Jove 2.0: DX11 Rendering System (Alpha Release)

@lazygunn
Yes, if you use Joves deferred pipeline all your base are belong to us. You do not get to choose shading model.

@Licarell
I mean writing my own terrain solution, in its entirety complete with streaming and deferred shading.

 

@lazygunn - Once again I agree, I will all come down to supply and demand,

 

There's a bit of a lack of consistency in what you're saying but no matter, this threads not the place for that. Anyways, road map looks fine to me, thankyou for posting it, and it's interesting to see ideas like new terrain systems being mentioned. A new terrain system on a fancypants renderer would be lovely. Im sure there's already thoughts put to it, just want my brain to uncook now and sleep then back to making the scene i've planned for Jove looking a bit more the piece - the only things i see really being needed to recreate it to some good state of its glory are in the next two milestones

 

Oh right,maybe it could be changed to water simulation? And on that note I think Assassins creed 4's ocean looked amazing

 

Well in Blender Cycles anyway theglass shader is whats used for shading oceans with a IOR of 1.4(at least what I use) and it looks very realistic

 

I know it sounds like it but, you made some vary valid points and got me thinking just as Aieth did when I was sure SpectraGI was the way to go.

I have a feeling I'll never look back...

 

Blender Cycles is a paradigm away as far as i can tell, it wouldn't really apply here, raytracers can simulate light transport in participating media with a powerful and comparitively slow process using the same principles light itself uses, realtime rendering relies on some smart trickery for deducing fast yet consistent methods to create the same effect, but it's not the same approach at all (I dont know if im preaching to the choir here). Even more to the point is that water these days is expected to have realistically generated waves and local displacent and such, wave generation can be very costly, the FFT approach well known to create beautiful but expensive waves and things like perlin noise are kinder in realtime but not as appropriate for recreating realistic waves. I've someones masters somewhere where they combined the two and it was nice. Additionally, the water surface would be a gameplay surface in any practical application of a water renderer and an api for gathering and setting information about the water for gameplay use would be expected. It's quite a bit to do, to make a water shader for your project and yours alone might not be too tough at all, but to make something that's a product is another matter

 

Bloody great results for an object without any proper maps, without baked AO and so on. Damn I love that renderer.
Stock comparison.

 

Downloaded it...loaded it up...and this is one of the most beautiful things I have ever seen in Unity. And this is without SSRR, GI, or some of these other effects you are writing for it??? Dude, please don't stop developing this. This is truly incredible!

 

I think with Jove being as different from Unity's pipeline as it could get compatibility with everything shouldn't be expected. I trust Aieth to make his own shaders designed specifically for the capabilities of Jove

And wow bac9 everytime you post new shots using Jove my jaw drops!!

 

I know the differen

I know the difference between raytraced and rasterized renderers,but this is the actual shader im talking about,not the ray-traced factor of the renderer. Blenders glass shader looks more or less the same as jove and looks like water when you have the correct settings applied.

 

Als

I am aware that this would not simulate actual oceans,but it is perfectly viable for things like puddles or a pond. I also wasnt telling Aieth not to implement ocean simulation,but pointing out that the wording in the road map was a bit vague

 

@ZedalisDesign
I have no intention of stopping any time soon

@bac9-flcl
Another great comparison, looking great. I have to ask though, what exactly is that a model of?

 

I'm having a whale of a time with this atm. its not perfect, i got a lot to learn, but it makes things so much easier than messing around with competing assets and unity's limitations

Will have screenies before too long i hope

I imagine youve seen this but a pal in a uk studio fired it at the guys and the vr one specifically at me but this is about as wonderful as it gets for latest gfx info http://blog.selfshadow.com/2014/08/14/siggraph-2014-links/

Only got a week left so get downloading!

Suggestions so far are: A one click tool for replacing the entire scenes regular materials with Jove materials. This would make some things all sorts of times faster when importing large scenes from external software

Tiny bit more documentation on some features, such as what the different boundaries on probes mean and suchlike

 

@lazygunn Thanks for the link, I have seen most of those around the web but I missed two of them

I will add creating a helper script to the to do list. How would you want such a tool to work though? Just a one click all your materials are now Joves Deferred Bumped shader?
I'll add in some more documentation It is a little unclear

The outer size of a probe is the size used for projection, meaning this size should align with the room/area you want the reflections to reflect. The reflection strength is then faded in starting at outer size and reaching full strength at inner size. The third parameter, affection multiplier, basically multiplies the outer size but only for what it affects, it is still projected to the same size. It is useful if you want it to affect a little more than it does but the projection is perfect, or the inverse if you want it to affect a lesser area.

