Official Unity Experimental HDRP DXR

keeponshading yes that I was thinking. Thanks!

 

You could use this Volvo scene/car instead - https://blogs.unity3d.com/2021/01/2...room-sample-template-and-volvo-xc40-recharge/

You just download it from the Unity Hub

 

Grimreaper358 can you provide for me link ? ...I cannot find it....

 

You can only get it through the Unity Hub with Unity 2020.3 installed.

You create a new project from a 2020.3 version and you can download it as a template then open it.

 

Grimreaper358 thanks!

 

btw....I check this car room show...looks nice but for us is like "old school system"
Here are my CPU very old renders from Lightwave/Kray and this is what I want from Unity DXR:

 

Yea, it's just there for you to get the car and add that to your DXR project and go on from there. You could also convert the whole scene to use Raytracing.

 

Thanks, but I'm asking on a very practical level for gaming: is dx12 something that is nearing production ready for real-time games, or still year(s) away?

I'm very aware what this does for offline rendering (I've tried unsuccessfully to get our company to switch renders to this for your reasoning), but thats not my question. I'd love to _think_ about using this for a procedural game that realistically wouldnt be done until 2022 (I'm well into the mechanics, but still far from real art, etc).

My probably unrealistic wish would be that this is getting in such good shape, that I could use it and allow non-rtx gamers to just turn off rtx GI / Shadows / Reflections through settings by then, but ... I'm very skeptical .

 

DGordon I think in your case you should stay with traditional Unity system. DX12 is more advanced and directed to RTX features.

 

Yeah, I was hoping to use RTX but allow people with lower-end machines to turn it off without getting a performance hit. If I can't do that, then I can't even think about if RTX would make sense for those who have it. Hence my curiosity about dx12. But I'm realistically expecting to stick with dx11.

 

You can work in DX12 without anything Raytracing or DXR. The only difference between DX12 and 11 right now is mainly Performance (DX12 is slower) and you can enable Raytracing/DXR, and DLSS for Nvidia GPUs. Other than that it's all the same.

 

Yeah, my initial question was how much slower and what the expected progress is. A dev noted a while back that they intend to get dx12 faster than dx11 eventually. I'm just curious what the current progress on that is and how its going.

 

Yeah DX12 for me, just gets skipped over as the performance is just really bad.

For example, in my open world project, the fps drops from 90FPS to about 45FPS on 5950X and RTX2080
I would very much like to use Raytracing and DLSS but the base performance overhead at the moment is ridiculous and forces me to stay with DX11.

Seriously though...what is up with DX12 performance in Unity?

 

It wasn't the main focus as getting HDRP in a propper state would take top priority. DX12 performance parity or better than DX11 should be sooner than later as it's on the roadmap.

https://portal.productboard.com/8ufdwj59ehtmsvxenjumxo82/c/201-directx12-out-of-preview

With the new consoles (mainly Xbox Series X) and DX12 Ultimate (with all its features for both PC and the Series X), we should be seeing DX12/U getting some love soon. Performance in a lot of areas will probably increase including raytracing since DX12U has some raytracing optimizations as well. I'm more excited about the rasterization performance parts, Mesh Shaders, Geometry Culling, VRS, etc. Still looking forward to the Raytracing/DXR performance as well. DX12U would probably take more time as GPU adoption is needed but general DX12 should be up to speed soon especially with support for the console.

 

Grimreaper358 yes, DX12U is simple next evolution technology...faster or later everybody must switch to this console.
DX12U offers new support for new GPU cards and this is like future happening now

For me as off line render guy using Lightwave 3D since 25 years is more than evolution ...it is REVOLUTION!
Having renders at least 5000 times faster (with same quality) is like paradise for me

btw
I just updated my free work portfolio gallery with my old CPU (LW/Kray) rendering at:
https://januszbiela.myportfolio.com/

 

Made this 5 second shot for a challenge (every entry has the same length of shot - hence why only 5 seconds, mocap for the character - I decided to use the otter - and 'circular' element - mine being the negative space top right). Rendered out at 8k, all DXR enabled.

