Upgrade to Unity 6.x

[Frooxius]

One of the possibilities that has been floated around is upgrading to a much newer version of Unity from the one we're using.

I have previously attempted to do this in the past, but ran into issue where newer versions of Unity do not support Deferred + Single Pass Stereo, meaning either one of those would have to be dropped.

• Dropping Deferred - any scenes with large amounts of lights would either significantly spike in performance requirements or the number of lights would have to be limited and result in artifacts
• Dropping Single Pass Stereo (SPS) - this can potentially drop performance in VR depending on HW

Based on discussions, there's a potential where the performance benefits of newer version of Unity would outweigh the performance drop without Single Pass Stereo. If that's the case, we'd consider merging this officially.

However we'd need some solid data to ensure that dropping (SPS) does not actually affect users negatively. We'd need to see some variety of benchmarks to determine this:

• Wide variety of Resonite content
  • Especially complex worlds with lots of drawcalls
  • It might be good to compile list of content for everyone to go through for some standardized results
  • People should be also encouraged to "field-test" and just use it normally for a bit and watch out for differences
• Wide variety of hardware (CPU, GPU, VR)
  • E.g. it's important to determine that users with weaker CPU's will not be affected negatively
  • Make sure everyone who helps test this includes what HW they have

If anyone's interested in helping organize this, I'd suggest following:

• Prep a build that upgrades Unity (you could make PR against this repo if you're okay with us not merging it if results are negative)
• Prep list of content to test
• Create Discussion here with instructions
• Organize community users to do the testing
• We can help increase visibility with a ping in our Discord
• Gather a bunch of data

If stuff looks good, especially on weaker HW, we'd likely merge it.

[0xFLOATINGPOINT]

I can fork this repo and make the needed changes to it for what would need to happen for a upgrade to Unity 6000.x. It will be a pretty long list of files that get upgraded with such a huge engine version upgrade. Good thing is i have already done this once before with my own version of Renderite so up-streaming those changes into a fork of this repo should not be too hard

[Gyztor]

Got a mostly working Unity 6000 branch going, Only problem I have right now is Quest Controllers aren't tracking for me. But index people seem fine if anyone wants to try it its here. Also fully works on linux from a sample size of 2:

https://github.com/Gyztor/Renderite.Unity.Renderer/tree/6000-Testing

EDIT: there are a few edits I had to make to the Renderite Repo as well to fix XRDevice issues and also to get a bit of performance back to fix the dash double rendering. So if you want to do a full clean build of all of it from renderite to the renderer you may need this: https://github.com/Gyztor/Renderite/tree/6000-Testing

EDIT 2: i say than i magically just rebuild and quest controllers work now. guess i will post a release on my GH to have a working build there.

[Gyztor]

Took some inspiration from #2 and gave it a shot and managed to build il2cpp on unity 6000 too. might be worth looking at.

https://github.com/Gyztor/Renderite.Unity.Renderer/releases/tag/v6000il2cpp

EDIT: i didn't performance test when I posted this but I am getting nearly 170 frames in local. I usually only hit 100-120

[Zyro1331]

I ran some performance tests on your Unity 6000 renderer and threw the numbers into the mix with the other tests I ran on the IL2CPP prerelease renderer, here's what I've gathered;

Local Home

	Average	Lowest	Highest
Stable	3.7 ms (~267 FPS)	2.8 ms (~360 FPS)	9.9 ms (~101 FPS)
IL2CPP	3.7 ms (~270 FPS)	3 ms (~335 FPS)	12 ms (~81 FPS)
6000 IL2CPP	3.3 ms (~302 FPS)	2.5 ms (~406 FPS)	24 ms (~41 FPS)
Stable (Decoupled)	1.3 ms (~749 FPS)	0.87 ms (~1150 FPS)	4.5 ms (~224 FPS)
IL2CPP (Decoupled)	1.1 ms (~909 FPS)	0.74 ms (~1344 FPS)	2.9 ms (~350 FPS)
6000 IL2CPP (Decoupled)	1 ms (~995 FPS)	0.56 ms (~1801 FPS)	9.1 ms (~109 FPS)