 

Yep the tool would be a 1 click thing you use after importing your scene model in along with its materials and gives everything deferred bumped, or forward glass/transparent (A setting possibly) to anything with an alpha lower than 1, filling in maps where necessary. Can forsee it as a nice time saver and in addition, an instant look as to how your scene looks straight way. The scene looks a bit funky until the materials are replaced which isnt ideal

 

Actually in addition, the textures for the masks of the material types they described almost work like quick visualisation presets. A broader supply of these would be ace, obviously they're easy to make but someone new to the system would be up and running quickly if they knew of them. In this regard the tool i described could automatically add the dialectric or glass mask textures as is most suitable - much much faster workflow if you're modelling scenes in your modeller of choice

 

Also while the scene view isn't representative of the game view, until its rectified, if its on the cards, perhaps some simple tools to switch between cameras or move a camera round a scene would be great, because at the moment im finding materialing different ends of the area quite fussy, i should probably stop asking for stuff and make it myself but eh

 

Hi Jove,

Just noticed this thread. Jove 2 looks amazing from bac9's screenshots of the rendering. I have a quick question though. You said that Jove is a complete overhaul of Unity's rendering pipeline and that your are only using a few basic components of it. Will custom and bought image effects still work. We use amplify colour for all my colour grading, would that still work in Jove 2? Also, what about something like proflares? Basically, what can we expect to work and not work with Jove. Not really overly knowledgeable about rending techniques, so any clarification would be great.

Cheers and thanks,

 

As far as I see every image effect still works as image effects generally use their own separate render textures. Of course, some things should absolutely not be used (like SSAO image effect and tonemapping) because Jove implements far superior (quality and performance wise) versions of many effects that benefit from proper integration into rendering pipeline and reuse of data.

But using stuff like screen space anti aliasing image effects is alright and, as I've said, they seem to work without any porting required.

I'll ask my colleagues if they use the assets you have mentioned and can test them with Jove.

 

You can expect anything that only operates on an image to work. For example, I believe amplify color just takes colors and remaps them to other colors. This means it does not rely on anything else such as light direction or a depth texture. Basically it works no matter what texture you give to it, so it is gonna work fine.

Proflares I am not so sure of. Do you know how it works? Does it work by finding bright pixels and creating flares from them? Or does it work by tracking where lights all? If all it needs is an input image to generate flares, then it works. Otherwise it would need adaption to Jove.

Simply put, any post process effect that just works on the image, operation on a bunch of pixels with only the pixels as input data, works fine. Everything else needs help. So things like Unitys vignetting works, anti aliasing post process, color correction etc works fine. Light shafts on the other hand would not (since it uses the depth buffer, which does not exist in Jove).

 

Hi Aieth

I've got a question about how the forward shaders deal with clustered rendering. Am I right in thinking that only the deferred shaders will work efficiently with the clustered rendering or can Jove use it efficiently with both. For example suppose I had 200 small lights lighting up a metal object with a deferred shader. I'm assuming this would be no problem for Jove. If I then added a glass ball next to this metal object with a forward shader what would Jove do ? Would the 200 lights light it without much of a penalty, would jove simply limit the number of lights that effected the glass shader or would Jove just take a large performance hit ?

Also would it be possible to make a script that can take a forward shader designed to use lighting for unity and convert it into a shader designed to accept jove lights or perhaps make a plugin for shader forge where you tell it to use jove's lighting in the same way you can switch it to use skyshop or pbr.

Another suggestion that I have would be to create a shader that can accept textures in the disney brdf format as that appears to be the format that a lot of graphics packages as well as pbr substances seem to be trying to make a standard for pbr textures.

Thanks,

Ian

 

Jove works efficiently with both. The limitation that plagues the opaque forward shaders is that they have to be rendered twice, once for depth and normals to the GBuffer and then once to be lit. You could skip that, but then you would have pixels being shaded twice (like if you turn ZTest off) and no SSAO for forward opaque.

