Mesa OpenGL Threading Now Ready For Community Testing, Can Bring Big Wins

[marek]

Is Mesa 17.1.4 + LLVM 4.0.0 (default Fedora 26 install) sufficient to test this properly? I know glthread makes a difference (I tested it with Bioshock Infinite), but I don't know whether I'm not missing some important commits currently available only in master.

For BioShock, I used Core i5 3570 and Radeon Fury X = decent CPU, fast GPU.

[gnarlin]

Maybe if the phoronix test suite could automatically turn threading on and off and mark the setting in the results that could be useful for finding the regressions?

[schmidtbag]

Of the games that do have regressions, how severe are they? If the game still gets playable framerates, I don't think it's necessary to focus on blacklisting or whitelisting, assuming most games aren't negatively affected.

In the event a game does have weirdly bad regressions, I wonder if that correlates to the CPU itself. For example: Marek says he tests on an i5. What if the regressions happens because a game already utilizes 4 threads, leaving the CPU over-stressed? I guess what I'm getting at is perhaps this should be disabled on CPUs with 4 total threads or fewer, but with the option to override it.

[marek]

Of the games that do have regressions, how severe are they? If the game still gets playable framerates, I don't think it's necessary to focus on blacklisting or whitelisting, assuming most games aren't negatively affected.

The sad reality of the industry is that customers buy GPUs based on published game/benchmark results, not based on whether framerates are playable or "good enough".

In the event a game does have weirdly bad regressions, I wonder if that correlates to the CPU itself. For example: Marek says he tests on an i5. What if the regressions happens because a game already utilizes 4 threads, leaving the CPU over-stressed? I guess what I'm getting at is perhaps this should be disabled on CPUs with 4 total threads or fewer, but with the option to override it.

4 cores/4 threads is plenty. 2 cores might not be enough, but can still be better than without glthread.

[caligula]

On a mid-range GPU with this mesa_glthread functionality enabled, Alien Isolation is faster by 60%, Border Lands 2 by 50%, BioShock Infinite by around 76%, and Civilization 6 by about 20%

Holy crap. How can it speed up that much?

[Nille_kungen]

The sad reality of the industry is that customers buy GPUs based on published game/benchmark results, not based on whether framerates are playable or "good enough".

It's more like what's in stock due too miners.
And also i think there is a big difference in how linux users choose graphics cards compared to windows users.
I think i would find myself in the "good enough" category.
But i think the reality is that most customers doesn't know what there buying and is left out to the sellers recommendation.
Thanks for this, i think i will play some Alien Isolation with mesa git now.

[schmidtbag]

The sad reality of the industry is that customers buy GPUs based on published game/benchmark results, not based on whether framerates are playable or "good enough".

AMD has been trying to catch up to Nvidia for a while, yet that doesn't stop AMD from making sales. People are slowly transitioning to AMD on Linux due to the overall better user experience and complacency with Linux. People were doing this despite the Mesa drivers performing overall worse than the Windows drivers. Only very recently have the drivers begun to be on-par with Windows, with some exceptions that seem to have a correlation with detail level. I'm sure many people are willing to sacrifice 5-10FPS on a game that already exceeds 150FPS if it means games that struggle to reach 60FPS can go faster. But again, I don't know how severe the regressions are or what the cause is.

4 cores/4 threads is plenty. 2 cores might not be enough, but can still be better than without glthread.

But is it plenty in the scenario I described? I'm aware almost no games in Linux take advantage of more than 4 threads, but to my recollection there are some that do see a performance difference between 3 and 4. The way I see it, if a game sees a noticeable performance gain when adding more cores (without GL threading) then surely the CPU is under too much stress when GL threading is introduced.

At the very least, I think it would be worth testing games with regressions on both 4 thread and 8 thread CPUs. If the regressions persist, then I'd say continue with your current plans. If the regression goes away then I'd say GL threading should blacklist thread count rather than games. Or, perhaps games based on thread count.

[SaucyJack]

Of the games that do have regressions, how severe are they? If the game still gets playable framerates, I don't think it's necessary to focus on blacklisting or whitelisting, assuming most games aren't negatively affected.

In the event a game does have weirdly bad regressions, I wonder if that correlates to the CPU itself. For example: Marek says he tests on an i5. What if the regressions happens because a game already utilizes 4 threads, leaving the CPU over-stressed? I guess what I'm getting at is perhaps this should be disabled on CPUs with 4 total threads or fewer, but with the option to override it.

...is that a joke? So who gets to decide what's playable then? I don't consider anything below 90 fps playable. I'd get a console if I did.

[schmidtbag]

...is that a joke? So who gets to decide what's playable then? I don't consider anything below 90 fps playable. I'd get a console if I did.

Even if I were to give you the benefit of the doubt that you actually are enough of an elitist where 90FPS is your bare minimum (I know of people with G-sync displays who are perfectly satisfied with 75FPS), where did I say anything about a specific framerate being playable? More importantly, how does your comment have anything to do with the snippet you quoted? Are you not aware that most AMD hardware on Linux does not regularly reach 90FPS on modern AAA titles?

If anything here is a joke, it's your level of comprehension. I'm trying to focus on improving everyone's experience in the most efficient way possible, so go troll somewhere else.

[smitty3268]

Even if I were to give you the benefit of the doubt that you actually are enough of an elitist where 90FPS is your bare minimum (I know of people with G-sync displays who are perfectly satisfied with 75FPS), where did I say anything about a specific framerate being playable? More importantly, how does your comment have anything to do with the snippet you quoted? Are you not aware that most AMD hardware on Linux does not regularly reach 90FPS on AAA titles?

If anything here is a joke, it's your level of comprehension. I'm trying to focus on improving everyone's experience in the most efficient way possible, so go troll somewhere else.

No offense, but some of your comments in this thread don't make much sense.

The wide range of hardware, games, settings, and how people define playable means that almost any loss at all is probably going to take some game from playable to non-playable for someone. The only question is at what resolution and for what card.

Also, if a game speeds up with increasing core counts, that doesn't mean this wouldn't help it. It's all about allowing the game to continue doing work rather than stalling out while transferring assets into memory or blocking on some other operation. That can help even if you only have a single-core cpu.