 

Love it keep all this juice coming guys, looks like you're the champions of HDRP!

 

I find it hard to separate HDRP and DX12 as DXR is exclusive to DX12, which by all means, separating DX12 and HDRP is fundamentally not possible provided that HDRP is to be utilized fully. What is the point of HDRP if DXR can't be leveraged due to DX12 issues? Sure it will still work for some projects and small prototypes but it will not work for majority of people.

Also, its not like DX12 came out yesterday...UT had plenty of time...I don't know why they act like DX12 is still new stuff and that they are about to begin working on it...there is already DX12 + @ out and the world is moving on but we are still imagining the day DX12 will reach performance parity with DX11...

DX12 was announced in 2014...which means most tech guys had some access and was launched in mid 2015....its mid 2021 now.

 

Hence I'm asking. Since, Unity being Unity, for all I love it ... "we want to reach parity" could mean in the next year, or DX13 will someday come out and they'll decide to work on it then. Heh. So all that really matters is what they've currently already done, until they do something else.

 

Yea, I'm looking forward to propper DX12 support, DX12 Ultimate Features, and improvements to DXR as much as everyone else. With the new console (Series X) DX12 will need to be worked on, so I think it will get better sooner than later. Especially seeing commits on GitHub for the new console support.

Spoiler: Extra Ramble

Currently, everything works with DX12 it's mainly performance that's below average. As we all know DX12, Vulkan, and Metal will be the main APIs for high-end Graphics moving forward and the ones that will be supporting real-time raytracing, and other Pipeline Optimizations.

I wonder how long they will keep DX11 support for HDRP after DX12 is on par with/better than DX11? I wouldn't touch DX11 at that point and anyone using HDRP to make games would have a DX12 capable GPU as well as anyone who would be able to run those games. I mean even the previous-gen console that did support DX11 (Xbox One/One S/One X) also supported DX12 from early on so even if that's the target you'd be safe with DX12. So DX11 would be mostly irrelevant as I don't think anyone would be targeting lower hardware than that with HDRP. I'd even argue for URP to be used if that's the target hardware

 

Hi. Is there a way to get lightmapped geometry in raytraced screen space reflections?
Seems the lightmaps are not visible in raytraces reflections.

 

Edit
I made a post about DLSS + DXR earlier. After playing with it some more, I recommend turning off the "optimal settings" and play with the forced screen percentage in the dynamic resolution settings to find out exactly what each option does and find the sweet spot for your game.

 

Thanks. Your tips from the old post helped me.
Probably the problem is that most of the Post Processing effects, in special the DOF, need some special handling for DLSS which is currently not implemented.

 

Does anyone has an idea or the deeper technical knowledge if
MSAA and Raytraced Reflections are technical possible and probably somehow available.

For design review of cars still standing and driving at low and high speeds MSAA gives the quality and stability needed.
On the other side realtime raytraced reflections would add a lot of realism to this.

I cannot wrap my head around if possible or not in general.

 

m0nsky please provide link to example scene when you get next update!

 

Some dev shots from latest Builmark beta 5:

Not UE5 of course but at least something ).

Water shader - well, found only simplest of those on Asset store for HDRP + DXR (just a plane with animated texture/depth based effects), but with decal projector for wetness at water level + fog volume for water god rays = quite nice; still no caustics and chromatic aberrations in refraction ...

VFX graph for desert hot air distortion - quite nice (but works very different on high and low fps (amount of particles and their speed is different, no idea how to balance that)).

No terrain for DXR ray tracing - well, currently it's mesh desert terrain with 450K poly per 2x2km block and it's low poly version as ray tracing proxy on separate layer. Hard to modify but at least it's working fast enough.

 

Amazing as always
After playing around with Lumen this post was really needed.

What do you think would be the best automatic proxy generation method?

