One can only hope.

[TimberWolf]

This could improve VR performance significantly, if the DCS engine could be re-written to incorporate DX12 Ultimate.

 

https://www.pcgamer.com/nvidias-rtx-3080-destroys-amds-rx-6800-xt-in-new-3dmark-test/

[speed-of-heat]

except we know that the work effort is going into Vulkan at present

[Eaglewings]

Interesting article. The mesh shader thing would be an absolute game changer if it becomes mainstream.
Hard to say what Vulkan is going to offer in dcs but looking at those figures in that article, performance boost with this mesh shader is insane.

Higher end graphic cards would come down their elevated position should this become implemented as you would not need those cards to enjoy gaming with all eye candies settings turned on.
Horizon looking good with this rendering innovation.

[TAW_Impalor]

Vulcan also includes mesh shaders from Turing cards and up: https://developer.nvidia.com/vulkan-driver

[Eaglewings]

Vulcan also includes mesh shaders from Turing cards and up: https://developer.nvidia.com/vulkan-driver

This is good to know.
I would hope dcs developers would take advantage of the this for their upcoming Vulkan integration into dcs world engine.

[mhe]

This would be a dream come true and tremendously help frames, especially in VR. And timing is just right since Vulkan was announced for later this year. Does the dev team have Mesh Shaders on their radar? @BIGNEWY

[BIGNEWY]

We have no plans for DX12 currently, but interesting stuff for sure. 

[FoxTwo]

42 minutes ago, BIGNEWY said:

We have no plans for DX12 currently, but interesting stuff for sure. 

 

Mesh shaders are not DX12 exclusive. Vulkan has support for them as well.

There are other APIs as well like variable rate shading that if properly leveraged could probably get DCS banging against the refresh rate of modern headsets without the need for motion interpolation. But I think it's unlikely for DCS to go from barely leveraging DX11 support to using bleeding edge vulkan apis. It's more likely we're going to get something that is mostly a direct port of the current engine to the new framework which has some built in efficiencies. At least at the beginning.

 

It would be nice if there was just a team that worked on the graphics engine exclusively and could roll out progressive updates to it from patch to patch. Maybe there will be, who knows.

[GunSlingerAUS]

On 2/13/2021 at 8:26 AM, TimberWolf said:

This could improve VR performance significantly, if the DCS engine could be re-written to incorporate DX12 Ultimate.

 

https://www.pcgamer.com/nvidias-rtx-3080-destroys-amds-rx-6800-xt-in-new-3dmark-test/

 

One major issue with mesh shaders (according to the Star Citizen devs) is that moving to this rendering pipeline would mean only the latest architectures from AMD and NVIDIA would be able to run it (AMD 6XXX and NVIDIA 2XXX/3XXX). Which means 80% or more of the DCS audience would no longer be able to play it. This reason alone means that it'll be at least five years - at the very least - before the vast majority of GPUs on the market support mesh shading. There are other limitations as well, but this single reason means it's sadly a non-starter for the time being. Still, if it does become the norm it could lead to some incredible performance improvements without a brute-force silicon approach.

[ShaunX]

5 hours ago, GunSlingerAUS said:

 

One major issue with mesh shaders (according to the Star Citizen devs) is that moving to this rendering pipeline would mean only the latest architectures from AMD and NVIDIA would be able to run it (AMD 6XXX and NVIDIA 2XXX/3XXX). Which means 80% or more of the DCS audience would no longer be able to play it. This reason alone means that it'll be at least five years - at the very least - before the vast majority of GPUs on the market support mesh shading. There are other limitations as well, but this single reason means it's sadly a non-starter for the time being. Still, if it does become the norm it could lead to some incredible performance improvements without a brute-force silicon approach.

 

Would that preclude ED from being able to perhaps offer mesh shading as a configurable option with the Vulkan implementation somewhere down the line for those with such devices ?

[GunSlingerAUS]

9 hours ago, shaun57 said:

 

Would that preclude ED from being able to perhaps offer mesh shading as a configurable option with the Vulkan implementation somewhere down the line for those with such devices ?

 

It requires a totally different design of the rendering pipeline; while Vulkan does support these features, it's not as simple as selecting a different rendering compositor, and would mean ED would need to maintain two totally diff APIs. Given the difficulty they have supporting one, I don't think we could ever expect them to start supporting two.

[ShaunX]

10 hours ago, GunSlingerAUS said:

 

It requires a totally different design of the rendering pipeline; while Vulkan does support these features, it's not as simple as selecting a different rendering compositor, and would mean ED would need to maintain two totally diff APIs. Given the difficulty they have supporting one, I don't think we could ever expect them to start supporting two.

oh well, just thought it was worth asking

[Brun]

These benchmarks seem rather misleading. The improvements only apply to a specific part of the rendering process, and the tests are setup to stress that.

 

If rendering a frame is like your weekly supermarket shop, 'mesh shaders' could simply make the checkout process eight times faster. Doesnt have any effect on how long it takes to drive there, park, get all your stuff in the trolley and drive home.

