Intel's Mesa On-Disk Shader Cache Maturing, Radeon Devs Not Yet Convinced

[pal666]

These days, if you have a single-core CPU, chances are GPU performance is not a priority.

shader cache does not affect gpu performance

[dungeon]

it has no stutter

Cool if does not now i did not tried it recently, so possible that Marek make some workaround there. Or we talk about different dirvers here? Or you might just have fastest CPU on planet, so did not spot an issue?

it's just cpu-bound

Well i know that generally MCCaps or __GL_THREADED_OPTIMIZATIONS or threadedGL feature missing

[duby229]

there is secret patented technology called page cache

wow!!

[schmidtbag]

shader cache does not affect gpu performance

And that's relevant because...?
The shader cache is inevitably for the GPU. My point is if you have shaders complex enough that caching them has a substantial impact on performance, then that means you're running something that a single-core CPU is probably not going to keep up with. Therefore, focusing on multi-threading mesa should offer a significant performance difference without the need of permanent storage.

EDIT:
Again - I have no problem with caching shaders; any performance enhancement is welcome. I just think it should've been a lower priority, and I'd like to know if I can [easily] disable it.

[stqn]

An on-disk shader cache is super important to speed up the loading times of Dead Island (the original, not so much the remake, but the remake runs slower and looks just like Dying Light, so it’s not as good). I’m sure there are other games that benefit significantly from it, but Dead Island is the worse example I know of. Deadfall Adventures also has super long loading times that are much shorter when you launch it the second time.

[dungeon]

I just think it should've been a lower priority

I would put CPU related issues on highest priority, by giving AMD Temash laptops to all developers (please do not complain devs, that is still double faster then what Stallman use ) But developers, eh those developers... they always like to use fastest CPU to save time on code compilation, so it is always lowest priority for them

and I'd like to know if I can [easily] disable it.

Windows drivers has GUI to click on/off it, we will probably have some console command line or so or if you compile mesa there is a switch to not compile it or you can change permission to not write to folder when it suppose to store caches , etc...

[SaucyJack]

This is stupid. Any dev who doesn't see a dire need for an on-disk cache doesn't play games. You want an insane example, go try out NS2. I recommend you have some coffee or a sandwich to make. Unreal Tournament suffers greatly as well. Trying playing a twitch FPS that microstutters (are full 1 second freezes micro at that point?) for the first half of every match due to on-the-fly shader compilation. The whole 'I don't see the benefit beyond x' argument is just idiocy. There's a reason the proprietary drivers use it.

[justmy2cents]

I think at least in theory, the linux file cache should handle that automatically? Once a file is read it should be cached to ram on the fly right?

unless some game abuses ram to insanity where first thing being dropped are things that are cached. loading it inside game guarantees persistence

as far as startup goes, there is no benefit in kernel caching unless you start game twice where only 2nd time would use them. there is also fact that reading one big file is much faster than reading 4000 small ones/

[smitty3268]

This is stupid. Any dev who doesn't see a dire need for an on-disk cache doesn't play games. You want an insane example, go try out NS2. I recommend you have some coffee or a sandwich to make. Unreal Tournament suffers greatly as well. Trying playing a twitch FPS that microstutters (are full 1 second freezes micro at that point?) for the first half of every match due to on-the-fly shader compilation. The whole 'I don't see the benefit beyond x' argument is just idiocy. There's a reason the proprietary drivers use it.

A lot of the Unreal Engine 3 games are pretty bad at this, but apparently UE4 no longer needs it now that the GL4.3 path is working.

Anyway, I don't think Marek is against the idea of a shader cache, I just think he wants to take care of some other performance problems first which he thinks can help more people faster. For example, he just committed this: https://cgit.freedesktop.org/mesa/me...4a10e67cdee073 which looks like it will pretty massively reduce register spilling in games with complex shaders.

[bridgman]

Anyway, I don't think Marek is against the idea of a shader cache, I just think he wants to take care of some other performance problems first which he thinks can help more people faster. For example, he just committed this: https://cgit.freedesktop.org/mesa/me...4a10e67cdee073 which looks like it will pretty massively reduce register spilling in games with complex shaders.

There have also been conflicting views about how many games would actually benefit from a persistent shader cache (ie how many games compile shaders for the first time during the middle of gameplay)... there are still discussions going on about that.