After ~70% FPS Boost For Zink, The OpenGL-on-Vulkan Code Is ~50% The GL Native Speed

[phoronix]

Phoronix: After ~70% FPS Boost For Zink, The OpenGL-on-Vulkan Code Is ~50% The GL Native Speed

For those following the development of Zink as a software OpenGL driver built atop the Vulkan API, there are some new performance numbers to discuss this weekend...

http://www.phoronix.com/scan.php?page=news_item&px=Zink-Heaven-Iris

[tildearrow]

Is it true that DXVK has approached/exceeded native DirectX speeds?

[Snaipersky]

Is it true that DXVK has approached/exceeded native DirectX speeds?

It depends on a lot of things. "A Hat in Time" is the poster child for "DXVK is faster than native", with a hefty 50% performance boost, though most of the time it's +/-15% native, especially once shaders have been compiled and pipelines cached. Do note however, that DXVK also sacrifices certain features, such as I believe shared surfaces, important for things like GDI controlled playback of media in a directx application.

[R41N3R]

Great performance improvements! Yes, we need this to go upstream in Mesa 20.3 :-)

[eydee]

Is it true that DXVK has approached/exceeded native DirectX speeds?

Mostly in select few games, when running under Windows, on AMD cards, due to bad Directx drivers. Generally, no. When running games in Wine under Linux, the performance hit is still very noticeable, in all games, often exceeding 50%. Many of the core components of Wine were written 20+ years ago, for old Pentiums, and they lack support for both multicore architectures and modern CPU instructions. This is also the reason why native Vulkan games tend to get a severe performance hit, like DOOM (2016) does.

[curfew]

Emulation layer cannot be as fast as the native implementation that properly utilizes the underlying hardware. Previously some people said the Vulkan implementation allows for skipping some redundant error checks, but that is only one slight theoretical improvement, whilst there must be a ten-fold amount of drawbacks.

[DebianLinuxero]

I don't see the point on this.

I remember the 3dfx days, when voodoo hardware hasn't a native OpenGL implementation, but an OpenGL to Glide wrapper.

Something like this Zink.

That wrapper costed performance to the voodoos.

I remember a press note of one 3dfx manager saying that the next drivers they'll do will be native OpenGL because of this.

Too late I "Zink"

[nokipaike]

it would be interesting to see the percentage results on much more interesting hardware, with all the respect for intel ... but there is better in circulation ..

[George99]

I am not the honorary editor but: Blumenkratz? (scratching flowers)

[smitty3268]

it would be interesting to see the percentage results on much more interesting hardware, with all the respect for intel ... but there is better in circulation ..

Yes, I kind of suspect that choosing Intel for these tests is more likely to paint Zink in a positive light than RadeonSI, although I don't know that for sure.

Anyway, for those that will only read the headlines and not the actual blog post, it's important to note that parts of the demo were only running at 30%, rather than 50%.

Also, the tests were run with tessellation turned off. I actually think that probably makes the test more representative than if it was turned on, but it is important to know when comparing the results.