Question URP on Quest 2 Depth Texture enable Destroys the framerate

Hello everybody. I am developing a VR game with quest 2 using Vulkan as Graphics api. I need to activate the Depth Txture in the URP settings to do some fog effects. But when I activate this Depth Texture the framerate goes from 90 to 45/50... without doing anything else, without adding anything to the scene, just activating DepthTexture... it's normal? is there a way to optimize it?

I noticed that with OpenGLES3 it is better, should I make the switch to OpenGLES3? I prefer to avoid...

 

There indeed is an issue with vulkan and depth which is really annoying. I assume you are using Unity 2021? (2020 doesn't have this issue afaik)
If so, have you updated to the latest release already?
OpenGLES can definitely have better performance, so if your game does run with all your requirements on it definitely use it. Oculus is pushing vulkan with new features like phase sync, but it's not required

 

I think in the past someone already found the PR causing the issue. Is there any ETA for it? Will it be fixed in the 2021 LTS or will ot just be 2022?

 

So the testing shows that just regenerating the depth buffer is faster than copying it in the general cases the team tested?

 

Hello, any news on the fix?

 

On quest 2, the gpu is using a tiled deferred rendering architecture. To copy the depth buffer you need to resolve this texture. Depending on your resolution and MSAA level this can take several ms per copy. If generating a new one costs less than that, it's worth it. Really though, you should avoid needing the depth buffer at all on quest.

 

Same problem.

 

@rjonaitis
Hi,
Has this 2021 URP MSAA problem been solved? Is there a link? Where is the progress now? Thanks

 

I think it has been fixed yes

 

Has it been fixed?
https://issuetracker.unity3d.com/is...tween-urp-versions-when-built?page=1#comments

 

Was it fixed on unity2021.3.14?
This bug with the link? Thanks
https://issuetracker.unity3d.com/is...tween-urp-versions-when-built?page=1#comments

 

I tested 2021.3.16 and didn't feel any performance improvement,With 2021 URP and MSAA 4X problem.

 

Yes, it was a noticeable improvement for me.

If you have a project which performs more than 5% faster on the latest 2020lts vs 2021lts with the same settings etc make a bug report including both projects so they can fix it.

 

Same here. Let's stay in touch. I want to have a nice fog in the game. But I can't. Maybe 2022 version solved the problem?

 

You can have fog on quest 2 hardware, you just need to do it in your fragment shader.



Here's a screenshot of our recent game (minus the shadows and bloom, it looks the same on quest). Fog and a few other "post process"-like effects are possible, but you'll need to be writing HLSL.

 

I've tried this one, but it didn't help me. But thanks anyway.

My setup is as follows:
Unity 2021.3.16f1
URP 12.1.8
XR Plugin 4.2.1

I would love to hear your thoughts about my setup

 

Hmmm, I seem to be having a very noticeable performance hit when turning depth texture on and off in my URP settings. Vulkan performance is especially abysmal. My application might be unique in that I use camera stacking extensively (4 cameras in the camera stack + 1 camera for render to texture for a total of 5 cameras in a given scene)

Could this cause perf issues in VR?

 

Is there any reason for the difference a lot of people mention between OpenGLES and Vulkan?

 

Manuele, I was wondering how I should best handle some of my camera stacking use-cases:

- Base camera (skybox): 3d elements in the skybox: I guess I can replace this with a flat texture for every map, but its rather laborious to create a 2d skybox for every map.... This exists mainly to render far off objects without needing to have a massive draw distance.
- Avatar camera: This is the camera attached to the spaceship which is always sync'd with a camera in the cockpit
- Target box camera: This camera renders the target boxes so they always appear in front of the avatar camera
- Cockpit camera: This is a camera attached to the XR rig itself. Its motion is always sync'd with the avatar camera
- Display camera: This renders a small display inside the cockpit showing a readout of the enemy. I use a render pass here with a fresnel shader to both stylize and reduce rendering strain.


I suppose it might be possible to do the target box camera as a render pass, but I'm not sure exactly how since each box has a ui sprite on it which must need some kind of custom shader? Can render passes fix my problem here? Wouldn't they take even more resources to achieve the same effect?

I do feel like I need all these cameras, and the setup (just barely) works on the Quest 1, and can work on the Quest 2. I can tune down my graphics to get 72 fps on the latter platform, but it is difficult to say the least.


A picture is worth a thousand words. Perhaps this is helpful:

Edit: After investigating a little, it seems like a separate camera for targetbox is unnecessary. I can actually set the layer for target box rendering to -2 and ditch the camera, it seems

 

Upcoming cockpit VR demo? I hope ya'll didn't build it off my debug sample from 2 years ago!

Edit: Can ya'll please not release a cockpit VR demo? I haven't released my game yet...

 

Followup: I created a shader for the targetboxes, and it seems like sprite layering does not work at all with custom sprite shaders implemented in URP/shader graph, so I had to go back to adding a camera for target box camera stacking.

 

Ok, that cockpit demo looks quite nice. I can also see now that you guys are doing a great job in toning down the gfx for the demos (I don't say this sarcastically...that thing looks like it can actually run on lightweight SoCs)
I look forward to swiping the those skybox assets and that cool capital ship for Rogue Stargun VR.

 

That demo looks visually so much better than my solo developed VR starfighter game. I'm getting a bit self conscious even. I kinda hope y'all hold off releasing that demo for another few months although I am curious about the techniques the team came up with for lighting, shading, and camera synchronization

I'm actually shocked such a complicated shader with edge detection runs on the quest 2 smoothly. Is it expected to do better than simple lit.

 

AFAIK, Unity sample scenes are not commercial games. They're provided to demonstrate new features of the render pipeline.

 

Don't worry. Instead learn from it, improve your own game, and use your own creative input.
Just how something looks won't sell a game

 

My concern is someone will take the demo, add like 2 levels to it, and there will be many clones of this on the applab...

But in either case kudos. When I started with URP in 2020, camera stacking didn't even work properly for VR on the Quest 1 (the camera stack would disappear in the left eye). The tech has come a long way (although may still need improvement)

Edit: looking at the presentation, it also looks like the nodes that are used in that demo are not currently in URP. Anyways, I definitely look forward to trying out these shaders. I wanted to use an edge detection shader for the targeting display in my game, but ended up going with fresnel (which is probably computationally cheaper)

 

Also, I wanted to follow up on the sprite layering issue with URP/shadergraph. When you create a custom sprite shader, it does not respect layers and I believe this is a bug. I want targetboxes in my hud to always be in front, and right now the only way to do that is to have a dedicated camera to force the rendering order.