Windows 10 vs. Linux 4.15 + Mesa 17.4-dev Radeon Gaming Performance

[papajo]

Did any of that work (significantly) better on Nvidia than on Mesa? I don't have any of those games except Tomb Raider so I can't really compare.

In my experience, games that worked well on my GTX 670 still run well or even better on my RX 480 (among them: Metro Last Light, Metro 2033 Redux, Witcher 2, Bioshock Infinite), and games that don't run well on rhe 480 didn't run any better on my 670 either (Tomb Raider being one of them).

The fundamental issue is not that the drivers from either vendor are bad, it's just that those D3D11->OpenGL ports tend to require absurd amounts of CPU power compared to Windows D3D11, and your FX 8320 simply doesn't deliver on that front (and neither does my overclocked 1090T).


It sort of works when you set the h264 profile to mainline (default is high although it is known not to work).

But yeah, if you intend to stream games with it, well, you are indeed screwed. This still needs a lot of work.

Are you sure you use the latest nvidia drivers? maybe you use nouveou?

Because for example the guy I showed you in my previous post (in case you did not check it out https://www.youtube.com/watch?v=uF62TIW_NmU ) plays on ubuntu mate 16.04 and uses the nvidia 361.42 drivers and has a pretty solid frame-rate ingame (which I noticed is much more taxing than the actual benchmark)

At this point it would not surprise me that linux driver could for some reason respond significantly differently on different systems (with same GPU but e.g different CPU and RAM ) lol

Since you mentioned it the Witcher 2 plays fine as well I just forgot about that ... I didnt really play it in Ubuntu that much but the few seconds I tackled with it, it was a steady 60 FPS (or it had vsync on ) I am going to play it a little now just to check if it keeps that frame-rate throughout the game.

Also as I mentioned before.. avg FPS is misleading in the linux benchmarks because (at least in my rig I dont know if I am the exception) the jittering caused by frame-drops is real, noticeable and frequent.

And mind you that I set the CPU governor to performance and generally have my system tweaked a little to perform on games I just didnt install the latest upstream kernel or any firmware ,, I even tried to disable the compositor to see if I squeeze a few FPS (on xorg because wayland is made in such a way that you cant disable the compositor )

And yes I also checked my performance without those tweaks (and before even enabling them for the first time ) and it was same slowish as after so my bad performance is not because I made a noob setting that slows down my system.

and your FX 8320 simply doesn't deliver on that front

I thought about that but a) its an OCed one b) I dont reach 100% in all the cores and rarely reach 100% in a few ones c) whatever bottleneck is caused by my CPU is surely not the reason for such a bad performance I mean it doesn't suck that much maybe its responsible for 10% framedrop or something... or in other words if my GPU was 10% faster than what it is now it still would suck....

[VikingGe]

Because for example the guy I showed you in my previous post (in case you did not check it out https://www.youtube.com/watch?v=uF62TIW_NmU ) plays on ubuntu mate 16.04 and uses the nvidia 361.42 drivers and has a pretty solid frame-rate ingame (which I noticed is much more taxing than the actual benchmark)

And that part of the game isn't even remotely as taxing as the rest of the game. Getting decent-ish performance on Mesa as well with FPS ranging anywhere from 40 to 80, depending on how much geometry the game renders (screenshot). And that's not even with a super-optimized Mesa build, just the one that currently ships with Arch. Move on in the game and you will drop back into the 20s though, or even below that.

Also as I mentioned before.. avg FPS is misleading in the linux benchmarks because (at least in my rig I dont know if I am the exception) the jittering caused by frame-drops ir real, noticeable and frequent.

The games that do run at an acceptable framerate feel pretty smooth to me, and actually that's something that the performance governor should help with.

I also make sure to run my GPU at maximum frequency when gaming, although that affects Wine a lot more than native Linux games. Still, might be worth testing.

[papajo]

And that part of the game isn't even remotely as taxing as the rest of the game. Getting decent-ish performance on Mesa as well with FPS ranging anywhere from 40 to 80, depending on how much geometry the game renders (screenshot). And that's not even with a super-optimized Mesa build, just the one that currently ships with Arch. Move on in the game and you will drop back into the 20s though, or even below that.

Mate trust me I played this game recently 40 to 80 FPS is only plausible in the benchmark... during the actual game it doesnt come even near... yes that part of the game is not the most taxing it starts being taxing when those russian guys try to kill you but never the less when I played that part of the game my FPS was 40-50 (with occasional noticeable frame-drops to 20-30)on a mix of high and normal settings with AA set to 2 this guy plays the same part at 50 to 60+ at ultra.

And while being on yesteryears older software/ drivers and hardware... I may not be on bleeding edge but for my distro I am pretty much up to date and my GPU should be much better than his (in windows it is competitive/better than a GTX 1060 which is better than a GTX 980 which is better than a GTX 780 which is better than a GTX 680 )

But a GTX 680 in ubuntu 16 +old nvidia drivers (current for when ubuntu 16 was the latest version) > RX 580 Ubuntu 17.10 MESA 17.3

Also I already told you that I set all cores with gpu governor to performance.

[papajo]

This comparison came handy. I was just about to remove my windows partition, since I play ported to linux games. Now I am not sure I can afford 100% performance hit on my mid-end rig. Too bad.

If you use a recent nvidia GPU (GTX 970 and above) your experience will be ok (not as good as windows but many games will run on a playable framerate at decent or ultra settings especially the vulkan ones will be identical to windows performance)

If you are looking for an AMD from what I experience my self I would advise you either to not get an AMD GPU or wait until there is some significant breakthrough in the drivers and then buy an amd card.

