17-Way Linux Graphics Card Comparison With Civilization Beyond Earth

[faldżip]

It's just that then we'd be testing the quality of the port, not the drivers. In fact, the only meaningful comparison even for that would be to test nvidia/Linux against nvidia/windows. Then we'd at least have a best case reference point.

I agree that testing drivers is important, but we see it twice a week on phoronix, while it would be nice to see overall linux gaming capabilities. For example, it is nice that opensource radeon drivers are as fast as proprietary ones, but it means nothing if for the decent gaming experience you still need to use windows (and I am talking about AAA titles). I'm happy to see more and more games coming to linux, but I'm afraid that there's still a long way to go.

[Kano]

It is hard to say you should stay with radeon oss when a game does not run. Kanotix has got a simple method to switch between binary and mesa drivers but i don't think that is so much fun to use AMD hardware for gaming.

The interesting thing about this game is the hard limit of about 50 fps for all newer AMD cards. As it is the same for a variety of cards, it can not be thruput related, only CPU bound is a weird assumption as well, most likely fglrx gets stuck in a lock between concurrent threads. Especially fglrx devs should look at it carefully, but I would not expect wonders between the years. They could fix a race condition for their integrated vaapi/xbva wrapper too.

After a look at the Nvidia benchmarks i would not say that this engine is heavyly optimized as well. There are only very few engines that are fully OpenGL optimizied, this one is not. I would like to see Rage or other ID games with full Linux support, those games did not even have got a D3D backend and should be simple to adopt to Linux, maybe just with differnet Video/Audio support, but that should not be critical.

AMD introduced mantle on Windows to get rid of CPU bottlenecks, this game could use mantle but where is the Linux variant? I understand that AMD wanted to create something that games run on slower CPUs better - well better: on cpus with bad single core performance like their own ones. But why on earth is the multithreaded performance so extremely bad then as they REALLY should use more than 1 thread for the driver - and thats on a much faster Intel CPU.

[carewolf]

This is one of AMD's promotional games. They probably had all the connections in place to get themselves a good showing on Linux... if they wanted to.

On Windows using Mantle. This is the port done by another company who obviously doesn't have the same relationship with AMD.

[carewolf]

It is hard to say you should stay with radeon oss when a game does not run. Kanotix has got a simple method to switch between binary and mesa drivers but i don't think that is so much fun to use AMD hardware for gaming.

The interesting thing about this game is the hard limit of about 50 fps for all newer AMD cards. As it is the same for a variety of cards, it can not be thruput related, only CPU bound is a weird assumption as well, most likely fglrx gets stuck in a lock between concurrent threads. Especially fglrx devs should look at it carefully, but I would not expect wonders between the years. They could fix a race condition for their integrated vaapi/xbva wrapper too.

After a look at the Nvidia benchmarks i would not say that this engine is heavyly optimized as well. There are only very few engines that are fully OpenGL optimizied, this one is not. I would like to see Rage or other ID games with full Linux support, those games did not even have got a D3D backend and should be simple to adopt to Linux, maybe just with differnet Video/Audio support, but that should not be critical.

AMD introduced mantle on Windows to get rid of CPU bottlenecks, this game could use mantle but where is the Linux variant? I understand that AMD wanted to create something that games run on slower CPUs better - well better: on cpus with bad single core performance like their own ones. But why on earth is the multithreaded performance so extremely bad then as they REALLY should use more than 1 thread for the driver - and thats on a much faster Intel CPU.

The Linux variant of Mantle is using best practices with OpenGL 4.x, but that can be hard unless the whole engine is made for batching everything like that.

[Ardje]

And we all know NVidia has it's own... quirks with the way it handles OpenGL and stuff so it's definitely no surprise that the game runs like shit on any other cards.

One of those things is running gl shaders that are obviously coded wrong, and hence would emit a shader error on a correct implementation, but nvidia swallows it. And that leaves the status to: "It works on my nvidia card, something must be wrong with your setup".

[AnonymousCoward]

There is clearly something wrong with game itself. Because it is limited on nVidia GPU too. http://boilingsteam.com/civilization...ower-on-linux/ Only difference is that AMD is capped to 33FPS and nVidia 55.

[dungeon]

There is clearly something wrong with game itself. Because it is limited on nVidia GPU too. http://boilingsteam.com/civilization...ower-on-linux/ Only difference is that AMD is capped to 33FPS and nVidia 55.

Yeah for sure, both are capped Someone needs to lower down these settings and enable one per one option more, so hopefully that way he can found what effect/settings actually holding it

[Kano]

Did you notice that vendor specific driver optimisation (based on filename) could replace shaders by optimized ones? Something like that was tried for older source engine games (those which use the hl2 binary) but not for csgo - with limited success. But i don't think AMD will invest so much time for opimisations for 1% of their customers. But at least they should fix general problems.

[dungeon]

There is clearly something wrong with game itself. Because it is limited on nVidia GPU too. http://boilingsteam.com/civilization...ower-on-linux/ Only difference is that AMD is capped to 33FPS and nVidia 55.

Also blogger said, "Phoronix does not disclose anything but resolution" he can look in config there: http://openbenchmarking.org/innhold/...bde7a09a0be612

[GraphicsSettings]
OverlayLevel = 2
ShadowLevel = 2
ReflectionLevel = 0
FOWLevel = 1
TerrainWaterQuality = 0
VFXQuality = 1
FadeShadows = 0
Enable Constant Rebasing = 1
Enable Threaded Rendering = 1
Enable Bloom = 1
Enable UIBlur = 1
Enable DoF = 1
Version = 3
Enable MGPU = 0

[ScreenSettings]
MSAASamples = 1
WaitForVSync = 0
Refresh Rate = 0
WindowResX = \$1
WindowResY = \$2
FullScreen = 1
StereoConvergenceMin = 0.000000
StereoConvergenceMax = 0.000000
StereoCursorOffset = 0.000000
ScreenShotWidth = 0
ScreenShotHeight = 0
DisableAdvancedAAModes = 0
Version = 1

[TerrainSettings]
TerrainDetailLevel = 2
TerrainTessLevel = 2
TerrainShadowQuality = 3
TerrainPageinSpeedStill = 6
TerrainPageinSpeedMoving = 3
BlockOnLoad = 0
AutoUpdateCells = 1
Version = 1

[LeaderheadSettings]
LeaderTextureReduction = 0
LeaderTextureBackgroundLoad = 0
AllowSM41 = 1
AllowSM50 = 1
AllowLeaderAA = 1
EnableShadows = 1
EnableSoftShadows = 1
CubeShadowResolution = 1024
MaxResidentScenes = -1
LeaderQuality = 3
SkinQuality = 2
SkinResolution = 1024
AspectAdjust = 0.000000
TargetAspect = 1.777778
EnableBloom = 1
EnableDistortion = 1
UseScreenShots = 0
UseGPUDecompress = 1

Now, where are those configs for Windows game benchmark settings, so we can compare?

[dungeon]

Same config works fine on Windows , so someone who has both Windows/Linux versions should benchmark it with that one at first, because well that make more sense:

http://paste.debian.net/plain/137755

Copy>Paste that in both Windows/Linux GraphicsSettings.ini config.