Since Feral Interactive have fixed up the OpenGL renderer in the Mad Max [Steam] Beta, here are some fresh OpenGL vs Vulkan tests.
I already cleared up the issue before and included some manual testing, see here.
These new tests are re-done using their benchmark feature, which is unique to the Linux version. This should now give a much more accurate look at how OpenGL fares against Vulkan in some heavy areas of the game.
Also, Feral have now made public how they have worked around a Linux performance issue. You might remember my post about tuning your CPU performance governor for Vulkan games, well this is partly where that came from (as well as Serious Sam). Essentially, to prevent the CPU performance being reduced due to less CPU use with Vulkan, Feral are spinning their rendering thread while waiting for the GPU (see here).
Here we have a performance difference of between (approx) +6% to +15% when using Vulkan.
Here we have a performance difference of between (approx) -20% to +45% when using Vulkan. Looks like there's an odd issue with Vulkan performance there, but I'm not the only one who has seen Vulkan sometimes do a little worse.
Here we have a performance difference of between (approx) +3% to +11% when using Vulkan.
Here we have a performance difference of between (approx) +1% to +6% when using Vulkan.
Some thoughts: A fair amount of their time during the Beta so far has likely been spent fixing up the regressed OpenGL side and when taking into consideration Vulkan is considered "Beta", it's likely Feral has optimizations left to do with Vulkan.
You might not think much of a 5% increase, but for people on lower-end hardware a 5% increase can mean a world of difference. I've seen a few people on GOL and Reddit say Vulkan has made it go from sluggish to smooth, which is a clear win.
We also have to consider that the OpenGL renderer in the Mad Max beta has also seen some optimizations since the original release. The original release doesn't have the benchmark mode, so we can't test that in the same way along side the beta.
Then there's also the drivers, Vulkan drivers are new and evolving and likely have their own sets of issues to be fixed. Some of which could affect performance. Sure Vulkan is supposed to have smaller drivers, but with so few Vulkan games out no driver has truly been tested.
Be sure to share your latest results in the comments, I'm keen to see what it's like on your systems too.
Finally, Feral are keen for feedback, email them direct here: [email protected]
I already cleared up the issue before and included some manual testing, see here.
These new tests are re-done using their benchmark feature, which is unique to the Linux version. This should now give a much more accurate look at how OpenGL fares against Vulkan in some heavy areas of the game.
Also, Feral have now made public how they have worked around a Linux performance issue. You might remember my post about tuning your CPU performance governor for Vulkan games, well this is partly where that came from (as well as Serious Sam). Essentially, to prevent the CPU performance being reduced due to less CPU use with Vulkan, Feral are spinning their rendering thread while waiting for the GPU (see here).
Benchmarks
Here we have a performance difference of between (approx) +6% to +15% when using Vulkan.
Here we have a performance difference of between (approx) -20% to +45% when using Vulkan. Looks like there's an odd issue with Vulkan performance there, but I'm not the only one who has seen Vulkan sometimes do a little worse.
Here we have a performance difference of between (approx) +3% to +11% when using Vulkan.
Here we have a performance difference of between (approx) +1% to +6% when using Vulkan.
Some thoughts: A fair amount of their time during the Beta so far has likely been spent fixing up the regressed OpenGL side and when taking into consideration Vulkan is considered "Beta", it's likely Feral has optimizations left to do with Vulkan.
You might not think much of a 5% increase, but for people on lower-end hardware a 5% increase can mean a world of difference. I've seen a few people on GOL and Reddit say Vulkan has made it go from sluggish to smooth, which is a clear win.
We also have to consider that the OpenGL renderer in the Mad Max beta has also seen some optimizations since the original release. The original release doesn't have the benchmark mode, so we can't test that in the same way along side the beta.
Then there's also the drivers, Vulkan drivers are new and evolving and likely have their own sets of issues to be fixed. Some of which could affect performance. Sure Vulkan is supposed to have smaller drivers, but with so few Vulkan games out no driver has truly been tested.
Be sure to share your latest results in the comments, I'm keen to see what it's like on your systems too.
