HD RP vs Built-In Pipeline Performance comparison

Unity with Post Stack 2 SSR : 73 fps

 

In UE4 colors are better and the wall roughness is better, the ground lighting is better if you look around the chairs.
I don't understand people always comparing to UE4, just use it if your goal is it's graphics

 

Just comparing frame rate. Because the UE4's scene has been made by Evermotion. Unity by me.
Also customizing color (contrast,gamma,color) is not too hard .I always love too bright and higher smoothness setup

 

Reflections are too high on Unity, or material have very low roughness and metalness in UE4.

Those two pics have the same frame rate counter, a yellow number on top right, are those really Unity and UE4 screenshots ? Can you show the same scenes in play mode including the editor ?

 

I think both scenes abuse PBR and hide it all with way too much post so it looks like a smear that has very little resemblance to fantasy or the real world.

 

Yellow counter is the fraps
I need to upload Unity with Stochastic SSR for download same scene
Here you can download exe files with SSR v1 for unity :
https://drive.google.com/file/d/0B2WDRR3zNNn8ZWtBb3ZGRTlqT0U/view
Here you can watch the video (color setup is not same ):

I'm going to compare HD and Built in on this scene today

 

1080P - Baked Lighting - No SSR - No AO
https://drive.google.com/file/d/1v2DIQcaeGnwfsjEJRXO6RRZGsRKEFI1R/view?usp=sharing
Workflow:
1. Run the game in Editor and take screenshot from Built-In pipeline
2. Create a new HD pipeline asset and assign it into the graphics settings
3. Convert materials from Edit->HD render pipeline
4. Disable shadows (it's not important in baked mode)
5. Run the game in editor and take screenshots
Built - In:

HD:

Spoiler: 4K

Built-In

HD:

 

In keeping up with srp and mainly the hdrp I saw this thread. Although I don't really plan to use unity 2018 till the end of the year and srp stuff probably earliest in 2019, its nice to know whats coming. There seemed to be a lot of bias near the beginning and I wanted to see some tests with more actual assets and a project in each pipeline from scratch and also if 2018.1 with builtin offers some improvements over 2017.1..
I created two identical projects with 2017.1.2p4 and 2018.1.0b12.
Everything is default (only changed to deferred and linear in 2017)
Imported 3dforge interiors to both (ran material upgrade in 2018, which upgraded everything nicely, except one (Legacy/Transparent/Diffuse material used for stairs)
Using SleepyRavenInn scene, which has a bunch of realtime lights
Placed camera at the same spot in both
Copied post process volume from 2018 to 2017
Using cts fps counter, these are the screenshots from editor and build respectively

2017.1.2p4 editor


2018.1.0b12 (HDRP) editor


then upgraded 2017 project to 2018
2018.1.0b12 (Builtin) editor


surprisingly the fps was similar in the build

2018.1.0b12 (HDRP) build


2018.1.0b12 (Builtin) build


I think that I was able to reproduce the scene in the different environments, keeping everything equal, only varying the unity version and the rendering pipeline. With that in mind I believe that hdrp, in addition to the possible visual quality improvements if used properly (I just used the provided textures which are not targeted towards hdrp, and have not change light intensity, or anything else) also offers improvements in performance with the base settings, and in its current state (at least for this or similar scenes).

​

 

I agree, like Enlighten at beginning, HD pipeline is just starting and it will need many iterations and fix issues before it is stable and fully usable.

Unreal and Unity are different on how they manage reflections, and UE4 has some pushed graphics features Unity deosn't have. I'm not sure comparing is so important.
Specially for outdoor environment where i find UE4 Skylight and Directionnal light produces a lot more natural lighting than Unity.

Anyway, HD pipeline is coming and that's good for people in Archi Viz or studios trying to make ultra realistic graphics games in Unity.

