NVIDIA/Radeon Windows 10 vs. Ubuntu Linux Relative Gaming Performance

[Steffo]

Yup i think these are kind of perf comparison articles people wants, plus ideally combined with videos like this and it would be superb

1. Michael is alone doing all this...
2. Can you tell on this video the difference between 50 and 70 frame rates?

[schmidtbag]

Sorry but ports are ports. Even on Consoles <-> Windows, people are always amazed that you need beefy PCs to run windows games that run pretty decently on consoles that have pretty much shit hardware.

I'm aware, but we're not comparing ports from consoles, but rather porting from one PC platform to another. Porting from consoles to Windows is not comparable to porting from Windows to Linux. Most of the dirty work has already been done. So yes, very often games that run worse on Linux are in fact just poorly ported. Sometimes DirectX's integration with Windows offers it a performance enhancement too. But sometimes when a game performs worse, neither of those situations may be true.

Looking at the graphs supplied by the article, the performance drops seem a little too consistent when you consider the variation of titles tested. Most of the tests where Linux does worse, it seems to average at around 70%. That leads me to believe that it's not just poor porting, or, inherent losses in inefficiency. I'm not saying that isn't a factor, but it seems to be more than that.

EDIT:
When you consider the tests where Linux performs as good or better than Windows, the advantages of DX suggests that the inherent losses of OpenGL are not the issue.
When you consider that both AMD and Nvidia tend to have the same percentage loss of performance, that suggests the drivers are not the issue.
When you consider that Linux still falls behind in Vulkan, that too suggests drivers are not the only issue (though AMD's drivers definitely seem to have a problem).
In order for poor porting to definitively be the blame, the percentage loss of performance should be a lot more variable. That being said, Civ VI is a good example of a poorly ported game.

[dungeon]

1. Michael is alone doing all this...

I also posting comments alone, Michael do it alone and also author of the video also do it alone, so?

2. Can you tell on this video the difference between 50 and 70 frame rates?

I can't of course, this is just a video and these 50/70 are just averages in whole... i am interested at stats and counters in this video more and generally in smooth rendering, high enough perf together with being more smooth represent good gaming experience, otherwise bad if you lose any of these two factors.

Or if you don't have both, it is clearly worse

[dungeon]

Looking at the graphs supplied by the article, the performance drops seem a little too consistent when you consider the variation of titles tested. Most of the tests where Linux does worse, it seems to average at around 70%. That leads me to believe that it's not just poor porting, or, inherent losses in inefficiency. I'm not saying that isn't a factor, but it seems to be more than that.

No one say this is just because of poor porting, as we speak of 70% average then remaining 30% maybe we can divide to also as average. So, 10% to poor port, second 10% to poor driver and another 10% to poor OS

[darkbasic]

2. Can you tell on this video the difference between 50 and 70 frame rates?

The video is just to show if it renders the same or not...

[dungeon]

The video is just to show if it renders the same or not...

This video shows it even better, so conclusion is obviosly not the same... most noticable is that people moving randomly and different

[schmidtbag]

No one say this is just because of poor porting, as we speak of 70% average then remaining 30% maybe we can divide to also as average. So, 10% to poor port, second 10% to poor driver and another 10% to poor OS

Generally speaking that's a fair assessment, but the numbers seem to tell a more complicated story than that.

[dungeon]

Generally speaking that's a fair assessment, but the numbers seem to tell a more complicated story than that.

Of course it is rough estimate but story is much more complicated, as if one just change kernel let say on Ubuntu to lowlatency one story might be different, not to change whole picture but slightly different and so on ... beside drivers and game itself, whole performance depend on all components of OS and all dependencies which game uses, etc...

[indepe]

Yup i think these are kind of perf comparison articles people wants, plus ideally combined with videos like this and it would be superb

(video)

The statistics and measurement graphs at the end of the video are very interesting.

I was always wondering why, for the ported games, the port was slower, yet the speed still depended very much on the graphics card. In general I would assume that buffering between the GPU and CPU allows them to work at the same time, so that only the slower one will determine the speed. But here it seemed different.

The CPU usage graphs show that all cores are being used, but lower than 80%, while the GPU usage is almost always at the top. So maybe the performance difference is because the GPU commands sent to the GPU are less performant. (Like a compiler producing worse code.)

This could be either due to the port's specific translation into OpenGL, because of OpenGL itself, or because a translation of DirectX calls into OpenGL calls doesn't allow using OpenGL's faster features.

[starshipeleven]

I'm aware, but we're not comparing ports from consoles, but rather porting from one PC platform to another. Porting from consoles to Windows is not comparable to porting from Windows to Linux.

Huh? To run Windows games on linux they need to slap some kind of DX-to-OGL wrapper on top of something that was designed to use DX, period. AAA games are not designed to use openGL or even Vulkan in decent ways, they use mostly-DX11 engines that are barely using DX12.

When you consider the tests where Linux performs as good or better than Windows, the advantages of DX suggests that the inherent losses of OpenGL are not the issue.

It's not inherent losses of OpenGL, it's that to port a Windows game they need to wrap DX calls into OpenGL at some point in the game's stack, that adds overhead, you can't avoid doing so without rewriting the engine to use OpenGL in a multi-platform way in the first place.

In order for poor porting to definitively be the blame, the percentage loss of performance should be a lot more variable.

I'm not blaming "poor porting" because I'm not a moron like duby.
I said "porting". Even good ports give a performance hit due to reasons I said above in this post.
Performance hit of good portings should be more or less the same, as this is induced by the wrapper they used or by design differences between DX and OGL that can't be dealt with by the porting.