The Talos Principle has another stable build with Vulkan improvements, much better than OpenGL

First, I love this game and I think Croteam are amazing.

GTX 1050 Ti (purchased because of solid performance with decent power draw):

Low - VK: 223.5 GL: 110.2
Med - VK: 132.9 GL: 80.1
High - VK: 64.7 GL: 46.4
Ultra - VK: 43.9 GL: 34.9

Personally, I play on settings that aren't much different than Medium since I can't see much change going higher than that. I originally played it using a GTX 750 (non Ti) at roughly medium settings in OpenGL and noticed no problems.

The performance jump from Vulkan is jaw-droppingly amazing. I was worried that Vulkan would be hard to switch on, but I did nothing more difficult than turning it on in the game settings.

Once developers get good at optimizing Vulkan across platforms, this has the possibility of really boosting Linux performance.

I rarely look at performance numbers and benchmarks, as I am happy when game runs with enough FPS for me to be able to play it. I would however be curious if there are OTHER components holding back Linux performance other than OpenGL. In other words, if a game uses Vulkan on both Windows and Linux, would it perform exactly the same on both Linux and Windows? Or would Linux still be worse?

OK, disclaimer, I'm no developer and I can't back this up.
But the general zeitgeist I have from following vaguely this sort of thing for years comes down to: Aside from OpenGL problems, up to now graphics card drivers for Linux have tended to be slower, although that situation too has been improving and the nature of Vulkan seems to be such that it probably won't be nearly as much an issue for Vulkan stuff. But OS to OS, Linux itself is probably faster than Windows, and certainly lighter, so it wouldn't hog as much CPU or memory for itself, leaving more for the game. The "lighter" part of course depends on specifics like, KDE or XFCE or what? As I say, this is my general impression which I cannot cite any sources to defend, so take it for what it's worth.

It's easy to be blinded by the AAA releases from Feral & Aspyr, but these guys here are real Linux gaming heroes and deserves every little bit of praise - and applause from our wallets - for their work.

Tried it on an AMD R9 380X here on Mesa 17.1-git.

The improvement is indeed impressive. While it's not the same improvement as the prop. NVIDIA tests, before Vulkan ran a fair bit worse than OpenGL, whereas now it's hard to split them, both averaging somewhere around 45FPS (not a proper benchmark, just my casual testing with the FPS counter on). Though Vulkan seems to be a bit more consistent in average FPS I believe.

Regardless, it's much nicer to play under the Vulkan renderer now on ultra graphics @ 1080p.

This is really awesome to see! It would be nice to see the Windows vs Linux comparison too :)

Very interested too.

I have a dream...

Still can't get it to run without crashing with Vukan on my home-built SteamOS brewmaster box.. :( Tried from the menus and with launch option.. boo

It's easy to be blinded by the AAA releases from Feral & Aspyr, but these guys here are real Linux gaming heroes and deserves every little bit of praise - and applause from our wallets - for their work.

I think you're right. I wasn't that keen on The Talos Principle having played the demo (it seemed like Portal without the humor and charm), but you're right, these guys should be rewarded for their efforts.

If anyone is interested I can run this on an i5-3570K with a 980Ti on both Windows and Linux on Vulkan, DX11, and OpenGL.

Okay, I did some benchmarks with DirectX 11 under Windows 10, Vulkan under Linux, and OpenGL under Linux. I ran the benchmark on a GTX 960 and an i7-6700K using the 60 second benchmark. If it says Ultra, then all settings were set to Ultra, for Medium, all settings were set to Medium, etc. If anyone wants to know a particular combination feel free to ask.