...and why the tree looks so great?

 

Any polygon reduction tool will do, if it's working with HDRP.

Those are scanned from real world via photogrammetry - they're from "Rural Australia" package that is free in Unreal engine store. Heavy unique textures for each tree trunk, though.

 

...some of my friends showed me new UNREAL 5 Lumen engine.....it is horrible! But is in early phase development so let see after 6-12 months what they will improve.

 

I have mixed impressions - while Nanite brings very, very detailed geometry plus super effective "static batching" for 100K-1M static objects in the frame with playable fps it feels like a step forward and makes me feel jealous ... but this is example of the one aspect of the game maxed out but with some painful sacrifices from other parts: ray tracing cannot render that amount of geometry details (even ray traced shadows/AO are too slow to render for that level of detalization) so it was reduced & partially replaced by the set of other techniques (relatively low res voxels + screen space). Lumen lighting packs too much of "gameish" limitations & drawbacks - feels like a step back in some core aspects (light leaking through thin objects, reflections are screenspace/low res mess, shadows artifacts and so on).

So, on the one hand we have universal, relatively hardware-independent real time lighting tech that really good for games - and animated foliage objects working (with shadows!), fast rendering of very detailed geometry + 10K free ultra HQ assets directly loaded from Quixel lib and mobile-oriented and tough features integration (like sky is looking cool out of the box)... Unity clearly beaten here, at least for now:

But on the other hand when it comes to quality and per-pixel ray tracing awesomeness Lumen is failing in too much places (and terrain still not contribute to the lighting, like in Unity lol):

In the end I cross my fingers for Unity - if DOTS (that can support 100K+ of dynamic objects) comes to ray tracing with virtual texture streaming working fine with dynamic & transparent objects - it will be better (and more future proof) foundation for Unity (and my Buldmark project I hope).

 

Lumen and Nanite in current state is an very immpressive dynamic GI system for Quixel Mega Scans and currently a hell of optimized for organic scaned geometrys.
There are several square kilometers of highly detailed rock formations with an integrated nice streaming, loading and unloading system



an if you move/change and toggle directional light or the skylight it gives you after 2 seconds very nice dynamic GI.





Lower the Directional Light and Skylight or place an emmisive mesh.




The debug rendering for Lumen or Nanite shows you what happens under the hood.




So the SDF generation for the GI is fully automatic too.
The Nanite preparation is also already done for all Megascan assets in the libary.
But with smaller geoemtrys under 10cm thickness you get problems with current implementation.

All in all i was quite immpressed.
However, there a lot of current limitations but it s already a very good Quixel Megascans renderer with dynamic GI.

The complete Megascans libary is fully integrated per drag and drop in the editor.
This is the hardest part because it lets you generate several square km streamed levels with dynamic GI and highest quality scanned content for free.
I think that was the main target and is fully solved for the early access.

so
- streaming
- HLOD´s
- dynamic GI for this area size
- auto Lumen SDF and Nanite Geo-> Tex preparation
- thousands of scanned prepared assets for all needs

finished per exellence

Meantime i try to get the MegaCity Entity subscene with HLOD streaming and Hbyrid Renderer V2 to run since half a year to partly solve only a subset of real world problems solved here.
Try to get Texture Streaming to work with manual setups since Granite was aquired.
Have to manually generate SDF´s with Houdini or projects found on Github for my cars and environments that i can collide with Visual Effect Graph. Have to read funny dynamic GI deprecation and use it in current state messages for the next 2 years.
Have to manually generate proxys to speed up raytraced GI.

Main Lessons learned for now is that the way Epic did for this early access already to deliver are fully working higly automated system for this scale you can already use for creating a big streamed dynamic GI level at highest quality is marketing wise a very good move.

I also see and like the more general solutions in current HDRP and the way Lex4Art uses it already in Buildmark.
And i hope the HDRP team can collaborate and integrate the other topics for scale, streaming and auto proxy generation, HLOD´s , Texture Streaming and stuff like this now faster to get workable base for scale.

