This plus the talos lobby and space walk exterior in prey would be the most beautiful location in gaming...really hoping for a prey 2 that supports this or maybe support patched into the original.
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
-
We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.
CryEngine 5.5 to support Non-RTX ray tracing
- Thread starter Dictator
- Start date
Honestly that feels like a reasonable tradeoff if it makes ray tracing functional on non-specialized GPUs/
OP
OP
Interesting things to see, a recursive reflection shows a planar/rectangular area light instead of the neon wire stuff in one scene:
ANd you can see the trails from the moving drone and how it fades away in the opening:
ANd you can see the trails from the moving drone and how it fades away in the opening:
Haven't used CryEngine in years but I'll give it another go, hopefully the documentation has improved.
Crytek are an enigma. They make some great stuff and they also make some horrible business decisions.
Shadows are easier case, so I would guess that they are supported.
OP
OP
Looking again you can see some other optimisations - > lower LOD mesh in the reflection:
your mspaint skills are unmatched keep up the good work
OP
OP
Every DF video is in fact made in MS Paint (inb4 "that explains the quality"). :D
Not GIMP? I am disappoint.
Seriously though, very interesting tech, can't wait to see some performance numbers and see how they're handling samples per pixel and the BVH's.
The wire mesh is not replaced by a rectangular plane, instead you are looking at the light underneath the neon sign in the reflection. (the reflective panels that open up are fully open in that reflection).Interesting things to see, a recursive reflection shows a planar/rectangular area light instead of the neon wire stuff in one scene:
ANd you can see the trails from the moving drone and how it fades away in the opening:
edit: as you mention LOD optimisation, might be an LOD related bug or limitation issue when you are dealing with reflections in reflections. (Should be perfectly preventable when making LOD's)
Last edited:
Looks like the demo might use voxel based ray tracing. The "Total Illumination" portion of Cryengine is used when enabling SVOGI.
https://docs.cryengine.com/pages/viewpage.action?pageId=25535599
Thanks to Colbert for pointing this out in the PS5 thread.
edit: Dictator I saw that you said that a voxel based ray-tracing solution can't deliver results like this? Would this imply that the entire ray-tracing implementation has changed in Cryengine? I don't want to post something misleading.
https://docs.cryengine.com/pages/viewpage.action?pageId=25535599
Thanks to Colbert for pointing this out in the PS5 thread.
edit: Dictator I saw that you said that a voxel based ray-tracing solution can't deliver results like this? Would this imply that the entire ray-tracing implementation has changed in Cryengine? I don't want to post something misleading.
Ah thank you.
Also to those wondering how raytracing works without rtx. Here is a pathtracer running in a webbrowser using only shaders. Source code is available to play with in recompile right in the browser.
https://www.shadertoy.com/view/MsXfz4
https://www.shadertoy.com/view/MtGyWK (You might need to lower the sample numbers to 16 or lower. My integrated intel GPU runs it at 30fps with 8)
Here are more examples. They will even run on your phone browser lol.
https://www.shadertoy.com/results?query=pathtracer&sort=popular&from=0&num=12
Implementation of the prototype is available in their github repo. To my understanding the basic idea is storing triangle mesh in SVO leaf node and using SVO as the acceleration structures. In theory this should perform slower than ray tracing using dedicated hardware unit like RTX, but looks like Vlad has done a lot more improvements since his first commit. I only had a glance at the code months ago so I'm probably wrong.
Also, seems like there isn't any glossy reflection in the demo, but that's perfectly doable using their approach.
Last edited:
I didn't know Vladimir Kajalin was still at Crytek. Amazing. Dude has been there for nineteen years now. Most people probably know him as the inventor of Screen Space Ambient Occlusion. Imagine if this tech gained similar traction.
Why is the geometry and texture different in the reflection? Its like there's an actual object/mesh there. Do reflections need meshes? Are these meshes actual meshes placed by the developer? Considering that some of these bullets don't have reflections.Looking again you can see some other optimisations - > lower LOD mesh in the reflection:
It reminds me of how I faked reflections in SketchUp, copped the whole model pasted it, flipped it on the blue axis and attached to the base o the original model.
This working on a Vega 56 is very, very good news for the future of ray tracing across a lot of platforms. It may not be quite as complex as RTX but it's still damn impressive and gives a very similar overall feel to the scene. Hope we get a breakdown of how it works at GDC.
Reflections need meshes as soon as you don't do them in screen-space (or by using some voxel representation I guess).
With raytracing for reflections (in an otherwise rasterized scene) the overhead for shading the reflective surfaces will increase significantly with the amount and detail of the geometry that you are raytracing into to render the reflection. So using lower LoD models for that makes sense.
Also, if it's a dynamic scene, you need to update or rebuild your spatial structure / bounding volume hierarchy to always represent the current scene, which is a significant task that also gets more efficient with simpler and/or fewer meshes.
I've thought ever since the RTX reveal that Compute Shader based rendering has not been explored enough and that nVidia's jump to hardware based raytracing was extremely premature. Compute Shaders can work rendering magic, and I feel like we've just barely scratched the surface of their capability.