Here are the results (Edited: added the results from my followup post below to include Windows Vulkan and OpenGL results):

Vulkan (Linux):
Ultra: 41.6
Medium: 158.3
Lowest: 419.3

Vulkan (Windows):
Ultra: 42.6
Medium: 166.8
Lowest: 387.7

OpenGL (Linux):
Ultra: 36.5
Medium: 101.0
Lowest: 263.2

OpenGL (Windows):
Ultra: 36.3
Medium: 101.2
Lowest: 343.5

D3D 11 (Windows):
Ultra: 45.6
Medium: 172.6
Lowest: 355.2

So, we are getting much closer :) It's interesting when put on low settings that Vulkan completely smokes D3D11 by almost 70 fps. I'm wondering if this has to do with multi-threading vs the pseudo multi-threadedness of D3D11?


Last edited by natewardawg on 7 Mar 2017 at 5:44 pm UTC

Okay, I did some benchmarks with DirectX 11 under Windows 10, Vulkan under Linux, and OpenGL under Linux. I ran the benchmark on a GTX 960 and an i7-6700K using the 60 second benchmark. If it says Ultra, then all settings were set to Ultra, for Medium, all settings were set to Medium, etc. If anyone wants to know a particular combination feel free to ask.

Here are the results:

D3D 11:
Ultra: 45.6
Medium: 172.6
Lowest: 355.2

Vulkan Linux:
Ultra: 41.6
Medium: 158.3
Lowest: 419.3

OpenGL Linux:
Ultra: 36.5
Medium: 101.0
Lowest: 263.2

So, we are getting much closer :) It's interesting when put on low settings that Vulkan completely smokes D3D11 by almost 70 fps. I'm wondering if this has to do with multi-threading vs the pseudo multi-threadedness of D3D11?

Yeah,it is most likely that.Because on lowest settings cpu became the frame limiter,on highest gpu became limiter.Cpu utilization on Vulkan is better than Dx 11 normally because Vulkan is competitor to DX12;which it is also good on cpu utilization.

Okay, I did some benchmarks with DirectX 11 under Windows 10, Vulkan under Linux, and OpenGL under Linux. I ran the benchmark on a GTX 960 and an i7-6700K using the 60 second benchmark. If it says Ultra, then all settings were set to Ultra, for Medium, all settings were set to Medium, etc. If anyone wants to know a particular combination feel free to ask.

Here are the results:

D3D 11:
Ultra: 45.6
Medium: 172.6
Lowest: 355.2

Vulkan Linux:
Ultra: 41.6
Medium: 158.3
Lowest: 419.3

OpenGL Linux:
Ultra: 36.5
Medium: 101.0
Lowest: 263.2

So, we are getting much closer :) It's interesting when put on low settings that Vulkan completely smokes D3D11 by almost 70 fps. I'm wondering if this has to do with multi-threading vs the pseudo multi-threadedness of D3D11?

Can you do the Vulkan and the OpenGL benchmarks under windows too?
I want to see the difference between Vulkan-Windows vs Vulkan-Linux and OpenGL-Windows vs OpenGL-Linux.

I rarely look at performance numbers and benchmarks, as I am happy when game runs with enough FPS for me to be able to play it. I would however be curious if there are OTHER components holding back Linux performance other than OpenGL. In other words, if a game uses Vulkan on both Windows and Linux, would it perform exactly the same on both Linux and Windows? Or would Linux still be worse?

OK, disclaimer, I'm no developer and I can't back this up.
But the general zeitgeist I have from following vaguely this sort of thing for years comes down to: Aside from OpenGL problems, up to now graphics card drivers for Linux have tended to be slower, although that situation too has been improving and the nature of Vulkan seems to be such that it probably won't be nearly as much an issue for Vulkan stuff. But OS to OS, Linux itself is probably faster than Windows, and certainly lighter, so it wouldn't hog as much CPU or memory for itself, leaving more for the game. The "lighter" part of course depends on specifics like, KDE or XFCE or what? As I say, this is my general impression which I cannot cite any sources to defend, so take it for what it's worth.

All modern OS are comparable in performance ( i.e. no big advantages ). They have differing design philosophies which can have an impact, and as you point out the quality of drivers and other code libraries can also affect performance. My cross-platform Linux/Windows code tends to run a little faster in Linux, but as it is developed using Linux-based libraries ported through MinGW to Windows, that is probably to be expected.

