gains appear to be incredibly small to non-existent depending upon the title, even when using this Pentium CPU.
The CPU usage wasn't measurably different on this modern dual-core Pentium + HT CPU.
So overall not really exciting, even with a sub-$100 Intel CPU. But with most of the big ticket performance items for Mesa OpenGL having been addressed especially on the RadeonSI side, the developers are left on tweaking the remaining finer items for squeezing out every last frame.
Add A Comment