Valve just recently published a new Steam beta, and they have fixed up the Steam Overlay rendering issue that was causing it to bring down Vulkan performance.
I've run some fresh tests as requested on my Nvidia 980ti to show you the difference now with the Steam Overlay on and being used (FPS counter) vs before.
You can click the spoiler tags to expand the full output if that tickles your fancy.
Just Vulkan
All together
You can see Valve has made the Steam Overlay not impact performance really at all, so that's fantastic news. This is great, as when more games begin to use Vulkan there shouldn't be issues like this (hopefully).
The OpenGL performance has stayed the same as there wasn't really an issue with the Steam Overlay and OpenGL.
4K
Vulkan Average: 112.5 FPS (before it was 75.1 FPS)
OpenGL Average: 91.8 FPS
1080p
Vulkan Average: 136.4 FPS (before it was 82.7 FPS)
OpenGL Average: 102.7 FPS
When there are other games available that use Vulkan, I will most likely take a look at benchmarking them.
I've run some fresh tests as requested on my Nvidia 980ti to show you the difference now with the Steam Overlay on and being used (FPS counter) vs before.
You can click the spoiler tags to expand the full output if that tickles your fancy.
Just Vulkan
All together
You can see Valve has made the Steam Overlay not impact performance really at all, so that's fantastic news. This is great, as when more games begin to use Vulkan there shouldn't be issues like this (hopefully).
The OpenGL performance has stayed the same as there wasn't really an issue with the Steam Overlay and OpenGL.
4K
Vulkan Average: 112.5 FPS (before it was 75.1 FPS)
Spoiler, click me
20:56:53 INF: Gfx API: Vulkan
20:56:53 INF: Duration: 60.0 seconds (6747 frames)
20:56:53 INF: Average: 112.5 FPS (113.8 w/o extremes)
20:56:53 INF: Extremes: 267.8 max, 20.9 min
20:56:53 INF: Sections: AI=6%, physics=2%, sound=1%, scene=70%, shadows=13%, misc=7%
20:56:53 INF: Highs: 612 in 4.3 seconds (141.5 FPS)
20:56:53 INF: Lows: 743 in 8.3 seconds (89.8 FPS)
20:56:53 INF: > 60 FPS: 100%
20:56:53 INF: Duration: 60.0 seconds (6747 frames)
20:56:53 INF: Average: 112.5 FPS (113.8 w/o extremes)
20:56:53 INF: Extremes: 267.8 max, 20.9 min
20:56:53 INF: Sections: AI=6%, physics=2%, sound=1%, scene=70%, shadows=13%, misc=7%
20:56:53 INF: Highs: 612 in 4.3 seconds (141.5 FPS)
20:56:53 INF: Lows: 743 in 8.3 seconds (89.8 FPS)
20:56:53 INF: > 60 FPS: 100%
OpenGL Average: 91.8 FPS
Spoiler, click me
20:28:12 INF: Gfx API: OpenGL
20:28:12 INF: Duration: 59.3 seconds (5444 frames)
20:28:12 INF: Average: 91.8 FPS (100.7 w/o extremes)
20:28:12 INF: Extremes: 226.5 max, 4.5 min
20:28:12 INF: Sections: AI=5%, physics=2%, sound=1%, scene=70%, shadows=16%, misc=6%
20:28:12 INF: Lows: 352 in 8.8 seconds (40.2 FPS)
20:28:12 INF: 30-60 FPS: 7%
20:28:12 INF: > 60 FPS: 92%
20:28:12 INF: Duration: 59.3 seconds (5444 frames)
20:28:12 INF: Average: 91.8 FPS (100.7 w/o extremes)
20:28:12 INF: Extremes: 226.5 max, 4.5 min
20:28:12 INF: Sections: AI=5%, physics=2%, sound=1%, scene=70%, shadows=16%, misc=6%