Dunes

	Average	Lowest	Highest
Stable	5.4 ms (~184 FPS)	4.8 ms (~208 FPS)	22 ms (~45 FPS)
IL2CPP	5.4 ms (~185 FPS)	4.9 ms (~204 FPS)	11 ms (~89 FPS)
6000 IL2CPP	5.9 ms (~171 FPS)	4.5 ms (~223 FPS)	16 ms (~63 FPS)
Stable (Decoupled)	3.7 ms (~269 FPS)	3.1 ms (~321 FPS)	7.2 ms (~139 FPS)
IL2CPP (Decoupled)	3.6 ms (~279 FPS)	1.8 ms (~550 FPS)	27 ms (~36 FPS)
6000 IL2CPP (Decoupled)	3.1 ms (~321 FPS)	1.7 ms (~588 FPS)	9.4 ms (~106 FPS)

Essentially a Forest

	Average	Lowest	Highest
Stable	5.8 ms (~171 FPS)	5 ms (~200 FPS)	39 ms (~26 FPS)
IL2CPP	5.9 ms (~171 FPS)	5.1 ms (~196 FPS)	13 ms (~79 FPS)
6000 IL2CPP	5.8 ms (~171 FPS)	4.7 ms (~213 FPS)	15 ms (~66 FPS)
Stable (Decoupled)	1.9 ms (~525 FPS)	1.6 ms (~620 FPS)	6.9 ms (~145 FPS)
IL2CPP (Decoupled)	1.7 ms (~574 FPS)	1.4 ms (~704 FPS)	5.7 ms (~175 FPS)
6000 IL2CPP (Decoupled)	1.9 ms (~520 FPS)	1.5 ms (~678 FPS)	13 ms (~78 FPS)

Null Exception Eclipse

	Average	Lowest	Highest
Stable	4.3 ms (~233 FPS)	2.9 ms (~345 FPS)	9.9 ms (~102 FPS)
IL2CPP	4.5 ms (~223 FPS)	3.3 ms (~300 FPS)	12 ms (~85 FPS)
6000 IL2CPP	4.2 ms (~238 FPS)	2.9 ms (~344 FPS)	14 ms (~73 FPS)
Stable (Decoupled)	2.4 ms (~416 FPS)	1.1 ms (~925 FPS)	6.4 ms (~155 FPS)
IL2CPP (Decoupled)	2.4 ms (~410 FPS)	0.87 ms (~1155 FPS)	5.4 ms (~185 FPS)
6000 IL2CPP (Decoupled)	2.5 ms (~397 FPS)	1 ms (~998 FPS)	12 ms (~84 FPS)

Like the other tests, these were all captured in desktop mode with a low screen resolution and with VSync off, on a completely new database without being signed in. So far, it's really neck and neck with the normal 2019.4 build in terms of performance it seems.

[Frooxius]

For the Unity 6000 one, I think it'd be important to also test it in VR.

Since it doesn't use the single pass stereo, VR might be affected negatively, while desktop isn't.

[XDelta]

For unity 6000, should also add normal, non-il2cpp build in the comparison

[Gyztor]

For unity 6000, should also add normal, non-il2cpp build in the comparison

there is already one in the same repo.

Like the other tests, these were all captured in desktop mode with a low screen resolution and with VSync off, on a completely new database without being signed in. So far, it's really neck and neck with the normal 2019.4 build in terms of performance it seems.

thank you for such an in-depth test too. The performance gains in 6000 usually show up not in fps for everyone but more in lower CPU usage, as well as meshes generally loading faster. (mainly based on Mushroom's Starship being loaded). For me personally I get better fps but its hugely hardware based, but its still good to know it at least isn't having any negative impacts which definitely helps a lot to know!

[Zyro1331]

For the Unity 6000 one, I think it'd be important to also test it in VR.

