Mesa 19.0 Can Cut In Half The Amount Of Memory For Team Fortress 2

[phoronix]

Phoronix: Mesa 19.0 Can Cut In Half The Amount Of Memory For Team Fortress 2

Timothy Arceri of Valve's open-source Linux graphics driver team has landed patches in Mesa 19.0 that drastically reduce the amount of system memory used when firing up the Team Fortress 2 game...

http://www.phoronix.com/scan.php?page=news_item&px=Mesa-19.0-Halves-TF2-Memory

[M@yeulC]

Couldn't Steam cache shaders before compilation as well, then handle compiling them in the background before starting the game/as a background process while there are some spare resources?

Also, couldn't we imagine a quick compiler pass to approximate the output of a shader, to avoid stuttering while compiling the real one? I think some projects already do it, but pointers would be appreciated.

[nanonyme]

Well, they kind of do. Steam has shader cache pre-seeding.

[cybertraveler]

I don't understand the problem here.

If you pre-compile all shaders for a game, why not just save them to disk so zero memory is used. Now when the shaders are actually required you load them from disk into memory. This ensures that only the minimum amount of system memory is required.

Despite me not understanding the problem, I found this article fascinating. Thanks Michael

[Veerappan]

Couldn't Steam cache shaders before compilation as well, then handle compiling them in the background before starting the game/as a background process while there are some spare resources?

Steam does ship precompiled shader for some games/systems, but there are limits because the message precompiled shader are mesa version, llvm version (for Radeon), and GPU specific. That'd be a lot of combinations to precompiled and ship, given that you usually have to launch the game to get the combinations of compiled and linked shaders

Also, couldn't we imagine a quick compiler pass to approximate the output of a shader, to avoid stuttering while compiling the real one? I think some projects already do it, but pointers would be appreciated.

The Radeon drivers already do this one, I think. They compile the shaders the first time without optimizations (if not already cached) and then build optimized versions in a background thread to swap in once they're ready.

[Aleksei]

Good news. Another optimization path would be to remove Spy class from the game - less models to load and whatnot. Valve, please consider it. #removespy

[faph]

Not impressed...
Who cares about 800 MB of RAM usage??? (if it would be VRAM that would be another story...)

Loading a savegame in BATTLETECH takes at least 30-40 seconds, so thanks to this update I can expect to wait even longer?

[cybertraveler]

Not impressed...
Who cares about 800 MB of RAM usage??? (if it would be VRAM that would be another story...)

Loading a savegame in BATTLETECH takes at least 30-40 seconds, so thanks to this update I can expect to wait even longer?

There's a lot of people that play video games on old computers; especially in poorer countries. There's also people playing games on very portable, non-performance oriented "work" laptops. Games like TF2 are popular with these people. Free to play, actively developed, low system requirements & huge community.

[Weasel]

Not impressed...
Who cares about 800 MB of RAM usage???

Who cares about your opinion? I hope it will take you 5 minutes to load it from now on.

[smitty3268]

Not impressed...
Who cares about 800 MB of RAM usage??? (if it would be VRAM that would be another story...)

Loading a savegame in BATTLETECH takes at least 30-40 seconds, so thanks to this update I can expect to wait even longer?

It can sometimes be important for 32-bit games which only have a 4GB address space. That's surprisingly easy to max out and crash some titles.

I do think they could skip this for native 64 bit games, though. They probably just want to reduce the differences between 32 and 64 bit paths.