Finally, Feral are keen for feedback, email them direct here: [email protected]
Some you may have missed, popular articles from the last month:
Really good piece of work. I saw some guys suggesting that Feral had put the regression on purpose. Personally I think that some people are just too quick in making judgments and accusations, which I don't think it is fine at all.
4 Likes, Who?
Quoting: GuestEDIT: also please fix the graphs and keep the order consistent, my ocd is killing me :(Well, the order is correct for the intended purposes, showing which is the winner in each section. The colours are always the same.
0 Likes
In my i7 4790k + gtx 780 I did not have very significant difference. Only in some moments I had little framedrop with vulkan compared to opengl. I show in the video
https://www.youtube.com/watch?v=y5AvPP0o_c8
https://www.youtube.com/watch?v=y5AvPP0o_c8
0 Likes
Quoting: FastOSIn my i7 4790k + gtx 780 I did not have very significant difference. Only in some moments I had little framedrop with vulkan compared to opengl. I show in the videoWell for one thing, that will be compared the regressed OpenGL. That's the point of this article, the initial Beta had a regressed OpenGL renderer.
https://www.youtube.com/watch?v=y5AvPP0o_c8
0 Likes
I my experience, the improvement is much better in bad areas. In line with the Tyrant's Lash benchmark. So the improvement is really noticeable where it counts.
2 Likes, Who?
its great that they have made this improvment so the game is playable for computers that werent...
but deus ex is the game who need serious improvments, i was watching linux game cast and even with an nvidia980 they didnt get an good frame rate if my memory dont fail me...
this improvment is not enough to make the performance better than windows, so no one will install linux just to play an game that he/she cant play on windows due to weak hardware.
and feral could have spent this time porting other games, sure, it cost more, but they will not make many extra sales from improving the performance on this one, porting others they will...
but deus ex is the game who need serious improvments, i was watching linux game cast and even with an nvidia980 they didnt get an good frame rate if my memory dont fail me...
this improvment is not enough to make the performance better than windows, so no one will install linux just to play an game that he/she cant play on windows due to weak hardware.
and feral could have spent this time porting other games, sure, it cost more, but they will not make many extra sales from improving the performance on this one, porting others they will...
0 Likes
Quoting: elmapulthis improvment is not enough to make the performance better than windows, so no one will install linux just to play an game that he/she cant play on windows due to weak hardware.It's hard to hit Windows levels with games that were never designed with Linux or anything but DirectX in mind.
Quoting: elmapuland feral could have spent this time porting other games, sure, it cost more, but they will not make many extra sales from improving the performance on this one, porting others they will...Any work they do on their Vulkan rendering will get it ready for future ports I'm sure.
5 Likes, Who?
Quoting: EhvisI my experience, the improvement is much better in bad areas. In line with the Tyrant's Lash benchmark. So the improvement is really noticeable where it counts.
Yes this is what we'd expect the largest boosts come when previously the GPU was starved due to GL driver overhead using all the CPU time.
Vulkan has a much lower overhead which means in these high stress areas you can now utilise much more (usually 100%) of your GPU. The less powerful the CPU and the more powerful the GPU the greater the effect Vulkan can have.
This means you'll be a bit faster in most areas but the areas that really dipped down due to drivers should now only drop a small amount giving you a much nicer experience.
That said this is still a beta (and so are some of the drivers) so expect a few rough edges. That's why we released the beta and asked for feedback.
Thanks to everyone who's emailed in so far!
Last edited by edddeduck_feral on 4 April 2017 at 10:01 pm UTC
20 Likes, Who?
This is absolutely a noticeable improvement on my rig - i3-6100, GTX950.
The difference is I can max out the graphical settings, including using antialiasing, and get a smooth game at 30fps+.
With OGL I had to run without antialiasing, and with a few settings at either medium or high (vs everything at ultra).
The difference is I can max out the graphical settings, including using antialiasing, and get a smooth game at 30fps+.
With OGL I had to run without antialiasing, and with a few settings at either medium or high (vs everything at ultra).
1 Likes, Who?
Hm, the old OpenGL version is still faster than the new one, and in some places faster than Vulkan.
1 Likes, Who?
See more from me