How Open-Source Allowed Valve To Implement VULKAN Much Faster On The Source 2 Engine

[Nille_kungen]

If we are talking about proprietary drivers, which we are if we are talking about Mantle drivers, then OpenGL has already "caught up." If we are talking about Open drivers, then that is still going to be a little while, but that little while could be before the end of the year.

Day one support does matter to push adoption. One major limiting factor of adoption of any new platform has always been platform reach. "Why should I support Mantle if it is only on AMD?" is not an uncommon nor unreasonable question. With day one drivers, even if they are beta drivers, from AMD and this Intel based driver, not to mention the ARM and Imagination Technologies driver they have shown, Engines proponents such as Unity and Unreal have a real use case to support multiple platforms quickly. As does any other engine developer.

The equation now becomes, "Should I support DX12 which only gets me Windows 10 and Xbox One, or should I support Vulkan which gets me Windows Vista, 7, 8, 8.1 and 10, as well as Steam Machines, Linux, presumably BSDs, OS X and iOS, not to mention Android?" Of course, Epic and Unity will support both, but other smaller vendors as well as in house engines will support the one where customers are now.

As you say we will see the support in game engines and since they already are supported on different platforms without vulcan there isn't any big issues of platform support.
Vulcan will be good for game engines and new hardware but the game engines will still support non vulcan hardware, vulcan might give you more fps but it won't keep you from running them.
Quick support is good for adoption of vulcan.
I mostly talk about OSS support.

[dragorth]

Good comment but vulcan support on day one is irrelevant since it's not like vulcan is that important that thing won't work without it.
Day one support would be nice and it might be possible but it doesn't really matter, i rather see openGL catching up first.

If we are talking about proprietary drivers, which we are if we are talking about Mantle drivers, then OpenGL has already "caught up." If we are talking about Open drivers, then that is still going to be a little while, but that little while could be before the end of the year.

Day one support does matter to push adoption. One major limiting factor of adoption of any new platform has always been platform reach. "Why should I support Mantle if it is only on AMD?" is not an uncommon nor unreasonable question. With day one drivers, even if they are beta drivers, from AMD and this Intel based driver, not to mention the ARM and Imagination Technologies driver they have shown, Engines proponents such as Unity and Unreal have a real use case to support multiple platforms quickly. As does any other engine developer.

The equation now becomes, "Should I support DX12 which only gets me Windows 10 and Xbox One, or should I support Vulkan which gets me Windows Vista, 7, 8, 8.1 and 10, as well as Steam Machines, Linux, presumably BSDs, OS X and iOS, not to mention Android?" Of course, Epic and Unity will support both, but other smaller vendors as well as in house engines will support the one where customers are now.

[bridgman]

To be honest, I'm more interested to read news about AMDGPU kernel driver than other. Shouldn't it be ready to show itself in the LKML for the end of 2014 - first month 2015? What's the progress actually?
Thanks.

It was never "supposed to be ready to show itself" at any particular time -- I don't know exactly what Alex said at XDC but at best it would have been a casual estimate based on progress so far. Unfortunately the pattern with IP review is always the same -- the first 75% is easy, the next 20% is much harder, and the last 5% seems like it will never happen.

It's down to maybe the last 1% now, so things are looking promising.

Probably not NIR, it is a in-memory format and does not have a serialization (AFAIK), which is all that SPIR-V is. It will, for most purposes, be the format that SPIR-V is parsed to. TGSI could very well be replaced, given that it served a similar purpose anyway. LLVM, I can't say.

Right -- SPIR-V is intended as a transfer IR not an internal IR, so if it replaced anything it would be TGSI. Don't think there has been much discussion of that yet though.

[Nille_kungen]

To be clear, you don't know whether they have or not. The only thing you know for sure is that they haven't released any support for Mantle on Linux. It should also be noted that NO vendor has released Vulkan drivers, as Vulkan's spec has not been nailed down. So, docking AMD for that is rather heavy handed.

Now, AMD has working drivers for Mantle on Windows. This code should be able to transition relatively easily to become Vulkan. Putting AMD in the position of having the most tested driver for working games in the market. They promised Linux drivers when Mantle was released. So, internally they were working on it. I don't know why they didn't release it, and I won't speculate. They did release Mantle as an open spec, in the form of Vulkan, so they made good on that promise. There would be no Vulkan this fast without the work AMD did on Mantle.