20:28:12 INF: Lows: 352 in 8.8 seconds (40.2 FPS)
20:28:12 INF: 30-60 FPS: 7%
20:28:12 INF: > 60 FPS: 92%
1080p
Vulkan Average: 136.4 FPS (before it was 82.7 FPS)
Spoiler, click me
20:43:11 INF: Gfx API: Vulkan
20:43:11 INF: Duration: 60.0 seconds (8186 frames)
20:43:11 INF: Average: 136.4 FPS (138.9 w/o extremes)
20:43:11 INF: Extremes: 261.2 max, 29.5 min
20:43:11 INF: Sections: AI=7%, physics=2%, sound=1%, scene=66%, shadows=15%, misc=9%
20:43:11 INF: Highs: 942 in 5.6 seconds (167.2 FPS)
20:43:11 INF: Lows: 1216 in 11.0 seconds (111.0 FPS)
20:43:11 INF: > 60 FPS: 100%
20:43:11 INF: Duration: 60.0 seconds (8186 frames)
20:43:11 INF: Average: 136.4 FPS (138.9 w/o extremes)
20:43:11 INF: Extremes: 261.2 max, 29.5 min
20:43:11 INF: Sections: AI=7%, physics=2%, sound=1%, scene=66%, shadows=15%, misc=9%
20:43:11 INF: Highs: 942 in 5.6 seconds (167.2 FPS)
20:43:11 INF: Lows: 1216 in 11.0 seconds (111.0 FPS)
20:43:11 INF: > 60 FPS: 100%
OpenGL Average: 102.7 FPS
Spoiler, click me
20:32:07 INF: Gfx API: OpenGL
20:32:07 INF: Duration: 60.0 seconds (6163 frames)
20:32:07 INF: Average: 102.7 FPS (111.7 w/o extremes)
20:32:07 INF: Extremes: 229.7 max, 19.1 min
20:32:07 INF: Sections: AI=6%, physics=2%, sound=1%, scene=67%, shadows=18%, misc=6%
20:32:07 INF: Highs: 530 in 3.2 seconds (166.2 FPS)
20:32:07 INF: Lows: 936 in 14.8 seconds (63.4 FPS)
20:32:07 INF: 30-60 FPS: 5%
20:32:07 INF: > 60 FPS: 95%
20:32:07 INF: Duration: 60.0 seconds (6163 frames)
20:32:07 INF: Average: 102.7 FPS (111.7 w/o extremes)
20:32:07 INF: Extremes: 229.7 max, 19.1 min
20:32:07 INF: Sections: AI=6%, physics=2%, sound=1%, scene=67%, shadows=18%, misc=6%
20:32:07 INF: Highs: 530 in 3.2 seconds (166.2 FPS)
20:32:07 INF: Lows: 936 in 14.8 seconds (63.4 FPS)
20:32:07 INF: 30-60 FPS: 5%
20:32:07 INF: > 60 FPS: 95%
When there are other games available that use Vulkan, I will most likely take a look at benchmarking them.
Some you may have missed, popular articles from the last month:
All posts need to follow our rules. For users logged in: please hit the Report Flag icon on any post that breaks the rules or contains illegal / harmful content. Guest readers can email us for any issues.
Serious improvement then !
2 Likes, Who?
Wow Vulkan is just murdering OpenGL
3 Likes, Who?
Added in a chart for them all together. Vulkan is clearly leading the way now for my testing, which is great news, and this isn't even final.
3 Likes, Who?
Wow Vulkan is just murdering OpenGL
While this is all nice for one particular game, and I'm sure there will be performance benefits across the board anyway, it's a bit early for such a generalised statement. Especially when the OpenGL version is of such importance. At least on my rig, Talos is using GL3.0, GLSL 130. Modern OpenGL (i.e GL4.5) would allow much greater performance improvements.
Again though, that they have something nicer with Vulkan is good. Just don't expect it everywhere.
While true, it depends on what they are doing. Using a higher OpenGL version wouldn't always result in higher performance.
1 Likes, Who?
hmmm... Arch Linux, Kernel 4.5.0, Nvidia 364.16, GTX850M ...