 

These are not technically correct, not empirically
From the technical point of view, a person invented a graphical technique, and everyone uses in their engines
All effects in Unity,Unreal and Frostbit are identical with various workflows.
For example Stochastic SSR has been used in Frostbite with 100 % identical quality. in UE4 with very similar result. PBR lighting model Invented by some persons and used in all 3d rendering softwares

Unreal just wants to change the names:
Eye-Adaptation -> Auto Exposure
Ambient Light->Sky Light
Tonempapping->Filmic ACES tonemapping
POM->Bump Offset


about sky light:
I've been researching this topic for months to see what makes it look better,and finally I realized that it's all about settings. UE4 default settings is very good .Sky light is not amazing thing, only the post processing are set correctly

About realistic:
The realistic result is only about the assets. in same scene, you can see same result:

HD for archviz:
It's not usable for archviz currently because of the some limitations in light settings (specially shadow radius on baking)

Comparing engines in terms of the quality of work is wrong. We just have to look at the frame rate :
Unreal blurs the image. And blurring always makes the result look more realistic

 

Your problem is shadows.
Shadows is On in Built in and is Off in HD (last two screenshots )
Turn off shadows for both and test it again
165k poly (HD) vs 1.3m poly (built-in). Shadows are actual 3d planes that castes into your scene

 

no, shadows are on in both setups, look under the table in hd.
the lighting is quite off, because I havent changed any settings

 

HD in default setting doesn't has shadows in long distance
But in Built In you have very long distance shadows with high quality
Also lighting is ON.
You can ope quality settings in Built in and reduce shadow distance to 10-30 to have same shadow distance and then compare the result. or complete disable shadows for both pipeline s

 

Why would you compare it without shadows though? As I understand thats one of the main improvements which the hd pipeline brings.

I have tried that but I havent found where to increase the shadow distance for the hd pipeline, the documentation is quite sparse (yes still is in very early preview). In general documentation for most of the goodness (mostly interested in the job system) that will be coming in unity 2018 is quite lacking considering the planned release date is around this time.

How do I increase the shadow distance? I will try with a greater shadow distance

 

HD shadow quality is poor currently . You can only reduce Built In distance to have same result
Also you can increase shadow resolution in light settings (512=>2048) but you can't increase shadow distance

 

I've not really tried to fiddle with shadow settings much myself yet but dont these settings work?



They are in the volume profile and can be overridden in the volume component.

 

yea that is mostly true. I have tried to change the resolution but it had near zero impact.

actually elbows is right there are the per volume shadow settings:

I have tried changing those but they dont seem to do much except turn stuff near the camera black when range is set to 0. There seems to be a hard coded max distance. Either these dont work, or I dont know how to use them. But dont really know since I cant check anywhere how to use that.

 

I do not have time to discuss
Just updated comparisons on 2018.2.12 :
You can try on your own system on unity's own templates

 

Sounds like you have a bug you could report?

What is your CPU and GPU for the tests?

 

The above test was on 8 GB RAM APU A4 6300

Current test on 24 GB RAM CPU 3770 and GTX 780 3GB

Please turn Off V Sync on Edit->Project Settings->Quality to see the actual result:

I tried different version of the HD RP and disabling Volume component. Result was same

You can try on your own pc:
HD RP:
1. Download latest unity version
2. Create a new project with HDRP template
3. Update HD RP from window->package manager
4. Test the result in 1080 P res on fixed resolution in Game view
5. Disable V Sync from Project->Quality settings

Built In:
1. Create a new project with 3D Extra template
2. Try in 1080 P on game view
3. Disable V Sync from Project->Quality settings

Animated Gifs :

 

Light Weight Pipeline vs Built-In :
Workflow :
Converted LW project to built-in :

 

Hi,

I am wondering why when you compare built-in vs HDRP with shadows:
built-in = 14.7 FPS.
View attachment 316570

But then you compare built-in vs LWRP with shadows:
built-in = 19FPS.
View attachment 316726

 

Because FPS-based "performance comparison" in the editor (and in general, but especially in the editor) is baseless and useless. Especially when someone just dumps a bunch of stuff in the scene without optimizing to the pipeline.