Forward transparent are not rendered twice so you naturally have to pay for lighting both the transparent object and whatever is behind it. The glass shader is the most expensive shader since it has to evaluate light in two directions (reflective and refractive light). It is more expensive but if it an issue or not I guess would be something you just have to test

As for the second question, that would not be possible. I cannot write a script to modify a shader without placing pretty severe constraints on what textures that shader uses. I mean, what do you treat as the metallic channel?

I do plan on refactoring a bunch of the lighting code so added lighting support to forward objects becomes easier. I recently had an e-mail discussion with a guy who had some pretty specific scenarios, and it sparked an idea in me on how I can make the system more modular. When you say a plugin for shader forge, do you mean shader forge actually supports plugins? I have not used it.

I am not sure what you mean by the Disney BRDF format? Do you mean Disneys shading model with sheen, anisotropy, subsurface scattering or do you mean some kind of texture layout?

 

Aieth, could you implement Adaptive SSAA?

 

@FPires
Trust me, you would not want that. There is a reason techniques like that one does not exist outside of tech demos

 

Why not, exactly? I assume it would be way faster than regular SSAA. I have to force 4x SGSSAA already on some deferred-games that are heavily affected by temporal aliasing like FFXIV but I get some fps drop in busy areas . Wouldn't Adaptive SSAA be faster than a full-screen SGSSAA?

 

What is the difference between what your doing and what marmoset skyshop or Lux are doing?

 

Almost everything would be faster than insanely expensive supersampling.

Most advanced anti-aliasing techniques used in production today (like temporal AA in UE4 or "HRAA" from new Siggraph papers) are temporal and avoid any form of supersampling or causing a pixel to be shaded more than once to maintain low performance footprint. They give very respectable quality, especially considering the resolutions going up and up while undesirable side effects like edge blur stay the same size. I fully expect supersampling and MSAA to become completely irrelevant as 4k resolutions become popular, but even 1080p is already enough space for those techniques to work well.

The first post explains everything. This is not a shader package working with stock Unity lighting pipeline but a complete replacement of that pipeline, with lots of corresponding advantages.

 

Again.. that tells me nothing. It looks like it does exactly the same thing. I see no difference. I scoured the forum post looking for something different... but it just looks like more image based lighting except now it cant support anything anyone else does. So again whats the point? If i can use lux or skyshop with something like candela screen space reflections and spectra GI... What benefit do i get from jove? And none of those require dx 11 either.

 

Well, yes, hence why I want it. Forced full-screen SSAA can get slow in some games.

Still I would like to see the option in at least one deferred pipeline to offer AA based on multiple sampling, and adaptive SSAA seems like the best implementation (that I've seen), since I'll take the sharp look of a technique that uses more information than one that removes information anyday. And given that I can see the edge blur on the pixel density of a LG G2 I don't think post-process AA is gonna do the trick for me for the next 10 years, 4k or not. UE4's temporal AA doesn't do it either.

 

It is as different as it gets. To start with it uses a completely different rendering model than Unity, which opens up a lot of freedom.
Jove supports, in the current version, up to 512 point dynamic point lights at a time. It also supports up to 8 local IBL probes in view at almost zero cost, compared to a single global. This works the same for any objects, no matter if it is dynamic or static, or if it transparent or opaque.
Jove also uses its own shadow mapping, which is of a lot higher quality than the default Unity shadow mapping. Supporting up to 6 cascades with very tweakable settings.
Then there is also the dynamic sky system. There are glass shaders tightly integrated into the pipeline, able to provide complete support for reflective/refractive lighting, both from IBL and analytical light sources.
Jove uses its own SSAO algorithm, which turned out really great at a low performance cost.
There is also built in physically based bloom, lens dirt and a tonemapping algorithm.
Lux and SkyShop are great, but you cannot compare a plugin which does a single thing to the unified approach that Jove brings. There is incredible value to be had in a unified approach, both for me as a developer but also for you as a customer. Performance wise the difference is huge and it also helps with work flow. I have said this before, but if Jove had only been about surface shading and IBL it would have been done in February.
As for the other products you mention I am not going to comment on an unremarkable implementation of SSR nor the vaporware which flat out lies in its promo material.