Last edited by z3ntu on 15 April 2016 at 8:47 pm UTC
22:43:59 INF: - benchmark results -
22:43:59 INF:
22:43:59 INF: Gfx API: Vulkan
22:43:59 INF: Duration: 60.0 seconds (3117 frames)
22:43:59 INF: Average: 52.0 FPS (52.8 w/o extremes)
22:43:59 INF: Extremes: 266.2 max, 15.0 min
22:43:59 INF: Sections: AI=4%, physics=1%, sound=1%, scene=83%, shadows=8%, misc=3%
22:43:59 INF: Highs: 110 in 1.4 seconds (80.3 FPS)
22:43:59 INF: Lows: 260 in 6.6 seconds (39.3 FPS)
22:43:59 INF: 30-60 FPS: 89%
22:43:59 INF: > 60 FPS: 10%
22:41:41 INF: - benchmark results -
22:41:41 INF:
22:41:41 INF: Gfx API: OpenGL
22:41:41 INF: Duration: 60.0 seconds (11336 frames)
22:41:41 INF: Average: 189.0 FPS (204.4 w/o extremes)
22:41:41 INF: Extremes: 354.0 max, 12.7 min
22:41:41 INF: Sections: AI=14%, physics=3%, sound=2%, scene=72%, shadows=0%, misc=9%
22:41:41 INF: Highs: 51 in 0.1 seconds (344.3 FPS)
22:41:41 INF: Lows: 856 in 8.8 seconds (97.1 FPS)
22:41:41 INF: > 60 FPS: 100%
Last edited by z3ntu on 15 April 2016 at 8:47 pm UTC
0 Likes
Liam, thanks for your work. Amazing benchmark, also the graphs are looking more professional and easy to read. Keep up!
Now, it would be nice to have a post on how to install this new driver. Is there a ppa? Should it be used together with bumblebee? This kind of stuff.
Best,
Now, it would be nice to have a post on how to install this new driver. Is there a ppa? Should it be used together with bumblebee? This kind of stuff.
Best,
2 Likes, Who?
If we could keep crap out of the punch bowl that would be great.
Grats to Valve on the very needed fix.
Last edited by ElectricPrism on 16 April 2016 at 4:47 am UTC
Grats to Valve on the very needed fix.
Last edited by ElectricPrism on 16 April 2016 at 4:47 am UTC
0 Likes
Where does the bottleneck come from, though? There is about 20% performance difference between 1080p and 4K which makes zero sense.
0 Likes
If by changing the resolution there's not much difference, then it should be a CPU botteneck.
Well, vulkan is supposed to be way less CPU dependent, as much as an i3 would give more or less the same results as an i7 on CPU demanding games.
But still Talos isnt really CPU demanding and show that 'lack' of differences between 1k and 4k, I don't have any idea and i'm wondering too.
Last edited by Armand Raynal on 16 April 2016 at 10:30 am UTC
Well, vulkan is supposed to be way less CPU dependent, as much as an i3 would give more or less the same results as an i7 on CPU demanding games.
But still Talos isnt really CPU demanding and show that 'lack' of differences between 1k and 4k, I don't have any idea and i'm wondering too.
Last edited by Armand Raynal on 16 April 2016 at 10:30 am UTC
0 Likes
Resolution is only one part of the performance equation unless you're doing extremely simplistic rendering. Both tests shuffle the same amount of textures and geometry and much of the processing and memory bandwidth requirements are the same, unless Liam changed other settings as well.
0 Likes
All settings were the same of course. 4K always uses more resources in all games I've tested. Some games are unplayable at 4K.
0 Likes
Serious improvement then !
I see what you did there ;)
1 Likes, Who?
Where does the bottleneck come from, though? There is about 20% performance difference between 1080p and 4K which makes zero sense.
The fact that it's 4k makes a lot of sense. It's that demanding.
0 Likes
Where does the bottleneck come from, though? There is about 20% performance difference between 1080p and 4K which makes zero sense.
It makes all the sense in the world? 4k is only EIGHT TIMES as many pixels to calculate and draw for each frame.
Last edited by Beamboom on 16 April 2016 at 4:34 pm UTC
0 Likes
Where does the bottleneck come from, though? There is about 20% performance difference between 1080p and 4K which makes zero sense.
It makes all the sense in the world? 4k is only EIGHT TIMES as many pixels to calculate and draw for each frame.
