I decided to take a look at this Vulkan nonsense with The Talos Principle, since I hadn't actually tried it yet myself. Here are some benchmarks done on my 980ti. I will look to do some on my 970 if people want to see them too.
Tested across two different driver versions, both at 4K and 1080p using the built-in benchmark tool. This is using the beta version of Talos, not the stable version.
From my observations Vulkan seems smoother, with little to no hitching/stutters, where as OpenGL did show some notable hitching/stutters for brief moments.
It seems that the Steam Overlay brings down the overall Vulkan performance if something is using it. If you have the FPS counter on, or you get notifications the Vulkan performance will go down. For best results, turn it off. Valve are aware of the Steam Overlay rendering issue.
Test done on Intel i7, Nvidia 980ti, 16GB RAM, standard hard drive for the game as my SSD is used for my OS. You can click each spoiler section for the full benchmark log from the game.
Game settings for all tests:
2x MSAA
CPU Speed: Ultra
GPU Speed: High
GPU Memory: Ultra
Level Caching: High
Tests with Steam Overlay turned off
364.16
4K
Vulkan Average: 114.4 FPS (115.9 w/o extremes)
OpenGL Average: 92.1 FPS (99.4 w/o extremes)
1080p
Vulkan Average: 134.1 FPS (136.1 w/o extremes)
OpenGL Average: 102.0 FPS (110.3 w/o extremes)
See Page 2 for tests with Steam Overlay turned on.
Article taken from GamingOnLinux.com.
Tested across two different driver versions, both at 4K and 1080p using the built-in benchmark tool. This is using the beta version of Talos, not the stable version.
From my observations Vulkan seems smoother, with little to no hitching/stutters, where as OpenGL did show some notable hitching/stutters for brief moments.
It seems that the Steam Overlay brings down the overall Vulkan performance if something is using it. If you have the FPS counter on, or you get notifications the Vulkan performance will go down. For best results, turn it off. Valve are aware of the Steam Overlay rendering issue.
Test done on Intel i7, Nvidia 980ti, 16GB RAM, standard hard drive for the game as my SSD is used for my OS. You can click each spoiler section for the full benchmark log from the game.
Game settings for all tests:
2x MSAA
CPU Speed: Ultra
GPU Speed: High
GPU Memory: Ultra
Level Caching: High
Tests with Steam Overlay turned off
364.16
4K
Vulkan Average: 114.4 FPS (115.9 w/o extremes)
Spoiler, click me
20:40:40 INF: Gfx API: Vulkan
20:40:40 INF: Duration: 60.0 seconds (6861 frames)
20:40:40 INF: Average: 114.4 FPS (115.9 w/o extremes)
20:40:40 INF: Extremes: 323.8 max, 34.5 min
20:40:40 INF: Sections: AI=6%, physics=3%, sound=1%, scene=68%, shadows=14%, misc=8%
20:40:40 INF: Highs: 540 in 3.6 seconds (148.6 FPS)
20:40:40 INF: Lows: 643 in 7.4 seconds (87.2 FPS)
20:40:40 INF: > 60 FPS: 100%
20:40:40 INF: Duration: 60.0 seconds (6861 frames)
20:40:40 INF: Average: 114.4 FPS (115.9 w/o extremes)
20:40:40 INF: Extremes: 323.8 max, 34.5 min
20:40:40 INF: Sections: AI=6%, physics=3%, sound=1%, scene=68%, shadows=14%, misc=8%
20:40:40 INF: Highs: 540 in 3.6 seconds (148.6 FPS)
20:40:40 INF: Lows: 643 in 7.4 seconds (87.2 FPS)
20:40:40 INF: > 60 FPS: 100%
OpenGL Average: 92.1 FPS (99.4 w/o extremes)
Spoiler, click me
20:45:11 INF: Gfx API: OpenGL
20:45:11 INF: Duration: 60.0 seconds (5527 frames)
20:45:11 INF: Average: 92.1 FPS (99.4 w/o extremes)
20:45:11 INF: Extremes: 237.9 max, 27.4 min
20:45:11 INF: Sections: AI=5%, physics=1%, sound=1%, scene=71%, shadows=15%, misc=6%
20:45:11 INF: Highs: 545 in 3.6 seconds (152.7 FPS)
20:45:11 INF: Lows: 859 in 14.9 seconds (57.5 FPS)
20:45:11 INF: 30-60 FPS: 8%
20:45:11 INF: > 60 FPS: 92%
20:45:11 INF: Duration: 60.0 seconds (5527 frames)
20:45:11 INF: Average: 92.1 FPS (99.4 w/o extremes)
20:45:11 INF: Extremes: 237.9 max, 27.4 min
20:45:11 INF: Sections: AI=5%, physics=1%, sound=1%, scene=71%, shadows=15%, misc=6%
20:45:11 INF: Highs: 545 in 3.6 seconds (152.7 FPS)
20:45:11 INF: Lows: 859 in 14.9 seconds (57.5 FPS)
20:45:11 INF: 30-60 FPS: 8%
20:45:11 INF: > 60 FPS: 92%
1080p