 

Obviously a pretty abstract comparison, but illustrates the point. Might not be as game changing as the articles suggest.

[TimberWolf]

1 hour ago, Brun said:

These benchmarks seem rather misleading. The improvements only apply to a specific part of the rendering process, and the tests are setup to stress that.

 

If rendering a frame is like your weekly supermarket shop, 'mesh shaders' could simply make the checkout process eight times faster. Doesnt have any effect on how long it takes to drive there, park, get all your stuff in the trolley and drive home.

 

Obviously a pretty abstract comparison, but illustrates the point. Might not be as game changing as the articles suggest.

Sure, it's a tech demo. But even a 1.5 to 2x increase would be fantastic. I'm sure many people would upgrade their video cards for that (assuming any were available).

[mhe]

@BIGNEWY Mesh Shading is available in Vulkan. Hence the question. Any statement on that? Of course it is not supported by all the graphics cards dating back to the stone ages, but as an option it would be tremendously helpful, especially for the VR players as they tend to run cards that can make use of it. 

[BIGNEWY]

33 minutes ago, mhe said:

@BIGNEWY Mesh Shading is available in Vulkan. Hence the question. Any statement on that? Of course it is not supported by all the graphics cards dating back to the stone ages, but as an option it would be tremendously helpful, especially for the VR players as they tend to run cards that can make use of it. 

 

I will mention it to the team, but I have no news to share about vulkan yet. 

 

thank you

[Lurker]

One other thing to consider. Brun touched upon this already. We will still be limited in how the game engine interacts with the whole system. Not just the graphics API. The major cause for slowdowns in DCS at the moment I believe are because the game engine is very CPU dependent. Unless this is changed significantly in the engine rewrites, switching APIs to DX12 or Vulcan will not significantly change this. 

 

Take one example, draw distance. Switching draw distance from MEDIUM, to HIGH or ULTRA in DCS has an huge effect on frames per second. And it's got nothing to do with your graphics card. I run a 3090RTX and in VR these draw calls are hugely taxing on our CPUs. 

 

Some of these things need to be offloaded on our GPUs, but that would take significant engine rewrites. I am sure that the the engine team at Eagle Dynamics is aware of this, hopefully the engine upgrades we are getting near the end of the year will improve this somewhat. 

[Brun]

13 hours ago, TimberWolf said:

Sure, it's a tech demo. But even a 1.5 to 2x increase would be fantastic. I'm sure many people would upgrade their video cards for that (assuming any were available).

In terms of tempering expectations, I don't think there's a cat in hell's chance of it being responsible for a 50-100% increase in frame rates. I'd guess 5-10% would be more realistic, if that.

 

As for CPU dependence, that's supposedly one of the key benefits of DX12/Vulkan over previous APIs. However one thing I understand about both technologies is that they're lower level than DX11 and maximising efficiency requires more work on the part of developers.

[mhe]

On 2/18/2021 at 10:53 AM, Lurker said:

One other thing to consider. Brun touched upon this already. We will still be limited in how the game engine interacts with the whole system. Not just the graphics API. The major cause for slowdowns in DCS at the moment I believe are because the game engine is very CPU dependent. Unless this is changed significantly in the engine rewrites, switching APIs to DX12 or Vulcan will not significantly change this. 

 

Take one example, draw distance. Switching draw distance from MEDIUM, to HIGH or ULTRA in DCS has an huge effect on frames per second. And it's got nothing to do with your graphics card. I run a 3090RTX and in VR these draw calls are hugely taxing on our CPUs. 

 

Some of these things need to be offloaded on our GPUs, but that would take significant engine rewrites. I am sure that the the engine team at Eagle Dynamics is aware of this, hopefully the engine upgrades we are getting near the end of the year will improve this somewhat. 

 

Yes, which is why ED is not only switching to Vulkan, but also overhauling the threading system of the engine, so it finally uses more than one (well, technically two) CPU cores. It is correct that Vulkan alone doesn't help much unless you also get rid of other bottlenecks. But the multihreading overhaul is on the roadmap for Q3 just like Vulkan is. 

[dburne]

10 minutes ago, mhe said:

 

Yes, which is why ED is not only switching to Vulkan, but also overhauling the threading system of the engine, so it finally uses more than one (well, technically two) CPU cores. It is correct that Vulkan alone doesn't help much unless you also get rid of other bottlenecks. But the multihreading overhaul is on the roadmap for Q3 just like Vulkan is. 

 

Nice hope we see good things from it.

[mhe]

So do we all, especially the VR users will rejoice with any performance improvements that can be had. I had the pleasure of occasionally experiencing 90fps without reprojection in very particular situations in DCS and it was something to behold. Having that buttery smooth performance without having to dial down your visuals so much would be pure joy. Flying through those new clouds in VR with framerate so stable that it could be real life, it would be mesmerizing to say the very least.