Mobile Shader / Lightweight Rnder Pipeline FPS Comparison

Test Summary:

#Edit: I went back and toggled on MipMaps for the textures on the ships and got a massive performance gain... The same scene ran at 30 FPS+ with mip maps enabled that ran at 2 FPS without. I had no idea MipMaps were so important.

I've been doing some real world tests of Unity's default Mobile Shaders and the new Lightweight Render Pipeline for an Android game I'm making and I figured I'd share the results since there are such drastic differences in FPS in my tests. Maybe it can help someone out.
I will also test Non-Mobile Shaders and Legacy Shaders just to see how they compare.

This is using unity 2018.2 and a Samsung Galaxy S4.

Here's the test setup:

24 Space Ships with around 10k polys each and 2048x2048 Textures / 2048x2048 Normals (on shaders that use normals) with a simple bobbing up and down animation.

Everything remains the same in all tests except the shader used.
All shaders have GPU Instancing enabled.

The Data:

Standard:
Diffuse + NormalMap only, Specular highlights on, Reflections off - 12 FPS
Diffuse + NormalMap only, Specular highlights off, Reflections off - 13 FPS

Mobile Shaders:
Mobile / Diffuse - 22 FPS
Mobile / Bumped Diffuse - 15 FPS
Mobile / Bumped Specular - 13 FPS (best performance with shininess)
Mobile / Bumped Specular (1 Directional Lights) - 8 FPS (God this is what I was using before this test. WORSE than the Standard PC shader with specular highlights.)
Mobile / Unlit - 15 FPS (How is it this bad?)
Mobile / VertexLit - 26 fps! (Winner for best performance in my test, no Normalmap or shininess)
Mobile / VertexLit (1 Directional Lights) - 13 FPS

Unlit:
Unlit / Texture - 16 FPS

Legacy: (Didn't expect them to be good)
Legacy Shaders / Bumped Diffuse - 10 FPS
Legacy Shaders / Diffuse Fast - 24 FPS
Legacy Shaders / Specular - 17 FPS (Surprisingly high)

Digesting the results:

One misconception I had was that the 1 Directional Lights versions of shaders were higher performance versions that didn't work with point lights or spot lights.
I don't exactly know what these variants do under the hood, but in a real world scenario they get HALF the framerate of their non-(1 Directional Lights) variants. These were the shaders I was using in my game before these tests for this precieved performance gain in my 1 Directional Light only scene. Turns out I was losing FPS for no quality gain.

I found Mobile/VertexLit to be the most performant of all the shaders I tested and will be using it for the things that don't need Normal Maps or shininess. I will also be using it for my lower LOD levels.

I see zero reason for Mobile/Unlit's existence based on these results.

Good thing I didn't automatically use Mobile/Unlit assuming "surely Unlit is more performant than VertexLit".

Lightweight Render Pipeline:

This one was a real shocker for me:
Lightweight Render Pipeline (Standard) - 2 FPS

Updated with tests from Simple LRP shader:
Lightweight Render Pipeline (Simple with specular) - 5 FPS
Lightweight Render Pipeline (Simple without specular) - 4 FPS
Lightweight Render Pipeline (Standard Unlit) - 13 FPS

So hang on a second... LRP gets 2 FPS on my phone but Standard gets 12 FPS?
I thought the LRP was built to be MORE performant than Standard for VR and such?

I assumed the LRP was a lot more mobile friendly than it is.
My message to anyone thinking of upgrading to the lightweight render pipeline:
Test both the mobile shaders and the LRP and see if LRP is a good idea.

LRP does support a lot more options than mobile shaders which are quite basic, so if the FPS isn't a huge deal for you and your scenes are extremely basic you might be able to get by with LRP and better graphics. But I have to say I'm a bit disappointed in LRP's performance. Hopefully they'll improve it because it seemed like a huge step forward for unity on mobile and VR.

Final Words:

These tests obviously don't apply to all devices, but mine is a pretty standard albeit outdated mobile device in 2018.

Also I guess don't assume something is more performant because it sounds like it would be.
Test it.