Vulkan Average: 134.1 FPS (136.1 w/o extremes)
Spoiler, click me
20:51:16 INF: Gfx API: Vulkan
20:51:16 INF: Duration: 60.0 seconds (8042 frames)
20:51:16 INF: Average: 134.1 FPS (136.1 w/o extremes)
20:51:16 INF: Extremes: 317.6 max, 18.3 min
20:51:16 INF: Sections: AI=7%, physics=2%, sound=1%, scene=67%, shadows=15%, misc=8%
20:51:16 INF: Highs: 789 in 4.8 seconds (164.9 FPS)
20:51:16 INF: Lows: 1050 in 9.6 seconds (109.2 FPS)
20:51:16 INF: > 60 FPS: 100%
20:51:16 INF: Duration: 60.0 seconds (8042 frames)
20:51:16 INF: Average: 134.1 FPS (136.1 w/o extremes)
20:51:16 INF: Extremes: 317.6 max, 18.3 min
20:51:16 INF: Sections: AI=7%, physics=2%, sound=1%, scene=67%, shadows=15%, misc=8%
20:51:16 INF: Highs: 789 in 4.8 seconds (164.9 FPS)
20:51:16 INF: Lows: 1050 in 9.6 seconds (109.2 FPS)
20:51:16 INF: > 60 FPS: 100%
OpenGL Average: 102.0 FPS (110.3 w/o extremes)
Spoiler, click me
20:55:04 INF: Gfx API: OpenGL
20:55:04 INF: Duration: 60.0 seconds (6118 frames)
20:55:04 INF: Average: 102.0 FPS (110.3 w/o extremes)
20:55:04 INF: Extremes: 280.9 max, 24.8 min
20:55:04 INF: Sections: AI=5%, physics=2%, sound=1%, scene=69%, shadows=17%, misc=6%
20:55:04 INF: Highs: 600 in 3.6 seconds (165.3 FPS)
20:55:04 INF: Lows: 942 in 14.9 seconds (63.4 FPS)
20:55:04 INF: 30-60 FPS: 5%
20:55:04 INF: > 60 FPS: 95%
20:55:04 INF: Duration: 60.0 seconds (6118 frames)
20:55:04 INF: Average: 102.0 FPS (110.3 w/o extremes)
20:55:04 INF: Extremes: 280.9 max, 24.8 min
20:55:04 INF: Sections: AI=5%, physics=2%, sound=1%, scene=69%, shadows=17%, misc=6%
20:55:04 INF: Highs: 600 in 3.6 seconds (165.3 FPS)
20:55:04 INF: Lows: 942 in 14.9 seconds (63.4 FPS)
20:55:04 INF: 30-60 FPS: 5%
20:55:04 INF: > 60 FPS: 95%
See Page 2 for tests with Steam Overlay turned on.
Article taken from GamingOnLinux.com.
0 Likes
Some you may have missed, popular articles from the last month:
The comments on this article are closed.
Quoting: Purple Library GuyQuoting: KimyrielleNice to see it outperforming OpenGL (was there any doubt that it would btw.?) But the much more important question is...will it outperform DX12? Because if it doesn't devs will go "Why would we add a Vulkan render path for these 1% that use Linux?" To succeed on the market, being the second fastest API won't be good enough when the fastest one already controls 90% of the market.
Definitely an important question. Seems like so far we're looking at rapid improvement, so good sign there. From what I've heard about how similar they are, though, I have this feeling they'll end up at rough parity.
I'm thinking if I were making the decision of DX12 vs Vulkan as someone wanting to sell a game (once Vulkan is properly off the ground), it would come down to "Do I want DX12 for Windows and Xbox, or Vulkan for Windows and (everything else)?" And probably my answer would be "I'll just use an engine that does both and get all of the above for almost free."
This to me is the answer. Whatever will reduce development time and give them easier platform options. Linux might be only 1% but android on the other hand...
0 Likes
Quoting: GuestThe steam overlay is pretty bad for performance, and is really a hack job more than anything else. Plus, it has to work across the board - so it's using an ancient version of OpenGL, not even running in the same context as the application. So it's no surprise that it impacts performance so much.Is the steam overlay even necessary anymore? I remember when I still played steam games on a windows partition it was useful because using alt+tab from a fullscreen game would freeze for quite a while before you'd get the desktop. I think this was the motivation behind steam developing the overlay in the first place.
But my experience on Linux has always been that you can rapidly switch back and forth to other applications or desktops no matter how demanding the game is, even when running the game in fullscreen at a lower resolution than your desktop. It's actually faster for me to pull up a steam page by switching desktops and opening it in firefox than it is for me to pull it up in the overlay.
