Early OpenGL vs. Vulkan Linux Benchmarks With Talos Principle

[pal666]

so, we have slow, unreleased and immature drivers and only one of them is open source atm

[rabcor]

That's really bizarre that Vulkan performance is faster than GL on windows but slower on Linux. The NVidia drivers are generally performing about the same on both.

It sounds to me like the NVidia Vulkan performance on Linux just isn't as good right now, for whatever reason. Unless someone can show that the GL performance is better under linux on this game, which seems doubtful.

Check this: http://arstechnica.com/gaming/2016/0...rk-to-be-done/

Same resolution, not sure if same settings, and of course slightly different hardware, but GPU is the most important here. And our GTX 980 (85fps) on linux is dominating over that site's GTX 980-Ti (63fps) which is a better card. Vulkan on our GTX 980(72fps) is also outperforming that GTX 980-Ti(68fps) on windows, but by a narrower margin.

So in fact the assumptions about vulkan performing worse on linux on Nvidia's driver seem pretty false to me when an inferior card is beating a superior one like that. What's causing this confusion is probably the fact that vulkan was actually faster than opengl on the windows benchmark, but you failed to check the actual numbers; opengl was just performing much better on linux to begin with, and vulkan was also performing better it seems.

We should actually be pretty happy about this. Unless our tests are not being run at ultra quality. If they're running on ultra with 4xAA then... it's probably the same benchmark with a slightly different CPU.

[blackout23]

Check this: http://arstechnica.com/gaming/2016/0...rk-to-be-done/
but you failed to check the actual numbers; opengl was just performing much better on linux to begin with,

Now you are comparing tow different system with maybe different settings and different hardware to conclude that the OpenGL performance is just better on Linux.
Talos Principle exposes an ass ton of graphics settings and for example the maximum framebuffer size for internal rendering doesn't automatically go up to QHD if you change to 2560x1440. Without doing the test with exactly the same settings on the same hardware you can't conclude anything.

[duby229]

Well, it what it is now... AMD doesn't have support ready to set the record straight yet so...... Disappointing as hell.

EDIT: I just mean that if AMD had been prepared, they could have set the precedent instead of letting nVidia set this screwed up one.

[boffo]

Well, it what it is now... AMD doesn't have support ready to set the record straight yet so...... Disappointing as hell.

EDIT: I just mean that if AMD had been prepared, they could have set the precedent instead of letting nVidia set this screwed up one.

All of the Vulkan drivers are in alpha/beta, there is no record.

[zanny]

so, we have slow, unreleased and immature drivers and only one of them is open source atm

This is probably not the drivers fault. Vulkan is not OpenGL - and Phoronix has a bad habit of blaming drivers on every FPS discrepancy as it is, but with Vulkan much more of the work and blame need to fall with the developer implementing a renderer on top of Vulkan.

I'm sure Intel can optimize their shader compiler / routine handling of the Vulkan API over time, but we will not see 10% performance every quarter perpetually because Vulkan does much less of the work than an OpenGL driver would.

[rabcor]

Now you are comparing tow different system with maybe different settings and different hardware to conclude that the OpenGL performance is just better on Linux.
Talos Principle exposes an ass ton of graphics settings and for example the maximum framebuffer size for internal rendering doesn't automatically go up to QHD if you change to 2560x1440. Without doing the test with exactly the same settings on the same hardware you can't conclude anything.

I know this, but at least it gives us the option of giving these benchmarks benefit of the doubt. Let's hope what I said mostly stands.

[johnc]

They would have been better off keeping this benchmark under wraps. First impressions are important and now this thing is being trumpeted everywhere as slower than older-than-dirt DX11.

Notice how MS was smart and kept DX12 benchmarks contained until they could actually show advantages over DX11.

[Kano]

Michael

How is it possible that you did NOT use Nvidia 355.00.27 for the Vulkan tests?

https://openbenchmarking.org/result/1602191-GA-TALOSP12801

[Michael]

355 was used. Ignore that table hence why it wasn't shown since I manually had to modify the test result until adding Vulkan option to the Talos test profile etc.

Guest

How is it possible that you did NOT use Nvidia 355.00.27 for the Vulkan tests?

https://openbenchmarking.org/result/...GA-TALOSP12801