Edit; This was posted with the understanding that this was a Compute Shader based implementation. I might definitely be wrong about that; however, I still have the above thoughts about the technique in general.
Edit; This was posted with the understanding that this was a Compute Shader based implementation. I might definitely be wrong about that; however, I still have the above thoughts about the technique in general.
The thing is that traversing non-uniform, hierarchical, potentially sparse data structures is not something current general GPU architectures are particularly well suited for.
Of course you can do raytracing on them -- you can run pretty much any algorithm on them -- but it makes perfect sense to have specific hardware for this particular job, especially in 2018 and beyond when we can put more transistors on a chip than we can use at the same time without it going up in flames (or, less spectacularly, throttling severely).
I was listening to a podcast and Arthur Gies was saying that, based on the amount of Ray-Tracing stuff at GDC this year, he thinks that it'll be a feature on next gen consoles.
Pretty wild to think about.
Pretty wild to think about.
I think you are on to something with this. Looks like only the bullets whose bottom edge is directly on the reflective surface are being reflected, even if some part of the bullets that aren't resting directly on the reflective surface should be naturally reflected in the water.
Last edited:
OP
OP
Yeah that is what I meant, the light underneath it, the artist placed rectangular area light that CE supports.
Everything I know about DXR or as to what VK RT will be is that devs do not do anything with the RT core, it is all in the driver. So devs do not do much with it, it just accelerates, so it will only become dead weight when there is some other way without hardware acceleration that is faster or offers up benefits it currently does not provide - no idea what the heck that future looks like.
This is what I get for not following CE so closely anymore :( If it is similar to that that would sure be unique, any idea in your understanding why some objects would not be represented in the reflections? A numbe of those shell casings for example.
The lack of glossy reflections sure is a curious in the demo...
As Cry_nic is saying, there will be an interview next week with Vlad and others, so we will get the details then I am sure.
That exodus to CIG and id did not claim him and a few others - the rocky core remains in some way and it shows.
Last edited:
Ultimately RTXnexiaits to accelerate certain functions around ray tracing
You can certainly do,that in software - possibly leveraging GPU compute where it's possible to accelerate further. And if they have optimisations/approximation methods to make the technique 'cheaper' it could be a possibility to get a reasonable improvement in reflections without dedicated RTX hardware.
Whether the cost is worth it (it will have overhead like when SVOGI was proposed for UE and then dropped when current consoles didn't seem up to it), is tbc of course. And it only seems to be delivering reflections, not the lighting parts.
You can certainly do,that in software - possibly leveraging GPU compute where it's possible to accelerate further. And if they have optimisations/approximation methods to make the technique 'cheaper' it could be a possibility to get a reasonable improvement in reflections without dedicated RTX hardware.
Whether the cost is worth it (it will have overhead like when SVOGI was proposed for UE and then dropped when current consoles didn't seem up to it), is tbc of course. And it only seems to be delivering reflections, not the lighting parts.
There is always a mesh but in screen space reflections (SSR).
It does not make too much of a difference if you use old tricks such as mirroring the mesh (what is a good trick for calculating reflections against a plane but not practical against almost any other shape) or ray tracing with the RTX cores, the presence of a "reflection" mesh that gets rendered mirrored or ray traced against is there in one form or another. It can be the same geometry, or a lower LOD, or a dummy mesh, or even a bounding box or sphere containing the geometry.
And also the fact that the calculation output of the ray collision or the rendered output data from the mirrored mesh in a G buffer is only half of the story, because with that data the engine must later shade it, that is another big deal on its own specially for deferred renders as for a semi transparent reflection surface you have two sets of deferred data to shade per pixel, and this just for a single bounce.
If this tech can actually be used in games without tanking performance or resolution then next gen might be more interesting in the graphics department than I thought.
I think the issue is (or atleast was) that despite being developed as a third party engine its toolset was still not as good as UE.
Not really tbh.
It might be present but won't be widespread I feel. Look at tesselation for instance, the current gen hardware supports it but barely any games use it. Or look at variable soft shadows with penumbra, Nvidia has had PCSS since like 2009 and even has a more advanced version HFTS, AMD's equivalent is CHS but barely any game use it even on PC...on consoles the only game to have any sort of variable penumbra soft shadows is RDR2.
There's a lot of smoke around next gen consoles having dedicated raytracing hardware. I personally haven't checked but Arthur from RebelFM has mentioned that AMD and MS have raytracing talks, and it's been reported that AMD is delivering raytracing hardware later this year, and MS is supporting raytracing in DirectX.
Not a lot of "dead weight", especially in the future. Also I'd imagine the opposite will happen actually - RT cores will become more capable and fast and more IHVs will adopt something like them.
This is the underlying idea. However the fact that only one GPU vendor has support for RT APIs right now likely mean that there are a lot of vendor specific code in any renderer using this functionality so far. Doesn't mean that the same code would be hard to run on a different h/w RT implementation but it may not happen automatically as is implied by DXR specs.
I highly doubt that Navi will downgrade compute comparing to Vega in any way.
So, put simply, if seen through wire frame mode only, rendered ray traced objects are still polygonal objects?