We can speculate that they will have well implemented drivers for Vulkan on day one, at least in Beta, thanks to their leg up on their already existing work. We will see if that is the case when Vulkan is released.

Good comment but vulcan support on day one is irrelevant since it's not like vulcan is that important that thing won't work without it.
Day one support would be nice and it might be possible but it doesn't really matter, i rather see openGL catching up first.

[Nille_kungen]

To be honest, I'm more interested to read news about AMDGPU kernel driver than other. Shouldn't it be ready to show itself in the LKML for the end of 2014 - first month 2015? What's the progress actually?
Thanks.

Bridgman has posted it month+ ago:

Regarding amdgpu, it's probably obvious but it's going through the usual IP review/approval exercises so as usual you'll probably see it within 30 minutes of agd5f receiving final approval. It's a fairly big slug of IP (new HW generation etc..) so not a trivial exercise.

He also talked about 3.21 kernel = 4.1, he can have been ironic but i think it's actually that close but the IP review/approval might take some time.

Pretty sure I have provided tentative ETAs a couple of times already, but let me do it again.

We are trying to get the code out in time to make the 3.21 merge window if possible. Not sure if we will make it though.

I think we will see it as soon as the driver passes IP review/approval.

[nanonyme]

I wonder if vulkan and its new shader representation can be used for doing cross-GPU hardware decoding : I am still experiencing hangs when using hardware decoding on my 6870 with open source drivers (And I am not talking about UE4 titles).

I can't wait for some vulkan demos to run on my PC !

It doesn't make any sense to decode with shaders where dedicated decoder hardware is included in the GPU. Have you filed bugs on your hangs?

[M@yeulC]

I wonder if vulkan and its new shader representation can be used for doing cross-GPU hardware decoding : I am still experiencing hangs when using hardware decoding on my 6870 with open source drivers (And I am not talking about UE4 titles).

I can't wait for some vulkan demos to run on my PC !

[Michael]

What a letdown. Michael, your tease promised something more, this was Captain Obvious-tier news.

I didn't know what I was getting exactly.... Besides LunarG promising they'd get me an exclusive story by end of month. Yes, most of it was obvious, but resparks VULKAN discussions at least.

[nightmarex]

Seriously, given the way things worked out what good would have come from releasing a Linux Mantle driver ? None of the game devs would have ported their games to Linux on Mantle knowing that Vulkan was coming. Companies don't really have feelings to hurt, but if they did then calling the decision not to implement a Linux Mantle driver under the circumstances "bulllshit lazy" (or even "bullshit lazy") would be both unfair and insulting.

Remember that the goal of Mantle was to change the direction of development environments, not to push an AMD-centric graphics API standard.

Implementing one Mantle driver probably wouldn't have been that much work, but Mantle was a moving target right until the end and maintaining two evolving drivers would have meant a lot of wasted effort. By the time the API stabilized enough to be the easy development effort you imagine, Vulkan was on the way.

Just for once, could you (and I mean "you all", not just nightmarex and asdfblah) please stop assuming that every decision a company makes is driven by stupidity or ignorance ?

I did say that it would be pointless to release mantle at this junction. I think you're missing what I am saying, AMD is the only company lately with any vision it seems with forward thinking with it's end goals. I support that and purchase AMD because of their developing technologies. However I don't like being looked at as a afterthought or nuisance because I don't want to use these products on artificially monopolized OS that thinks it owns your software and hardware instead of the other way around. So even if I am being critical, as a customer I may dislike certain aspects but it's not like I am one of those Nvidia shills who never say a bad word about something just because I am afraid of the consequence.

[SXX⁣]

To be honest, I'm more interested to read news about AMDGPU kernel driver than other. Shouldn't it be ready to show itself in the LKML for the end of 2014 - first month 2015? What's the progress actually?
Thanks.

Bridgman has posted it month+ ago:

Regarding amdgpu, it's probably obvious but it's going through the usual IP review/approval exercises so as usual you'll probably see it within 30 minutes of agd5f receiving final approval. It's a fairly big slug of IP (new HW generation etc..) so not a trivial exercise.