Zink OpenGL-On-Vulkan Benchmarks Are Very Competitive To Radeon OpenGL Driver

[Anux]

AFAIK RadeonSi developers have a certain advantage when it comes to access to early hardware documentation.
So I think in order to phase out RadeonSi, AMD must first start officially supporting Radv and ACO.

Whatever there is can be "stolen" from AMDs official vulkan driver I guess.

To replace RadSI with Zink it must develop faster than RadeonSI. If vulkan + zink get more optimizations faster than RadSI there will be a point were RadeonSI falls out of favor. That might take 10 or more years.

But if you look at Intels new cards that come without DX 9-11 hardware support and probably without OGL support it might be useful right about now.

[user1]

Whatever there is can be "stolen" from AMDs official vulkan driver I guess.

I heard Radv reuses RadeonSi's winsys (the lowest component of user mode driver that communicates with the kernel driver). So I don't think they'll be able to rely solely on AMD's official vulkan driver.

But if you look at Intels new cards that come without DX 9-11 hardware support and probably without OGL support it might be useful right about now.

If you mean their Windows drivers, then only DX9 is run on top of DX12. Other drivers are native.

[Linuxxx]

When looking at the CPU usage over the span of all the OpenGL benchmarks carried out, to much surprise the Zink OpenGL-on-Vulkan run actually led to lower CPU usage than with the RadeonSI Gallium3D driver.

Exactly this!

I wonder why nobody else seems to regard this as the most important metric, especially if it means that Valve can improve Steam Deck's battery life even further by switching over to Zink for OpenGL games in the (near) future...

Michael

Still hoping that You could one day have a look at running the Steam Deck with AMD_PSTATE + performance governor and compare to the default acpi-cpufreq + schedutil, including power metrics.

One day...

[Eirikr1848]

Exactly this!

I wonder why nobody else seems to regard this as the most important metric, especially if it means that Valve can improve Steam Deck's battery life even further by switching over to Zink for OpenGL games in the (near) future...

Michael

Still hoping that You could one day have a look at running the Steam Deck with AMD_PSTATE + performance governor and compare to the default acpi-cpufreq + schedutil, including power metrics.

One day...

I have a Steam Deck I would be willing to lend to phoronix for that test if one isn’t already in the asset closet.

[Anux]

Exactly this!

I wonder why nobody else seems to regard this as the most important metric

Because geomean is also lower, less fps less cpu usage. Nothing special about it.

[qarium]

I have a Steam Deck I would be willing to lend to phoronix for that test if one isn’t already in the asset closet.

he already has a steam deck. he just need motivation means money to perform the work.

[smitty3268]

Ahhh got it! At what point do you think radeonsi will no longer be used and instead will be Zink on RADV for all GCN1/GFX6 and newer cards?

The only people really working on radeonsi for quite a while now is AMD, and they've made it pretty clear they have zero interest in supporting RADV. Maybe that changes eventually, but i don't think it's going to happen anytime soon.

They've also spent the last year extensively optimizing radeonsi for various "professional" workloads and benchmarks, which are a completely different type of workload to what you see in a typical game. I suspect Zink probably hasn't been optimized as much for those types of workloads yet, and I doubt AMD is going to just throw away the last year of optimization work they've done in order to start from scratch.

That said, ultimately it's going to be the distros that decide this. If at some point Fedora or Ubuntu thinks that Zink is good enough for most desktop users, they'll just put that in their repositories and make people go the extra step to install radeonsi if they really want to. I don't see AMD taking that step first, just like they continue to support their own amdvlk driver rather than radv which everyone uses.

[Eirikr1848]

he already has a steam deck. he just need motivation means money to perform the work.

When criticism is warranted, it’s worth stating but honestly friend this is… unwarranted and quite rude.

Unless I’m misunderstanding.

[zboszor]

What about ACO for Zink?

This question is meaningless. ACO is Amd COmpiler.
RADV is talking directly to the AMD hardware. It doesn't use LLVM but it has to use a compiler to generate the GPU "machine code", hence ACO.

Zink is a translation layer and works for any Vulkan driver with enough features.
Using AMD GPU machine code (the code generated by ACO) for Intel's ANV would just be garbage.

[Eirikr1848]

This question is meaningless. ACO is Amd COmpiler.
RADV is talking directly to the AMD hardware. It doesn't use LLVM but it has to use a compiler to generate the GPU "machine code", hence ACO.

Zink is a translation layer and works for any Vulkan driver with enough features.
Using AMD GPU machine code (the code generated by ACO) for Intel's ANV would just be garbage.

Because I pretty much derp out, I thought Zink was specific to OpenGL on RADV.

(With that incorrect thought process / I then wondered if there was a specific shortcut for Zink and why it was not compiled via ACO… so like… RADV but with “Zink GL hooks”

(Also — too much time on my hands while home ill... but not enough time to get things right! Whoops)