 

The issue with that algorithm is that it is a lot of work. And not a lot of people are going to want to spend that much of their frame budget on SSAA. As an asset developer I am afraid I have to cater to the needs of the majority, so I cannot justify spending the time required to implement such an algorithm. When the time comes I am going to be work on a temporal solution, I hope that you will find it acceptable though

 

Alright, that's understandable. Thanks for the reply!

 

Kurylo3d the main difference is that Jove uses clustered rendering. It's rendering performance is far higher with lots of lights. You could for example have 500 dynamic lights lighting up the scene. The lighting is also more accurate and you also can perform effects like translucency without much of a performance hit.

With conventional rendering techniques having 8 lights lighting an object requires 8 times the performance. With clustered rendering it doesn't. The limitation is that the initial performance hit from using clustered rendering is higher and it requires more advanced compute shaders hence the requirement for direct x 11.

If you have a project that needs to run on low end hardware like phones and only need a couple of lights you wouldn't use Jove if on the other hand you were making something for consoles or pc that needed very accurate lighting and 100s of dynamic lights then Jove would be the way to go.

 

This is stunning... and with the proper maps for objects, it will look like an offline renderer. I may have to give Jove a spin, but I don't want to break my pipeline (I'm using Alloy) and I want to wait until I see your skin shader and compare the quality to others. Keep at this man... I've been wanting offline render quality in Unity and this is damn near close to it.

 

It's not exact. There is no "conventional rendering", but some different techniques.
So yes, for forward rendering, 8 lights means 8 times the performances (assuming they illuminate the whole scene).
In Deferred rendering, no (that's why deferred rendering was created), the cost is almost the same. But it doesn't handle the transparency.
Jove 2.0 uses (if i understand well) clustered rendering, which combines advantages of forward (translucency and transparency) with the big number of lights.

 

Hi Aieth

Yes the Disney format I was referring to was a texture format used to record pbr textures. I think it's the format that UE4 uses and is also the format substance painter exports to.

It's detailed in this video

Yes shader forge does support plugins. If converting normal shaders to use your lights is not straightforward then I think it might be worth making a plugin for shader forge. Shaderforge does have a PBR pipeline so it would make sense. With shaderforge it is very easy to tell it what textures to use so that isn't the issue. I'm assuming any lighting nodes or functions that are used will not be recognised by Jove because shaderforge will output the shader assuming you are going to use unity's default lighting.

 

Deferred rendering already handles the number of lights. Thats built into unity. Doesnt require dx 11.

 

Do your probes fade into one another by location?.. is it box projected cube mapping?

As for the SSR.. dont know how u can call it or a lie or vaporware. They are shooting for a release for end of this month. Not quite sure how you can call that unremarkable either. Realtime GI and real time reflections can save a lot of man hours in designing an environment.

 

Hey Aieth, how is the texturing workflow handled with Jove? Are you working with a spec/gloss or metallic/roughness setup? Also what are the steps for texturing with this shader package? Also someone mentioned the use of Substances. I work with Substance Designer/Painter and it is very important that I get the look of Jove inside of the viewport while I am authoring textures... could you post some examples of common PBR surfaces, such as metal and stone.

 

kurylo, you may as well leave off really, if you don't know the difference between a shader pack and an entire rendering pipeline then your questions are going to be hard to answer in a 'this is bizarre, how do i answer this?' way

Needless to say it makes prettier things happen, faster, and the price is a high requirement on entry - the reward is a blinding fast very pretty bit of technology that is extremely easy to use

Which is where i'll say hi, the last day has been one of the most fun ive had with Unity, this thing is great, and everythinf feels so well integrated into other stuff that it feels weightless. I'l have a screenshot soon,, you'll be able to spot the yet-waiting-for additions but this is an alpha showing and only the start. It's a good start

 

First texture slot stores albedo and is multiplied by the main color property.
Second (PBR attribute slot) takes RGBA textures with the following order of maps inside:

R - metallic
G - occlusion
B - translucency
A - smoothness

Third texture slot is just normals.

Once you know that order you can quickly save Jove-compatible attribute maps using grayscale source images of roughness, metalness etc from any other PBR system or library.

 

This is good to see. And it is JUST like Alloy... so I could just plug in my textures and POOF, instant eyegasm... I would love to see how Jove handles skin as well when you get to it, I am planning on doing skin and metal in the same place (think about terminator)... it is good to have choices. If I focus only on DX11, I'll pick up Jove as well... and I also use Skyshop for my reflection probes, and that is not going anywhere. Keep this up man. It looks MARVELOUS.

