Unity Benchmark v0.1

I need to re-run all of these just to test aginst current OS/Driver but it looks like generally Unity has been improving its performance over time. (at least at drawing cubes with a bumped specular or equivalent standard shader).

 

I've deleted 5.3, got a massive performance hit with 5.3. Our game's runs like 3-4 times slower. Probably will skip till next major release.

 

Windows platform.

 

I'll prepare a bug report tomorrow and post the number here.

 

Are you using 3D physics, or elaborate shaders and have you profiled it and compared 5.2 and 5.3 to see where the slowdown was?

 

We are using 3D physics yes. We have custom CG PBR shaders, renderers, lighting and effects. We went from ~3.2ms on the CPU in 5.2 up to ~17ms in 5.3. That's enormous! I didn't investigated much where it comes but will do tomorrow after extracting a scene from our 60GB project.

 

5.3.1 Running 3 x Fantastic @ 2560 x 1600 full screen on my PC. (DX 11)

4.5 / 4.6 / 4.6.1 / 4.6.2 / 4.6.3 / 5.0.0 / 5.1 / 5.2 / 5.2.2 / 5.3 / 5.3.1
30 fps > 51,716 / 48,853 / 48,254 / 48,681 / 49,835 / 59,498 / 53,177 / 55,370 / 56,120 / 57,718 / 53,106
60 fps > 42,230 / 39,983 / 39,337 / 39,490 / 40,723 / 46,899 / 51,693 / 53,705 / 53,976 / 56,026 / 44,174

 

@Arowx
Why 5.2.2 instead 5.2.4?

 

I sometimes miss point releases due to work and life, or if I do build the benchmark against them and it's similar to the previous releases numbers just leave them out.

 

The latest 5.3.1 seems a bit shocking given the release notes claim to improve over 5.3 :/

 

Best to compare the performance on your own hardware or within your own project(s) for performance as build targets can vary e.g. different Graphical API's, Graphics Drivers and IL2CPP before you even factor in hardware and implementation differences.

 

Quick update ran the benchmark on Firefox nightly built to WebGL 2.0 (Unity 5.3.1p1)

Fastest
30 21,260
60 16,174

Fantastic
30 14,704
60 14,219

Not outstanding at the moment but it brings Firefox more in line with Chrome and Edge.

Note this version of WebGL 2.0 does not have all of the features fully working so fingers crossed things will improve and WebGL 2.0 will boost Chrome and Edge when they get the new API.

 

Just tried 5.4.3.f1 and got this...


5.3.4 Running 3 x Fantastic @ 2560 x 1600 full screen on my PC. (DX 11)

4.5 / 4.6 / 4.6.1 / 4.6.2 / 4.6.3 / 5.0.0 / 5.1 / 5.2 / 5.2.2 / 5.3 / 5.3.1 / 5.3.4
30 fps > 51,716 / 48,853 / 48,254 / 48,681 / 49,835 / 59,498 / 53,177 / 55,370 / 56,120 / 57,718 / 53,106 / 51,234
60 fps > 42,230 / 39,983 / 39,337 / 39,490 / 40,723 / 46,899 / 51,693 / 53,705 / 53,976 / 56,026 / 44,174 / 49,885

So a slight drop in 30fps and a boost in 60 fps.

 

Might want to enable instancing in the shader?

 

5.3.4 I'm going to do another test with 5.4 beta.

 

5.4b10 Running 3 x Fantastic @ 2560 x 1600 full screen on my PC. (DX 12)

4.5 / 4.6 / 4.6.1 / 4.6.2 / 4.6.3 / 5.0.0 / 5.1 / 5.2 / 5.2.2 / 5.3 / 5.3.1 / 5.3.4 / 5.4
30 fps > 51,716 / 48,853 / 48,254 / 48,681 / 49,835 / 59,498 / 53,177 / 55,370 / 56,120 / 57,718 / 53,106 / 51,234 / 85,418
60 fps > 42,230 / 39,983 / 39,337 / 39,490 / 40,723 / 46,899 / 51,693 / 53,705 / 53,976 / 56,026 / 44,174 / 49,885 / 82,719

So about 53% to 66% faster.

Note this is without instancing shaders being used.

 

5.4b10 Running 3 x Fantastic @ 2560 x 1600 full screen on my PC. (DX 12)

Note this is using the instancing shaders (found in the beta forum).

I used to use combine mesh static in the previous version, but when I changed over to the instanced shader it clashed and produced a drop in performance.