Here's the website for the game I'm doing these tests for in case anyone cares:
https://nullreferencegames.com/exoplanets-online/

 

I was wondering if you could also profile and post the details

 

Hi, I have a similar problem / question.
I'm working in a toy game for mid-range to high-end phones. I tested with 20 bears (a lot of polys) and I thought that using my own Unlit material will increase the performance comparing to the LWRP standard material but it's exacly the same.

Why I have this results? I thought that the Unlit material would be less expensive for the graphics.

 

The "Gfx.WaitForPresent" use arround 70-80%
I'm testing with a mid-range phone (huawei p10 lite)

 

The problem was the textures, my bad... didn't thought that the texture resolution affects the performance like that.
Now I got 60fps with 100k triangles
Thanks for all

 

Yeah, I might have. I'll go back and test the simple lighting version and add the results to the original post.
Edit after tests:
Okay I ran the test again with LRP (Simple) with everything I know of turned to lowest possible settings. (Vertex lighting disabled, Depth texture disabled, Opaque texture disabled, Anti-aliasing disabled, dynamic batching enabled, shadows disabled) and I still got around 4 FPS...
With standard unlit I got 13 FPS.

Maybe LRP only runs better than Standard shaders on newer devices?

 

Model SCH-I545 UD
Completely stock and relatively low hours used for it's age. I wish I had more devices to run the same test on but I don't... sorry.
Just out of curiosity, I assume during your tests Mobile shaders (like mobile bumped diffuse) are still more performant than LRP (Simple Lighting, lowest settings) correct? Or during your tests is LRP actually more performant than mobile shaders? I would assume less performant since it can do more?

 

Alright, I'll submit a bug report I guess.
I started this trying to be helpful, not to be complainy. =p

I only had a single One-Directional Light in the scene when testing, so I'm not exactly sure why one directional variants performed worse than their non-one directional counterparts. Like you say it should be the same. (or better if there are directional lights.)
Each test was so short and I did build and run between each test so heat shouldn't have played much of a role.

When I said Standard Unlit I meant the Standard Unlit version of the LRP shader, sorry.

When I tested Standard Shaders it was lit (Diffuse + Normal) and LRP (Simple, Diffuse + Normal only), both with a single one directional light, standard ran at 13 FPS and LRP (Simple) ran at 4 FPS. (same model, same textures) So I think that's a pretty fair test. Could just be device specific like @aleksandrk said. The S4 is getting pretty old now.

Edit:
The craziest thing to me is I toggled on MipMaps on the ship textures and my framerate jumped to 30 FPS+ for LRP and every other shader. I had no idea MipMaps helped that much...
With MipMaps enabled, LRP (simple) (37 FPS) outperforms old Standard shader (29 FPS). Strange that without mipmaps LRP performs worse than standard, and with mip maps it performs better as you would expect.

 

@SaturnCO thanks for testing. Would be great if you could update your original post with your new numbers with mipmaps - my takeaway after reading the whole discussion is "things work as one would expect - LWRP is faster than Standard", is that correct?

 

When I first tested without mip-maps, LWRP actually ran slower than Standard Shader for whatever reason. (mobile)
After enabling mip-maps however LWRP outperformed Standard Shader as expected.

I guess the takeaway is:
1: Mip-Maps for textures on 3D objects are a must-have.
2: If you're aiming for super-performant mobile and don't care so much about quality you still might want to think about using Vertex-lit mobile shaders, or one of the mobile-diffuse shaders. (you can still add tint and stuff to these manually)
If you want higher quality/lower performance than mobile shaders with more options built in, LWRP is more performant than the standard shader for mobile, and looks almost as good.

I will probably be switching to LWRP at some point, but since all old custom shaders no longer work with the new render pipelines, that is one big downside. It will probably take some time for all the custom shaders in the asset store to be recreated in some fashion with shader graph for the new pipelines.
(Specifically There's a planet shader I'm having a hard time recreating quite as well in shader graph for LWRP as the amplify shader version I'm currently using.)
So, if you need unique custom LWRP shaders for your project you may be looking at making them yourself. (for now)