Firstly, a correction to your numbers. HD is 1920x1080. 4k is NOT 8xHD pixels - 4K resolutions are typically 3840x2160, despite the implication that 4K has 4000 vertical lines. 4k has 4xHD pixels.
Back in the early days of GPU cards, once you hit the pixel fill rate, that was basically the bottleneck. So 4x the pixels would be 1/4th the frame rate.
Modern cards are highly parallel, typically 100+ pipelines even at the midrange. Pixel rate is no longer the bottleneck, shader texel rates and data across the GPU bus are more critical. If the texture data needed by the GPU fits into the VRAM, along with the compiled pixel and vertex shaders, very high frame rates at high resolutions will run nearly as fast as the lower resolutions.
0 Likes
Wow Vulkan is just murdering OpenGL
While this is all nice for one particular game, and I'm sure there will be performance benefits across the board anyway, it's a bit early for such a generalised statement. Especially when the OpenGL version is of such importance. At least on my rig, Talos is using GL3.0, GLSL 130. Modern OpenGL (i.e GL4.5) would allow much greater performance improvements.
Again though, that they have something nicer with Vulkan is good. Just don't expect it everywhere.
nope, it wouldn't for 3 reasons
1. only NVidia implements it well. just try looking trough all the threads where developers tried AZDO. and then there is the problem where Apple is at old GL and to make it worse, their GL is terrible which is what forces developers to go with older one
2. AZDO < Vulkan since it only covers part of functionality in Vulkan
3. AZDO is even harder to grasp/write than Vulkan
face it, GL problems will never be fixed since most vendors focused on Vulkan
Last edited by vulture on 16 April 2016 at 10:07 pm UTC
0 Likes
Serious improvement then !
I see what you did there ;)
0 Likes
[quote=AskQuestionsLater]
Firstly, a correction to your numbers. HD is 1920x1080. 4k is NOT 8xHD pixels - 4K resolutions are typically 3840x2160, despite the implication that 4K has 4000 vertical lines. 4k has 4xHD pixels.
/quote]
( 3840 * 2160) / (1920 * 1080) = 4 - so it's 4 times the amount of pixels! ;)
Your statement about the impact of resolution is still not true though: Yes, there's no difference in between resolution like 720p and 1080p. But 1080p vs 4k does have a big impact on performance! If you check out modern benchmarks on e.g. computerbase you'll notice that there's allmost no video card (not even 980ti, Titan X, Fury X) that can handle modern (windows) games in 4k resolution with modest frame rates (=> 60 FPS). One reason is the amount of video memory (even cards with 8 GB of VRAM can ran out of it) as well as the lack of raw performance.
Last edited by cRaZy-bisCuiT on 17 April 2016 at 11:10 am UTC
Where does the bottleneck come from, though? There is about 20% performance difference between 1080p and 4K which makes zero sense.
It makes all the sense in the world? 4k is only EIGHT TIMES as many pixels to calculate and draw for each frame.
Firstly, a correction to your numbers. HD is 1920x1080. 4k is NOT 8xHD pixels - 4K resolutions are typically 3840x2160, despite the implication that 4K has 4000 vertical lines. 4k has 4xHD pixels.
/quote]
( 3840 * 2160) / (1920 * 1080) = 4 - so it's 4 times the amount of pixels! ;)
Your statement about the impact of resolution is still not true though: Yes, there's no difference in between resolution like 720p and 1080p. But 1080p vs 4k does have a big impact on performance! If you check out modern benchmarks on e.g. computerbase you'll notice that there's allmost no video card (not even 980ti, Titan X, Fury X) that can handle modern (windows) games in 4k resolution with modest frame rates (=> 60 FPS). One reason is the amount of video memory (even cards with 8 GB of VRAM can ran out of it) as well as the lack of raw performance.
Last edited by cRaZy-bisCuiT on 17 April 2016 at 11:10 am UTC
0 Likes
What can I say - good work Khronos group and everyone involved. Seems you have nailed this one!
0 Likes
This also shows that Valve bet on Vulkan has been fully justified. This will make Steam Machines actually matter.
0 Likes
See more from me