 

Might want to see more data where the bottleneck is going forward such as is it CPU limiting it or GPU? From what I can tell, this is a CPU improvement.

 

5.4b10 Running 3 x @ 2560 x 1600 full screen on my PC. (DX 12)



Added 90 and 120 fps tests but this takes a very long time (Note lower score probably due to me watching a movie or two).

Massive difference between Fastest and Fast at lower fps, and very little difference at Fantastic across all the fps settings, and ditto for 120 fps column??

 

5.4b10 Running 1 x @ 2560 x 1600 full screen on my PC. (DX 12)

Odd, I need to run this again 3x solo just to check.

 

5.4b10 Running 2 x @ 2560 x 1600 full screen on my PC. (DX 12)

Must stop watching tv while benchmarking!

 

5.4b11 Running 3 x @ 2560 x 1600 full screen on my PC. (DX 12) Nothing else running.

 

Could you make a bar chart, so its easy to compare, and maybe calculate projected estimate for unity 6.0...

*I tested some numbers here, bit messy still..
https://docs.google.com/spreadsheets/d/19GzqxATpAHhK9SV_4MBuzxtKt7c_dijsQ_JzJbqk-6k/edit?usp=sharing

 

The idea is to allow developers to get an idea of what Unity can handle on various hardware.

It will be interesting to see how the Vulkan API compares to DirectX 12.

 

5.4b12 Running 3 x @ 2560 x 1600 full screen on my PC. (DX 12) Nothing else running.

 

That's not true anymore as you've set minimum price. Either fix store page or the OP so it is consistent.

 

Cheers fixed.

 

Unity 5.4 Beta 17 results. Running 3 x @ 2560 x 1600 full screen on my PC. (DX 12).

 

Seems to be getting slower.

 

Unity 5.3.5f1

4.5 / 4.6 / 4.6.1 / 4.6.2 / 4.6.3 / 5.0.0 / 5.1 / 5.2 / 5.2.2 / 5.3 / 5.3.1 / 5.3.4 / 5.3.5
30 fps > 51,716 / 48,853 / 48,254 / 48,681 / 49,835 / 59,498 / 53,177 / 55,370 / 56,120 / 57,718 / 53,106 / 51,234 / 50,796
60 fps > 42,230 / 39,983 / 39,337 / 39,490 / 40,723 / 46,899 / 51,693 / 53,705 / 53,976 / 56,026 / 44,174 / 49,885 / 49,418

 

Again...

 

Unity 5.4 Beta 19 results. Running 3 x @ 2560 x 1600 full screen on my PC. (DX 12).

 

Unity 5.4 Beta 20 results. Running 3 x @ 2560 x 1600 full screen on my PC. (DX 12).

 

Unity 5.4f3 Running 3 x @ 2560 x 1600 full screen on my PC. (DX 12).
Note using default Instanced shader which does not have bump/specular.

 

well if you're going to change things like resolution or shaders each time, then history begins at each change.

 

With bump and normal maps on "Instanced" shader.

 

Test of the New Vulkan API in the 5.5 beta:

DX12 @ 1440x900

Vulkan @ 1440x900

Only comparing standard shaders, as instancing does not work for me on Vulkan.

 

It looks like Vulkan is about 1.3x faster on my PC/GPU combo using Unity 5.5 beta and Standard Shaders (Instance shaders don't work for me in Vulkan yet).

Try it for yourself -> https://arowx.itch.io/unity-cube-mark

Bottom two files CubeMark55DirectX12.zip and CubeMark55Vulkan.zip

Standalone PC only.

 

Unity 5.4.3 @ 1600 x 1200

 

Unity 5.5f3 @ 1600 x 1200

 

Looks like cubemark won't be the same again in 5.6

 

How many cubes is this exactly?
Saw this on the 5.6 beta page.

So does this mean games like Rome Total War will now be possible with Unity?

 

Simple answer Yes.

Complex answer you will probably need a GPU based Instanced Animated Mesh shader/solution, with LOD and Imposters. Unless Unity have one in the works for 5.6?

Note: You will probably only see this many cubes and performance without physics colliders/rigidbodies.

 

Unity 5.6b1 @1600 x 1200 [Vulkan]

 

Unity 5.6b1 @1600 x 1200 [DX12]

 

Unity 5.6b1 @1600 x 1200 [DX12] Instanced Shader

 

Unity 5.6b1 @1600 x 1200 [Vulkan] Instanced Shader