Vulkan vs. OpenGL Performance For Linux Games

[phoronix]

Phoronix: Vulkan vs. OpenGL Performance For Linux Games

It has been a while since last publishing some Linux GPU driver benchmarks focused explicitly on the OpenGL vs. Vulkan performance, but that changed today with a fresh look at the performance between these two Khronos graphics APIs when tested with AMD and NVIDIA hardware on the latest RadeonSI/RADV and NVIDIA Linux graphics drivers.

http://www.phoronix.com/vr.php?view=26577

[Bitey]

I suspect that the min frame rates may be the main benefactor of vulcan in this gpu limited scenario...

[artivision]

Yep It's well established that Vulkan has less CPU overhead than openGL. Agree on that point.

His claim was that Nvidia's driver had more overhead than RadV, which seems to be untrue.

On Linux Nvidia has the best Vulkan driver. Performs better than the AMD community driver RadV.

The proprietary Radeon Software 18.20 does perform faster than RadV but according to Michael it doesn't work for half the games.

Mad_Max_default = less overhead per frame on OGL
Dawn_of_war_iii = less overhead per frame on both apis
Total_war_saga = complete parity
Dota_2 = parity per frame

[Brisse]

@humbug: That is true, and sometimes talent can even work around patents and come up with different solutions which achieves a similiar result without infringing on any patents.

[humbug]

It's possible that some performance optimizations are encumbered by patents though, right? Like the S3TC patent which recently expired which meant the tech could be enabled by default in the FOSS driver stack.

It's possible. But the AMD openGL open source drivers surpassing the Nvidia blob happened because of enough talented people spending enough man hours over the years to improve it.

I suspect the same will be true for Vulkan as well. Whoever has more talent contributing more effort will be faster, regardless of FOSS or blob.

[Brisse]

I don't think there is anything inherently slower about open source software. So I do not think we should have different performance standards...

It's possible that some performance optimizations are encumbered by patents though, right? Like the S3TC patent which recently expired which meant the tech could be enabled by default in the FOSS driver stack.

[humbug]

The performance from Vulkan driver for Nvidia is hardly surprising considering the black box approach from the company.

I don't think there is anything inherently slower about open source software. So I do not think we should have different performance standards...

However, that does not mean it is a better method. Too much quirks to resolve especially tainted kernel.

I agree, that's why I used AMD graphics cards. They just fit in better and are well integrated with the Linux ecosystem. Things just work out of the box and less headaches..

[finalzone]

On Linux Nvidia has the best Vulkan driver. Performs better than the AMD community driver RadV.

The performance from Vulkan driver for Nvidia is hardly surprising considering the black box approach from the company. However, that does not mean it is a better method. Too much quirks to resolve especially tainted kernel.

What makes RadV remarkable is the open implementation that matches and sometimes surpass Nvidia proprietary driver on some games or productive software not to mention the better interaction with AMD team in addition of Valve and in some extent Intel. RadV is quite functional and just needs refinements notably handling all GCN cards possible.

[F.Ultra]

Curious how times have changed.. it seems like nowdays ~40FPS is not considered a playable experience anymore.. I remember the times when I was more than happy with my 25FPS gaming experience

And back when I and friends where totally happy playing Elite on C64 where it would get below 1fps at times

[humbug]

Yep It's well established that Vulkan has less CPU overhead than openGL. Agree on that point.

His claim was that Nvidia's driver had more overhead than RadV, which seems to be untrue.

On Linux Nvidia has the best Vulkan driver. Performs better than the AMD community driver RadV.

The proprietary Radeon Software 18.20 does perform faster than RadV but according to Michael it doesn't work for half the games.