 

Yeah exactly.

HDRP still has allocations and so on and is built around the concept of a fully loaded scene of which the example is not.

But I think op's business model relies on builtin doing well.

 

Because i have changed some settings in Built in to have same level of quality as LW in Built in. Just limited Built in into the forward rendering path and some changes on shadows
LW is an optimized version of the Built in just on settings (forward,gamma,shadows,light limits, AA)

 

OK
Tested on an empty scene
Built-In version is same as HD (removed HD asset from Graphics setting). And switched to Deferred rendering path on the camera
Here is and built version test
Also i'm not an amateur user. I have optimized scene in maxed out to compare with UE4 .Also i have tested all possible setting to see any performance improvement on HD

 

You can show me any number of editor screenshots, it's not relevant. Editor is unreliable, it does things on multiple threads which may or may not hinder your entire measurement.

And I'm not saying that the HDRP isn't eating up more resources than the built-in, on the contrary, I expect it to be more resource-hungry, it's doing so much more, after all. Not to mention it is not really optimized properly yet, I guess.
So yeah, your numbers are cute and everything, but I wouldn't call them performance comparison.

But if you want to compare useless numbers: on my computer, the default HDRP scene runs (in the editor)

 

As I said before, I also put the build version with exactly identical results
Your image is ambiguous. GPU model, Resolution ,Built-In version

 

Identical to what? In your built versions of your own scene speciel the HDPR is better apparently (more FPS).
The FPSs are all over the place, so IDK what is 'identical results'.

My screenshot was made in editor (2018.3b5, 3.3.0 HDRP), the render resolution is on the picture: 1920x1080, the real resolution is 2560x1440, it's an Alienware 17 R4 laptop, with i7-7820HK, 32GB RAM, NVidia GTX 1080, Windows 10 Home.
In the background my Chrome is running and constantly eating 2-5%CPU and 5Gigs of ram (Youtube video is playing in HD).

 

In my screenshots (gtx 780 and hdrp) the fps was 111 on 1080 P
In your screenshots (GTX 1080 and HDRP) the fps was 160 on 1080 P

Your gpu score is 12315
My gpu score is 7960

so everything is normal

Now you must try Built-In scene and share the result to see higher than 600 fps on your machine

 

Built-In test build (exe for windows x64 x86)
HD RP test build (exe for windows x64 x86)

description:
3 moving cube with standard shader
no Sky box
switched from HD RP project
deferred rendering on Built-In
No shadows

 

The amount of rendering done inside HD pipeline is bigger than Built-in. More shading models, more lighting models, multiple passes for correct back-front transparent rendering, volumetrics, etc. They are very different renderers. Plus the C# overhead in-editor. This comparison is kinda silly.

 

HD render is more like HiFidelity render, of course it's cost more. You can have cheap render that run everywhere somewhat, cheaper that run on everything but with less feature (low fidelity) or luxury the render that only wealthy gpu can afford.

 

Batch Material converter to quickly switch between HD/BuiltIn pipeline materials for performance and quality comparison s :

 

Why is performance in HDRP worse than in built-in renderer ?

 

Because it does significantly more and also is significantly more new.

 

But on the same scene as in Built-in renderer, HDRP should be more performant while looking the same.
Then even task-heavy operations which are only in HDRP will worsen the performance even more.

Equivalent properties of materials and lights in HDRP and built-in renderer should be more perfomant on HDRP side by default.

 

That would be true if it were rendering the same scene in the same way.

 

It's not slower for me because I turn all the features on and actually use them. This even with the GC allocs the current experimental preview version does, it still faster than trying the same in builtin. And I don't have hundreds of volumetric areas with mesh decals with builtin.