Does windows still have hangups when alt+tabbing out of a game? It would be great if valve could just get rid of the overlay entirely, since I think a lot of the design decisions are tied into that (like having steam render its own windows on the desktop)
0 Likes
Mountain Man Apr 14, 2016
Quoting: KimyrielleBut the much more important question is...will it outperform DX12? Because if it doesn't devs will go "Why would we add a Vulkan render path for these 1% that use Linux?"Performance isn't the only advantage of Vulkan or even the primary advantage. It's the fact that it can work on pretty much every platform that is capable of running games.
0 Likes
so they fixed whatever was causing problems on cpu-bound scenarios? since on 4k it runs better on vulkan. It's good to see than even if the engine is still not optimized for vulkan, its already giving good results.
Last edited by edo on 14 April 2016 at 1:48 am UTC
Last edited by edo on 14 April 2016 at 1:48 am UTC
0 Likes
Kimyrielle Apr 14, 2016
Quoting: Mountain ManQuoting: KimyrielleBut the much more important question is...will it outperform DX12? Because if it doesn't devs will go "Why would we add a Vulkan render path for these 1% that use Linux?"Performance isn't the only advantage of Vulkan or even the primary advantage. It's the fact that it can work on pretty much every platform that is capable of running games.
You're preaching to the converted in my case (obviously), but you know how these Windows devs operate: "Wait, what? People play games on platforms other than Windows? Are you kidding me?" If you want to get that crowd to use Vulkan, you need to give them something other than "It's running flawlessly on these platforms you don't even care about".
0 Likes
Quoting: jo3fisQuoting: Purple Library GuyQuoting: KimyrielleNice to see it outperforming OpenGL (was there any doubt that it would btw.?) But the much more important question is...will it outperform DX12? Because if it doesn't devs will go "Why would we add a Vulkan render path for these 1% that use Linux?" To succeed on the market, being the second fastest API won't be good enough when the fastest one already controls 90% of the market.
Definitely an important question. Seems like so far we're looking at rapid improvement, so good sign there. From what I've heard about how similar they are, though, I have this feeling they'll end up at rough parity.
I'm thinking if I were making the decision of DX12 vs Vulkan as someone wanting to sell a game (once Vulkan is properly off the ground), it would come down to "Do I want DX12 for Windows and Xbox, or Vulkan for Windows and (everything else)?" And probably my answer would be "I'll just use an engine that does both and get all of the above for almost free."
This to me is the answer. Whatever will reduce development time and give them easier platform options. Linux might be only 1% but android on the other hand...
Well, and correct me if I'm wrong but I seem to recall hearing that PS4 will run Vulkan . . .
0 Likes
Quoting: GuestThe steam overlay is pretty bad for performance, and is really a hack job more than anything else. Plus, it has to work across the board - so it's using an ancient version of OpenGL, not even running in the same context as the application. So it's no surprise that it impacts performance so much.At this moment with OpenGL the game is equally using two CPUs, while with Vulkan everything is crammed into one.
I'm curious as to whether CPU usage evens out with Vulkan or not. It speaks a lot about where the improvements are really coming from.
0 Likes
It would be really useful to have the actual frame data of the benchmark, instead of a short summary (average, extremes etc.). It would be possible to calculate other important metrics (median, for example) and have a more in-depth comparison of vulkan vs opengl vs dx
EDIT: it looks like using voglperf (the 32 bit version) it is possible to capture all the frame timings... time to do some more benchmark!
Last edited by J_Salem on 14 April 2016 at 7:07 am UTC
EDIT: it looks like using voglperf (the 32 bit version) it is possible to capture all the frame timings... time to do some more benchmark!
Last edited by J_Salem on 14 April 2016 at 7:07 am UTC
0 Likes
Quoting: JSalemIt would be really useful to have the actual frame data of the benchmark, instead of a short summary (average, extremes etc.). It would be possible to calculate other important metrics (median, for example) and have a more in-depth comparison of vulkan vs opengl vs dxIt's a bit too early for that sort of scrutiny. Their Vulkan backend is very much a work in progress, and you wouldn't get useful data out of the analysis.
1 Likes, Who?
Quoting: tuubiQuoting: JSalemIt would be really useful to have the actual frame data of the benchmark, instead of a short summary (average, extremes etc.). It would be possible to calculate other important metrics (median, for example) and have a more in-depth comparison of vulkan vs opengl vs dxIt's a bit too early for that sort of scrutiny. Their Vulkan backend is very much a work in progress, and you wouldn't get useful data out of the analysis.
I think it would still be useful; ideally you should be able to see exactly where, in their automated demo, there are significant differences between the APIs. This would potentially be insightful for the developers themselves.
More data is always good (provided that the data is annotated clearly)
1 Likes, Who?
Patreon
PayPal
See more from me