Doing this on a real live project/game at this size would probably accelerate this and helps to focus on stuff people need first.
(e.g. No Crossfading instanced trees in HDRP! After years of smooth crossfaded trees LODtransitions with GPU Instancer i should show now blooping trees in HDRP? )
Then optimize it for workflow and in editor.

This could all be solved but to deliver an vast amount of HQ, prepared and now hell of optimized set of assets like the quixel megascans libary is surely an topic where it is hard to get in competion again.

 

keeponshading This Demo weights 100GB...forget...totally useless, hocus-pocus mass deluded action .
I saw Lumen GI...looks horrible:
- light leaks
- too strong bounce light GI
- too strong color bleeding
- too strong tone mapping
- horrible shadow quality
- horrible reflection blur quality
- flat GI in detailed geometry

Ghosting GI out of discussion (forget animation with this)
Of course This is early stage ...

 

The cards and SDF genration for dynamic GI in ArchViz needs a lot of debug rendering.
It s simply a limitation of the SDF and the automatic generation has limits too.
You need to model bigger walls for now and manually correct or generate your SDF´s .)
Under 10 to 15cm you get light bleeding.
Also nanite has no big benefit for regular flat surfaces you have in ArchViz.
But like i said.
I think the current drop is optimized for 10 square kilometers of streamed scanned organic geometry with dynamic GI.

And i think that was the plan to deliver, marketing wise and that smaller studios can start generting big levels and photorealistic outside areas for ArchViz, loacations and driving areas for automotive experiences and FPS stuff.

For a FPS with an large photorealistic outside area and dynamic GI it s already in a good state.

Automotive Exterior and Interior and ArchViz Interior stuff is currently not solved.

 

But Unity can do to.)
Thanks Lex4Art for some real valuable input!

Raytraced Reflections and GI on really fast move.
This is simply a test how to get Raytraced Reflections/GI to work on fast drive (300km/h +) and to show off the real RT value on car reflecting itself and reflections from the environment, to stetup an plan proxys for the environment and layers for RT in general.

Fast HDRP conversion of the Perinn424 project (GitHub) powered by Vehicle Physics Pro.

 

Some new tests:

 

#2

 

#3

 

And the question:
How we can deal with surfaces like mirror, there is some plans for DXR to improve this issue?

 

Short animation with CTAA plugin:

 

Did some more serious tests with DLSS (2560 pix, CTAA+TAA):
no DLSS:

with DLSS:


and yes, problem with DoF (hope will be solved soon)

 

Some extra shoots:

 

Extreme resolution:

 

Hi. Nice shots. Does CTAA HDRP stays stable at little faster realtime camera moves too?

Whats you opinion on partly mix such interior visualisations with an "partly" bake. Only this areas where you need technically 6 bounces+ to get the right visualisation?

ArchViz interiors like some areas in a car interior (foot area and behind seats and car sky) need some more bounces to get right..
I also ask me if Bakerys partly scene bake sector technology could help in RT interior cases too.
https://geom.io/bakery/wiki/index.php?title=Partial_scene_baking
Like using the sectors as proxys for raytraced GI or other trickery.

As an i example i tried MLS HDRP on the ArchVizPro3 interior.
It blends now

- Lightprobes
- Reflection Probes
- Lightmaps (+Directionals + Shadowmasks)

fluently in realtime.

Every Day, Evening, Blackout bake is a minimal 6min PLM bake for the complete scene.
Bakery is even faster.
So for now i prefer to spend this 20min one time bake and having high quaity result and interactive framerate over spending 25fps + in frameloss for much lower quailty realtime raytraced GI.







The framerate is over 90fps and also perfectly usable for VR.

Realtime Ratraced Reflections and in some cases Realtime Raytraced Shadows would add now real value to this scene.

But the raytraced feflections on mirrors and higly chrome parts doesn t seem to take my baked stuff into account and reflect not that what i precomputed.