So I do more and it looks better and it runs faster. But then, I'm not trying random scenes from asset store or the internet. I'm using features builtin truly struggles with.

So when people think they know what they're talking about with this subject I'm happy to ignore them.

 

You have a WIP thread for you game? interested to see how its looking

 

No, I never show anything until its basically finished, then I enter 6 months of marketing

 

So no dev logs etc? Pretty solid way of starting a community

 

Correct! I understand all about that and I don't need it

 

Dev logs are great if you just want to build a community of other devs, but ultimately they're nowhere near as important as wider community marketing.

 

Devlog are great when you have mostly locked feature set, constant progression and design mostly done. Ie when you aren't taking rmajor risk (especially creative) and are essentially implementing a well define idea.

I can't start a devlog, I have been struggling 5 years just to define a definitive visual features THAT NEED to be in. It's only when that feature is done I'm confident to start "safe" dev (even though many creative challenge remain).

 

While I would love to have a lot of money from my game, I don't particularly care if I don't. Because I've been dirt poor all my life and whenever I get money, I give it to people. It's because when I was very young, my grandfather taught me how money changes people and its worthless. It stuck so hard I've lived by it ever since.

I will try to earn the maximum amount of money so I can help people and pay for health care, plus I have plans to help with animal welfare issues (these are things dear to me) but I don't actually make the game FOR money if that makes sense, so decisions I make are not the same as advice I dish out for others, which I do, personalised for whoever is asking I guess.

You guys are so evil dragging me off topic.

 

Updated "Batch Material Converter" to have Lightweight shader option

Lightweight vs Built-In pipelines :
- No realtime Light
- No shadows
- No HDR
- Render scale 1

 

Hi!

I've been trying to work with HDRP 4.6 for a few weeks now but I don't get what you said: you are saying that since it does not render things the same way, it's normal that it doesn't have the same performances as the builtin renderer and I don't get that. I'm not comparing internet assets or editor builds but if I turn off every features of HDRP for an empty scene I should have at least as much FPS as the legacy one has, don't you agree ?

While it does a lot of things like you said and it is faster than third party plugins for specific things like SSS, I'm still wondering why it's normal that in the early stages of a project, the FPS would already be bad. From what I see, HDRP still needs optimizations (which is fine, it's a preview and people are still working on it) and can't be used in production before at least 19.3 from what Seblagarde said and that's a shame because it's a very cool tech.
Anyway, I was just looking for threads about performances with basic settings of HDRP proving that's it's not there yet and found this thread so sorry for necro but I'm glad to see it's not just me.

(Btw my project is targeting PS4 which made HDRP lack of optimizations even more obvious but then again they never said it was ready for consoles.)

 

1. book of the dead runs on ps4 pro at 30fps smoothly, and looks better than pretty much everything, This is using an older and slower version of HDRP. But you have to work with it: https://unity3d.com/how-to/optimize-high-end-graphics

2. HDRP should be slower, at the moment, yes for an empty scene. SebLagarde already explained why: it does more work in advance and the link to optimisation above also toggles a feature on that does even more work in advance (prepass depth) so that the bigger rendering job is executed faster than built-in could hope to do.

3. In future I expect, when it's finished that they will look into reducing the cost of the up-front work. But most projects that favour HDRP do require the up front work...

For example in builtin, you would be rendering buffers over and over per post effect, per pass and so on, so when you do something ambitious it will just fall over and cry, while HDRP will go on like nothing bothered it.

What you should not expect is for HDRP to get much faster for PS4 at this point, you should be looking in depth what is making your project slow with it - see above link again. Many people hope that it'll be optimised for something like switch one day, and I think it' possible but realistically we would never expect a switch version to run as fast as on an engine designed for it like LWRP.

It's designed for PS4 first by default.

This whole thread is riddled with ignorance and misconceptions and extremely bad assumptions (not you personally) so I wouldn't bother. If you're serious, post in the graphics experimental previews forum.