Since it doesn't use the single pass stereo, VR might be affected negatively, while desktop isn't.

I've run mostly the same tests in VR now. I used ALVR with Frame Pacing disabled so my headsets refresh rate wouldn't limit my framerate, and I was mostly GPU bound in all scenarios since I ran them all with Decoupling force enabled. I was actually quite surprised to see a trend with Unity 6000 IL2CPP trending towards lower frame times despite missing the single-pass stereo.

Local Home

	Average	Lowest	Highest ("1% Low")
Stable	6.3 ms (~158 FPS)	3.5 ms (~285 FPS)	11 ms (~92 FPS)
IL2CPP	6.4 ms (~157 FPS)	4.9 ms (~202 FPS)	7.5 ms (~133 FPS)
6000 IL2CPP	5.8 ms (~173 FPS)	3.8 ms (~260 FPS)	6.8 ms (~148 FPS)

Essentially a Forest

	Average	Lowest	Highest ("1% Low")
Stable	13 ms (~75 FPS)	13 ms (~79 FPS)	16 ms (~63 FPS)
IL2CPP	14 ms (~74 FPS)	9 ms (~111 FPS)	14 ms (~70 FPS)
6000 IL2CPP	11 ms (~88 FPS)	11 ms (~94 FPS)	12 ms (~81 FPS)

Null Exception Eclipse

	Average	Lowest	Highest ("1% Low")
Stable	8.1 ms (~123 FPS)	6.1 ms (~164 FPS)	33 ms (~30 FPS)
IL2CPP	8.3 ms (~120 FPS)	7.4 ms (~134 FPS)	10 ms (~100 FPS)
6000 IL2CPP	7.8 ms (~128 FPS)	4.7 ms (~214 FPS)	37 ms (~27 FPS)

Dunes

	Average	Lowest	Highest ("1% Low")
Stable	14 ms (~74 FPS)	9.8 ms (~102 FPS)	15 ms (~65 FPS)
IL2CPP	14 ms (~74 FPS)	7.3 ms (~137 FPS)	15 ms (~66 FPS)
6000 IL2CPP	12 ms (~82 FPS)	9.8 ms (~102 FPS)	14 ms (~74 FPS)

[ClimberJ]

I've just played for an hour on the Unity 6000 IL2CPP build in VR and I did not notice any significant problems. I did not do benchmarking, but it did feel better, except for some somewhat regular significant drops in fps, but i believe those to be caused in FrooxEngine, as the simulation rate seemed to drop as well.

[Frooxius]

Can you share what HW you run on please as well?

The results seem promising, but without knowing what HW it was on, it's not enough to make a decision on.

My main concern is that users with weaker CPU's would be affected most negatively - faster CPU's can potentially bulldoze through the extra overhead without much notice.

[ClimberJ]

I am running a Ryzen 9 9950X3D and a 7900XTX with an Index.

[Zyro1331]

I'm daily driving a roughly ~7-year-old CPU on my system, but here's my full system specs.

CPU	AMD Ryzen 7 3700X (@4.2 GHz)
RAM	Corsair Vengeance LPX 32 GB (2 x 16 GB - 3200 MHz)
GPU	XFX AMD Radeon™ RX 6750 XT (100-438412)
MB	MSI X570 GAMING EDGE WIFI (MS-7C37)
SSD	Samsung 980 Pro 1 TB

[Frooxius]

Ooo! That's quite promising then!

[ModernBalloonie]

After testing for a bit on Unity6 in VR, I get significant performance degradation.

I get 45-50 fps at spawn in the cloud home, and I get 90 on the il2cpp 2019 renderer.

CPU: AMD Ryzen 5 5800x3d
GPU: AMD Radeon 7900xtx
RAM: 32GB

[XDelta]

I get 45-50 fps at spawn in the cloud home, and I get 90 on the il2cpp 2019 renderer.

How about for the main release as a comparison, also a consideration if this is in part your headset/steamvr cutting the framerate in half to stabilize framerates when you other.