It is honestly very difficult to properly compare APIs or compare OS performance without actually viewing the source to verify that what you are comparing is reasonable. For example, the croteam OpenGL implementation may just not be as "good" in terms of best practice and optimization as their Vulkan implementation. All good developers learn constantly, but it doesn't mean you have the time or inclination to go back and improve older stuff once you realize it sucks :-)

I'm more with Kimyrielle, in that I don't care how many FPS, as long as it's good for my personal perception, which typically means consistency and responsiveness over pure frame-rate.

Can you do the Vulkan and the OpenGL benchmarks under windows too?
I want to see the difference between Vulkan-Windows vs Vulkan-Linux and OpenGL-Windows vs OpenGL-Linux.

Yep, here you go. I put them all here again for easy reading :)

Vulkan (Linux):
Ultra: 41.6
Medium: 158.3
Lowest: 419.3

Vulkan (Windows):
Ultra: 42.6
Medium: 166.8
Lowest: 387.7

OpenGL (Linux):
Ultra: 36.5
Medium: 101.0
Lowest: 263.2

OpenGL (Windows):
Ultra: 36.3
Medium: 101.2
Lowest: 343.5

D3D 11 (Windows):
Ultra: 45.6
Medium: 172.6
Lowest: 355.2

So, the numbers are about the same for Vulkan between Ultra and Medium, with it doing fairly better on Linux for Lowest settings (which is neat to see). OpenGL is almost exactly the same for Ultra and Medium, but doing significantly better on Windows (which I personally find interesting). Also, all tests were done with 1920x1080 resolution.


Last edited by natewardawg on 5 Mar 2017 at 3:13 am UTC

Vulkan just won't run for me in Linux. I get a black screen, then it reloads into OpenGL.

I got the following numbers though:

Settings:
Resolution 3840x2160
Max3d rendering Mpix 2.4
Multi-sample AA Medium
Vsync- disabled
Triple buffering- disabled
CPU Ultra
GPU Ultra
GPU Memory Ultra
Level Caching Ultra
Max fps Unlimited

With both DX9 & 11 it maxed out at around 155fps.
GPU utilization was 100% for GPU1 & 92% for GPU2.
CPU was at 33% with 1 core maxing, 3 more at around 40% & the rest idle.

Linux Mint 18 Nvidia driver version 375 (tried 367 as well with no change).
OpenGL 55.3 fps
Vulkan :(


Windows 10 Nvidia driver 378.66

OpenGL 65.4 fps

DX9 94.5 fps
DX9 SLI 156.2 fps

DX11 94.0 fps
DX11 SLI 154.6 fps

Vulkan 88.3 fps (did not use both GPUs, not sure if I need to enable multigpu somehow, or if it is not implemented yet).

https://www.youtube.com/watch?v=rLjHXyh1pps

Can you do the Vulkan and the OpenGL benchmarks under windows too?
I want to see the difference between Vulkan-Windows vs Vulkan-Linux and OpenGL-Windows vs OpenGL-Linux.

Yep, here you go. I put them all here again for easy reading :)

Vulkan (Linux):
Ultra: 41.6
Medium: 158.3
Lowest: 419.3

Vulkan (Windows):
Ultra: 42.6
Medium: 166.8
Lowest: 387.7

OpenGL (Linux):
Ultra: 36.5
Medium: 101.0
Lowest: 263.2

OpenGL (Windows):
Ultra: 36.3
Medium: 101.2
Lowest: 343.5

D3D 11 (Windows):
Ultra: 45.6
Medium: 172.6
Lowest: 355.2

So, the numbers are about the same for Vulkan between Ultra and Medium, with it doing fairly better on Linux for Lowest settings (which is neat to see). OpenGL is almost exactly the same for Ultra and Medium, but doing significantly better on Windows (which I personally find interesting). Also, all tests were done with 1920x1080 resolution.