I hear that older architectures (like r9 380x) have better support though so research a little on that if you are interested.

There is also a 3rd path instead of dual boot try to google "KVM windows gaming" you can have both worlds in one box without the need to reboot that way

[lucrus]

You need to use the terminal to install

That's not true, at least if you use Ubuntu. However I think the problem here is papajo wants a AMD desktop Linux gaming end user experience similar to what NVidia delivers, but withtout the drawbacks you face with NVidia when you stop using it for gaming. That's not going to happen. Desktop Linux is not for the masses. Stop craving for it. Desktop Linux is being made by programmers/sysadmins, for programmers/sysadmins. Even Canonical has failed at bringing the desktop linux to the masses.

If you want a no-brainer desktop gaming experience AND a no-brainer desktop non-gaming experience, then switch back to Windows and stop using your brain.

If you prefer Linux even for gaming, that's because, for you, the install/configure/patch/upgrade part is fun too, and realizing that effort gains you features and FPS it is even more fun. And that's why we love open source and AMD. Yes, you could gain more features and FPS just by booting Windows, but you loose the fun.

Don't your friends and family think it's fun to smucinate between terminal spells of mighty kernel upgrades and library dependencies resolutions? I know, it's sad they don't get the fun part, but you can't do much to change that.

you can not fix closed source bugs by yourself.

Please. You (impersonal "you") cannot fix AMD drivers by yourself either. Maybe you ( debianxfce ) are able to fix drivers by yourself, but that's not the case for the average Linux user.

[VikingGe]

this guy plays the same part at 50 to 60+ at ultra.

Yeah, and that guy uses an i5, which is quite a bit faster than your FX in games. Comparing two entirely different systems when the game is CPU-bound just doesn't make any sense.

yes that part of the game is not the most taxing it starts being taxing when those russian guys try to kill you but never the less when I played that part of the game my FPS was 40-50 (with occasional noticeable frame-drops to 20-30)

that's literally what I just wrote. If you want to compare AMD to Nvidia drivers, you need to
a) compare otherwise identical systems, or at least the same CPU
b) compare the same part of the game.

[papajo]

Yeah, and that guy uses an i5, which is quite a bit faster than your FX in games. Comparing two entirely different systems when the game is CPU-bound just doesn't make any sense.


that's literally what I just wrote. If you want to compare AMD to Nvidia drivers, you need to
a) compare otherwise identical systems, or at least the same CPU
b) compare the same part of the game.

ok so this is my CPU utilization https://i.snag.gy/Adktfn.jpg when I am on this http://oi68.tinypic.com/fe409d.jpg part of the game where I have 18FPS (It constantly is 18 FPS on this area I dont know why the screenshot shows 19.. I took several screenshots all showing 19 I presume the steam overlay adds in the print screen as an additional frame or something... also I am jumbing because I was trying to creating a commotion by running fast and change camera angles and move crazy to "push" my CPU for the other screen-shot) on LOW settings (I dont even have tressFX hair ) .. to be fair the hit with high settings is not that bad on this part it gets to 10-12 rarely going to 9 or even 8 lol....

So you can see that CPU is not even taxed and can not be the cause of the bottleneck...

and as you can see here https://www.youtube.com/watch?v=ofctsTusLZw the difference is negligible between the i5 he uses and the 8320...

Yes I know in the video its a 8350 but its the 100% same CPU no difference whatso ever compared to mine other than the base clock speed which is just a little higher (and which in my case is higher anyway since I OCed mine as I told you)


So you can confirm that my CPU plays no role here..

further more even if it did impact my performance negatively the sheer horsepower difference of the GPUs (GTX 680 vs RX 580 ) should be enough to make me having much better framerates even if my CPU is bottlenecking my avg FPS.. because the card would be so strong to have a better avg FPS anyway...

but the drivers suck so I am even worse than the ancient GTX 680

[VikingGe]

ok so this is my CPU utilization https://i.snag.gy/Adktfn.jpg

Irrelevant. fire up the game with GALLIUM_HUD=GPU-load and look at that. Not even close to 100% -> CPU bottleneck.

Anyway, I'm not going to waste any more time on your apples vs oranges comparions that drive you to all the wrong conclusions. Just get a new CPU if you are serious about Linux gaming, for god's sake.

[papajo]

Irrelevant. fire up the game with GALLIUM_HUD=GPU-load and look at that. Not even close to 100% -> CPU bottleneck.

Anyway, I'm not going to waste any more time on your apples vs oranges comparions that drive you to all the wrong conclusions. Just get a new CPU if you are serious about Linux gaming, for god's sake.

I may not be a linux hardcore poweruser/dev but if I know something very good it is PC hardware and I can assure you that this 4th generation 4 core 3 year old devils canyon CPU from intel is not a big leap forward compared to my FX 8320 I would even argue that in some scenarios my CPU is better and in all the other scenarios where it is not, it is slower just by a small margin.

And I also am a windows power user and can assure you that hardware wise my system would destroy his system even if he OCes everything and I am on stock... like he would not even be close to my avg FPS...

So the difference has to be caused by the OS the port of the game or the drivers.

Since the OS he uses is the same as mine only just an older version (he uses 16 I use 17) and since the port of the game is the SAME then only one thing is remaining that could influence the outcome and that thing is drivers..

[VikingGe]

and since the port of the game is the SAME then only one thing is remaining that could influence the outcome and that thing is drivers..

You still don't understand that a D3D11->OpenGL port on Linux has completely different performance characteristics than a D3D11 game running on Windows and usually requires significantly more single-threaded CPU power. If you want to blame AMD for an i5 being faster than your FX, go ahead, but the GPU drivers have nothing to do with it.