See raytraced refelctions in the mirror is way off (no bake?) and lighting becomes only visible when i lit it with a flash light before the mirror.
Is there a way to get this right?




This transfers 1:1 to my car interiors.
For the right quality and performance an bake blend like this helps a lot.to get the needed quality and performance.

I want simply add realtime raytraced reflections and shadows to it.
Exteriors and Environment are quite nice alread without bake. See BuildMark from Lex4Art or my test here
https://forum.unity.com/threads/unity-experimental-hdrp-dxr.656092/page-26#post-7194868

So i want to use realtime blendable bakes mainly for car and archviz interiors + rt reflections, +rt shadows.
Exterior and environment could live without.

 

Sorry for cutting you questions

...no idea, never needed fast camera in arch renderings. But there will be another problem: GI ghosting effect - the faster you move the worse effect you see.

That`s why I added Unity to my pipe line, after over 25 years rendering I am simply "tired" with waiting, I want everything instant and real time. I do not want make anymore step back (bake or other half way solutions).
Technically, one bounce light in UNITY is enough (this is not exactly one bounce GI, it is more). I did some tests in Cornell Box , also my experience tell me this is enough to produce high quality off line renders.

This is about skill, how you deal with that without tricks

I do not need real time ....for me 1-5 fps make me ultimate happy! Still 5000+ times faster than traditional CPU rendering!

I can not help you with your pipe line because I need completely differed than you:
- I need connection between Rhino/Lightwave 3D/Unity (pipe line 100% done) this we use in company
- I need ultra realistic surfaces, no matter time render (pipe line 80% done with StackLit system, still I need some work with this)
- I need Ultra GI quality. Practically this part in Unity is very well done comparing to Unreal (not mention how bad Lumen it is...). I Need only develop further DLSS
- clean mirror is the biggest problem now (also in Unreal)

This can work for gaming very well!

 

Any idea how I can deal with this error ? (Alpha17):

"Calling Deallocate on pointer, that can not be deallocated by allocator ALLOC_TEMP_THREAD
UnityEngine.GUIUtilityprocessEvent (int,intptr,bool&)"

It is forever going error....

 

This was fixed in recently released alpha 19.

Upd: nope, this errors still there, spamming console a lot ...

 

Lex4art my mistake ...did not check HUB is there new alpha. Thanks!

 

Alpha 19
When I select Use Optimal Settings I get problem with camera view: the more I watch up or down the more I lose perspective. I get really huge fps boost like 4x
When I disable it I get correct camera view and same frames per sec ratio as without DLSS (of course with AA edge smoothing)

 

Also IES not working:
"
Asset import failed, "Assets/System/3.IES" > NullReferenceException: Object reference not set to an instance of an object
UnityEditor.Rendering.IESImporter.OnImportAsset (UnityEditor.AssetImporters.AssetImportContext ctx) (at Packages/com.unity.render-pipelines.core/Editor/Lighting/IESImporter.cs:68)
UnityEditor.AssetImporters.ScriptedImporter.GenerateAssetData (UnityEditor.AssetImporters.AssetImportContext ctx) (at <6bba75ed57cb4aab97e7ea7a8713b0aa>:0)
UnityEditorInternal.InternalEditorUtilityrojectWindowDrag(HierarchyProperty, Boolean)
UnityEngine.GUIUtilityrocessEvent(Int32, IntPtr, Boolean&)

UnityEngine.GUIUtilityrocessEvent (int,intptr,bool&)
"

 

Besides that there is a lot problems with new Alpha19 I see light in the tunnel
Also I noticed that AA solution is replaced by DLSS which make AA perfect! (I do not know why FPS is the same as standard system without DLSS but AA looks great)
Huge lag Windows system with Write/Read from hard drives with new Alpha + GUI windows lags (when you operate inside Unity is ok, but when you want to do something in Windows that cause lags)
Very often crashes.
And some renders (Still problems with DLSS and DoF):