Thank you for the info. :)

Sadly, this is the exception to the rule, because Croteam are the exception to the rule..
They are true developers..
You can see the passion in their work and the evidence of that are the continuous upgrades and improvements in a game that is more than two years old...
The Linux version of The Talos Principle is not a port of a Windows game; is an in-house developed Linux game and that should be the standard of Linux games...

But, I don't see porters like Feral, VP or Aspyr following the path of Croteam in matter of upgrades..
Games like Tomb Raider, Life is Strange, Mad Max, Deus Ex Mankind Divided will benefit of an upgrade to Vulkan, but that won't happen because is Feral...
Maybe the future linux port of Rise of the Tomb Raider will use Vulkan from scratch..Who knows.

Games like both Metro Redux don't need Vulkan upgrades, because they are in-house Linux games with a performance that is close to the windows versions. The only update they need is to add a resolution control in the graphics options and Physx by GPU, but I don't think it will happen because those games are abandonware.

I purchased the game twice; one for my windows account and another for my Linux account.
I give Croteam 9 of 10

@Mora, you're running the 64bit Talos right? It only works on that as far as I know.

The Linux version of The Talos Principle is not a port of a Windows game; is an in-house developed Linux game and that should be the standard of Linux games...

Croteam is a cool developer, but this is sadly not true. They've admitted that the graphics engine is designed around D3D, and other backends suffer as a result. They're working on a redesign and their new VR releases are likely to benefit from this work, but Talos is very much a Windows game they ported (in-house) to Linux.

Sadly, this is the exception to the rule, because Croteam are the exception to the rule..
They are true developers..
You can see the passion in their work and the evidence of that are the continuous upgrades and improvements in a game that is more than two years old...
The Linux version of The Talos Principle is not a port of a Windows game; is an in-house developed Linux game and that should be the standard of Linux games...

But, I don't see porters like Feral, VP or Aspyr following the path of Croteam in matter of upgrades..

Unfortunately, it doesn't really makes sense *right now* that porters would even use Vulkan due to lack of experience. Feral did say they're going to start experimenting with Vulkan early this year, which means they're probably already doing it. My guess is with Deus Ex:MD first. It makes a lot of sense that Croteam is pushing hard with Vulkan in TP since, correct me if I'm wrong, Serious Sam 4 is using the same engine as Talos Principle. Feral porting the 2013 Tomb Raider game to Vulkan doesn't trickle down to any other game. In other words Croteam adding Vulkan support to TP directly affects other games, therefore they didn't really loose time or money because they have a fully finished title to test the Vulkan support with. Feral adding Vulkan to Tomb Raider doesn't affect anything else and therefore is a big time/money sink.

What I'm really trying to say is this... I don't think Feral, VP, Aspyr, etc not porting their games to Vulkan has anything to do with a lack of desire to do so, I think they really would love to, but it just doesn't make financial sense to do so. It makes more sense for them to focus on getting us other great games which is good for them and us. Croteam does have a financial incentive to port games like TP to Vulkan since it affects other future titles, which is good for both them and us.

While I'm with 100% you that I would love to see these games ported to Vulkan, I think these companies are making the right decision by not porting some of these older titles to Vulkan. Probably more to come from them, with Vulkan, in the future :)


Last edited by natewardawg on 6 Mar 2017 at 12:43 pm UTC

@Mora, you're running the 64bit Talos right? It only works on that as far as I know.

Yep, I sure am. It could have something to do with running 4k or SLI maybe. I've tried it on previous versions of Talos as well & have never had Vulkan work.

Still can't get it to run without crashing with Vukan on my home-built SteamOS brewmaster box.. :( Tried from the menus and with launch option.. boo

Are you running an AMD gpu? [...]

Nope, Nvidia 970..


Last edited by inlinuxdude on 13 Mar 2017 at 2:53 pm UTC