Trying Out The New OpenGL Threaded Dispatch In Mesa 17.1

[Michael]

I'd say that having the same frame rate with and without glthread on low settings is fairly unlikely. I don't know what went wrong there, whether steam didn't pick up the environment variable or whether PTS wiped out the environment.

Instead of messing with environment variables, you can create ~/.drirc with the following:

<driconf>
<device>
<application name="Default">
<option name="mesa_glthread" value="true" />
</application>
</device>
</driconf>

If you already have ~/.drirc with something in it and you don't know why, remove that, because generally ~/.drirc shouldn't mess with default Mesa settings in any way.

Steam seemed to pickup the env var and set it in the same way I did earlier for the shader cache var to go into effect. I thought the numbers looked a bit odd but when carrying out all the tests an additional time the results ended the same way, with regard to when glthread was set those few tests regressed and the rest stayed the same. PTS doesn't override any mesa_* vars. But anyhow will try it with the DRIRC method and see if it yields any different results.

[prazola]

How can I test this with steam games? Should I add an export in sh launchers or can I simply add "mesa_glthread=true" to the game launch options form? Ty

[lordheavy]

How can I test this with steam games? Should I add an export in sh launchers or can I simply add "mesa_glthread=true" to the game launch options form? Ty

yes, mesa_glthread=true %command%

[theghost]

I don't know but debianxfce's explanation seems pretty valid, assumed that Michael used Oibaf's PPA.
Btw. I don't know why Mesa Git tests are always run on Ubuntu. Always messing with PPA's and outdated LLVM.
Just install Arch or Tumbleweed for one time and switch on Mesa-git...

[prazola]

yes, mesa_glthread=true %command%

Ty. Performance loss in DE:MD, -12FPS avg. I would try Metro LL redux but it seems there's no integrated benchmark...is it only in the first episode?

[TheOne]

even if Michael tests are screwed the results reported by other people here having a 2x increase in FPS is marvelous , finally we are seeing some nice performance changes for mesa , even if it wouldn't work well with all games.

[LeJimster]

This is because all these Linux games are ported to OpenGL from DX for the most part. OpenGL isn't as well threaded therefore it doesn't show very good multicore scaling. With Intel having fewer, higher clocked cores even if the IPC is rather similar they get better frame rates (especially for OpenGL.)

Maybe I can buy that as an explanation. But then why does radv performance tank so bad. Vulkan is multi threaded by design.

[boffo]

Maybe I can buy that as an explanation. But then why does radv performance tank so bad. Vulkan is multi threaded by design.

mesa_glthread=true is for OpenGL (some or all drivers?), not for radv.

[mmstick]

Maybe I can buy that as an explanation. But then why does radv performance tank so bad. Vulkan is multi threaded by design.

It's an issue with the Vulkan driver tested.

[linuxgeex]

Strange. Something is obviously not working correctly. I remember seeing old benchmarks on borderlands and their was a big performance boost.

Reading other people testing it they noticed no framerate gains but claim it was often a smoother experience wit it enabled.

Hopefully we can see more testing and work done to figure out why its seemingly not giving the performance boost expected.

Yes it would be interesting to see how close to TDPmax the card was operating. Multithreading can't help if the card has nothing left to give.

@Michael: Maybe frame latency, thermals, and wattage would help shed some light on this.