 

Speaking of Alloy, their little utility that allows you to slot per-channel grayscale maps into a Unity editor window and save the RGBA file out of it looks pretty useful. I should look into writing something like that, shouldn't be too complex if I stick to plain saving and won't touch some of the fancier stuff like mip correction.

That will allow you to keep a library of attribute map channels and quickly slap complete maps together from selected components. Can be done in PS, but it's always nice to have the luxury of not leaving the editor and shaving a few minutes per material, heh.

 

Lazy gun, dont be a prick. Its a legitimate question. Its not asking whats the difference between shaders and rendering pipeline. Its a question of how is what JOVE does different asthetically from the competition. So far I see no difference. for the price you have to pay. Everything mentioned is something available else where. You want lots of lights. Use deferred rendering. You want this probe based shaders with roughness and shininess.. got at least 2 others to choose from and thats not counting whats coming in unity 5. Plus i have no idea if this one even does box projection or blends between probes by location. If it dont do that.. then its not useful for anything but rnedering stationary statues.

Dont treat my like i shot your favorite puppy. I am a potential customer deciding what route i go for a project that is starting soon.

 

Oop, not perfect, ive got a bunch to re-add, after being semi-conservative for its first outing. Only 5 and a bit million polys in view, according to Unity, heavy use of the glass shader etc ec

One day of work! Can't even dream of that kinda thing with just the usual shader things + banging your head against Unity

 

Woah @lazygunn, incredible work here! This is practically offline architectural visualization quality, not something you typically see realtime engines doing well.

Oh wow. I repeat, please read the first post, it answers exactly the question of difference between this deferred implementation and stock deferred implementation in detail.

Stock deferred is not actually that good performance wise when large amount of lights is involved, quickly starting to drag even at 50 overlapping light sources, to say nothing about 500 light sources Jove has no issue rendering at great framerate on comparatively old hardware (I think it was 60+ fps on GTX560?).

Aside from that, the first post explains in detail exactly why the traditional approach of slapping image effects to your camera is extremely problematic from performance, maintenance and quality standpoint and is incomparable to deeply integrating effects like screen space AO into the rendering pipeline.

As about your question about "aesthetic differences", I don't really understand what are you're asking. There is nothing subjective about physically based rendering, so the question about aesthetic differences is meaningless. Environmental assets, character textures and particle effects can differ in that, not stuff like light transport models.

 

Yep. This is why I use Alloy and Substance Designer. From SD, I can spit out the same textures very easily, and save myself a headache. I do have to wait for it to pack but that is okay... It looks REALLY good. And I am glad Jove is using Metallic/Roughness... and it has physically accurate lighting which I DO WANT but it is not a high priority. I am glad I don't do my texturing work in PS much. It is hard to do it and save, look, save, look, etc.

I need to read the roadmap again, I saw something about Post FX not working with this... or Aieth is making one just for Jove. And I hope you will support Substances as well... you already have the inputs.

In fact, @Aieth, once you start doing skin shader, we should discuss something. I am primarily a character artist and I would love to help develop your shaders.

 

This is pretty, even for a short time working on it. I am particularly loving the bronze sculptures... keep this up man.

 

I wasn't trying to be a prick, just indicate where you might as well leave it. You havent seen anything done by it, really, and its in an alpha stage, so why contest so vociferously

It doesnt do box correction, it does what seems to be called 'parallax correction', this seems to be both box and sphere correction, whatever suits your scene, and it's so well integrated that there is absolutely 0 hassle with it, and it gets very accurate and impressive results. If you say its been done before, and you recognise shader solutions are a different thing to a renderer, then can you point out where its been done before? PBS and SSR and all the other buzzwords you mention are a subset of the renderer, SSR is planned soon and given the pretty nifty state its already in i have no doubt it will be implemented extremely competently, and benefit from the cohesion of a one stop solution to most rendering things (is the idea i imagine). The only other asset i can think of off the top of my head that it could really benefit from as a whole that might be too much to take on and get Jove finished is Shader Forge, so Ian's comments on the subject are decent

(Investigate shader forge if you get the chance Aieth)