Mesa Open-Source GPU Drivers Enjoyed Near-Record Growth In 2021, Valve Dev Top Contributor

[Quackdoc]

"extra overhead" can't become so good, so that "no extra overhead" can be removed. zink is for supporting hardware with shitty drivers like novideo, not for pointlessly removing "native gallium"

there isn't really any reason that zink cannot become low enough overhead that it wouldn't make a significant difference for the vast majority of cases. Vulkan is low level enough and a growing spec. of course there will always be edge cases don't get me wrong, I'm sure there will be a few cases where zink simple cannot cope. but I don't see it being so large of an issue. I don't think we would remove gallium, but I can see them being put on life support. at least until gallium itself becomes obsolete... maybe one day?

I will trade a few percent overhead for simplicity, conformance, and minimum system redondancy with pleasure.
Wrappers on top of Vulkan do not suffer as much as some claim. DXVK already proved it.
If a driver like Zink, still in its infancy can get 75-100 percent of the perf of the best native drivers, established for years and years, that means it's on the way to being a potential new primary driver.

I DXVK is a prime example of why we WONT abandon native drivers. first of all, DXVK is an incomplete directx and dxgi implementation. and is not faithful to spec, which is why it will never replace native windows libraries and only works when used in an apps folder on windows. I am assuming zink aims to be conformant so it's really not a good example. of course you can argue that it is simply because of dxgi being closed source, but I would bet money that at least some of DXVK's performance gain in some cases is due to it not implementing some bloat DXGI has. and even at that there are still a good chunk of examples where dx11 games preform better.

[MadeUpName]

Bring on MESA 22. 🥳

[pal666]

I will trade a few percent overhead for simplicity, conformance, and minimum system redondancy with pleasure.

extra overhead reduces simplicity, doesn't add any conformance and doesn't reduces redundancy because now you have yet another driver(nobody will drop better gallium drivers just because you have some crazy ideas)

Wrappers on top of Vulkan do not suffer as much as some claim. DXVK already proved it.

dxvk didn't prove anything. to prove your crazy claim you have to implement dxvk as gallium backend(which will require adding frontend for dx>9, dx9 is handled by gallium nine) and benchmark it vs native backend.

.If a driver like Zink, still in its infancy can get 75-100 percent of the perf of the best native drivers, established for years and years, that means it's on the way to being a potential new primary driver.

you are crazy. zink's dev is topmost contributor to mesa for last year, it's not that much of infancy for wrapper driver. it can't get 100 percent and all its perf is derived from "best native drivers established for years and years", zink uses existing "established for years and years" opengl frontend and existing "established for years and years" vulkan drivers. but adds some overhead on top. there's no way to become primary driver by doing a lot of unnecessary work. zink can become primary driver for hardware with shitty opengl drivers like novideo, but it can't improve gallium drivers because they are exactly same as zink but with less extra work

[pal666]

there isn't really any reason that zink cannot become low enough overhead that it wouldn't make a significant difference for the vast majority of cases. Vulkan is low level enough and a growing spec. of course there will always be edge cases don't get me wrong, I'm sure there will be a few cases where zink simple cannot cope. but I don't see it being so large of an issue. I don't think we would remove gallium, but I can see them being put on life support. at least until gallium itself becomes obsolete... maybe one day?

there isn't really any reason for your fantasies to make any sense when you are proposing to remove gallium without realizing that zink is gallium driver and you'll have to remove zink too

[Venemo]

I will trade a few percent overhead for simplicity, conformance, and minimum system redondancy with pleasure.
Wrappers on top of Vulkan do not suffer as much as some claim. DXVK already proved it.
If a driver like Zink, still in its infancy can get 75-100 percent of the perf of the best native drivers, established for years and years, that means it's on the way to being a potential new primary driver.

The main issue now (with both Zink and DXVK) is not raw performance but the stuttering. Some work is being done to make it better, but currently not fully solvable.

[Quackdoc]

there isn't really any reason for your fantasies to make any sense when you are proposing to remove gallium without realizing that zink is gallium driver and you'll have to remove zink too

Did you even read what I said? I realize I didn't make my self clear, when i was specifically talking about gallium drivers, though I did allude to it when I said "them" but even then I said we wouldn't remove gallium. Ill make myself abundantly clear, only put the gallium drivers like radeonsi, for GPUs that have a conformant vulkan implementation capable of running zink, on bugfix only mode.

I don't think we would remove gallium, but I can see them being put on life support

[Rabiator]

Those who are calling for removing Gallium, don't forget that there are a lot of older GPUs that don't support Vulkan in hardware, but are still useful and widespread. How do you suggest supporting those in the future?
For instance the AMD TeraScale GPUs, marketing name HD 6000 (and earlier), plus some later low end models that are still using the TeraScale architecture.
For Intel, Gen7 is the first generation that supports Vulkan in hardware. Anything Sandy Bridge and older would be left in the cold.
And personally I don't care too much about Nvidia, but dropping their older GPUs would alienate too many people as well.

[Quackdoc]

Those who are calling for removing Gallium, don't forget that there are a lot of older GPUs that don't support Vulkan in hardware, but are still useful and widespread. How do you suggest supporting those in the future?
For instance the AMD TeraScale GPUs, marketing name HD 6000 (and earlier), plus some later low end models that are still using the TeraScale architecture.
For Intel, Gen7 is the first generation that supports Vulkan in hardware. Anything Sandy Bridge and older would be left in the cold.
And personally I don't care too much about Nvidia, but dropping their older GPUs would alienate too many people as well.

no one wants to remove gallium, what we want is to replace radeonsi, iris etc. on supported gpus with zink, once preformance hits a certain threshold, and then put the drivers into a "maintenance mode"

[rmfx]

Time will tell…
I bet Vulkan is low level enough to translate GL calls with near native perf and extensions regularly are added to help with that. And especially since VK drivers are more straightforward, they are less likely to hide bottlenecks. I don’t see why vk would waste that much perf if APIs are translated in a good way. In the end they do the same thing.

Zink imho can squish the remaining bottlenecks next year enough so that Zink prove its full value.

[smitty3268]

no one wants to remove gallium, what we want is to replace radeonsi, iris etc. on supported gpus with zink, once preformance hits a certain threshold, and then put the drivers into a "maintenance mode"

Those GL drivers are already essentially in maintenance mode. They've basically finished GL 4.6 support, and there isn't much more being added now.

It's not clear what switching over to zink would give you, if you aren't actually deleting those drivers. And then you're asking people to be ok with regressions to their driver when they update Mesa to a newer version, which seems unlikely to go very well.