• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Crytek demo DXR on Vega 56

There is a definite benefit to having dedicated hardware for Ray Tracing but you only need that for nVidia's implementation because like everything they do its so badly optimised, that doesn't take away from the RTX cards tho.

You could make the same scenes in Cryengine to the same quality and it would run above 30 FPS on my 1070. That's how bad nVidia's implementation of Ray Tracing is.
 
We are talking about Ray Tracing outside of RTX, don't be so bloody touchy, its pathetic.


And there are no comparison possible. Crytek demo has not been benchmarked on GTX cards and AMD haven't made DXR drivers so you can't look at Metro Exodus, Tomb Raider, BFV or the nvidia demos.


The only compariosns that can be done is Vega 56 vs RTX cards in Crytec demo where RTX cards are 8x faster, or the Turing vs Pascal cards in the games with DXR .
 
And there are no comparison possible. Crytek demo has not been benchmarked on GTX cards and AMD haven't made DXR drivers so you can't look at Metro Exodus, Tomb Raider, BFV or the nvidia demos.


The only compariosns that can be done is Vega 56 vs RTX cards in Crytec demo where RTX cards are 8x faster, or the Turing vs Pascal cards in the games with DXR .

Again with the "but but... RTX cards superdooper....." no one is disputing their Ray Tracing prowess

You misunderstand Ray Tracing in this instance, its GPU and API agnostic, it doesn't need special RTX hardware or software. That's the point of it.
 
Last edited:
And there are no comparison possible. Crytek demo has not been benchmarked on GTX cards and AMD haven't made DXR drivers so you can't look at Metro Exodus, Tomb Raider, BFV or the nvidia demos.


The only compariosns that can be done is Vega 56 vs RTX cards in Crytec demo where RTX cards are 8x faster, or the Turing vs Pascal cards in the games with DXR .

In the long run a hardware solution in theory takes away the need for the developer to special case stuff, frequently test and tweak their implementation to ensure quality and performance, etc. the Cryengine solution looks great but it will have situations where it doesn't quite work right, slows down significantly or has other artefacts, etc. unless the developer uses special casing and so on.

Unless someone makes a massive breakthrough - which definitely hasn't happened yet - CPUs simply don't have the performance to do software ray tracing even via tricking it up using things like voxel representations to compete even with first generation RTX hardware when the tech matures.
 
You are the one that is paranoid and making up numbers out of thin air to try and defend AMD.

Because we are talking about Ray Tracing on cards OTHER THAN RTX cards, you lot are just anxious that we are saying RTX cards are not necessary, its why you two are in here crying "oh but RTX cards are still better" go way with that paranoia...

I don't think DP is trying to defend AMD and is mentioning how 1080P 30 fps is still crap compared to what NVidia can offer with dedicated hardware. Come on Les, don't get all humpy over something that is a fact. Clearly you can even read that Crytek have to halve this and that to achieve any form of playable framerates?
 
There is a definite benefit to having dedicated hardware for Ray Tracing but you only need that for nVidia's implementation because like everything they do its so badly optimised, that doesn't take away from the RTX cards tho.

You could make the same scenes in Cryengine to the same quality and it would run above 30 FPS on my 1070. That's how bad nVidia's implementation of Ray Tracing is.
So now 30 fps is playable to you? I have read so many comments on how 30 fps isn't acceptable but when it comes to anything other than NVidia, it is fine?
 
I'm talking about a difference in performance Greg ^^^ you're arguing irrelevant semantics.

In the long run a hardware solution in theory takes away the need for the developer to special case stuff, frequently test and tweak their implementation to ensure quality and performance, etc. the Cryengine solution looks great but it will have situations where it doesn't quite work right, slows down significantly or has other artefacts, etc. unless the developer uses special casing and so on.

Unless someone makes a massive breakthrough - which definitely hasn't happened yet - CPUs simply don't have the performance to do software ray tracing even via tricking it up using things like voxel representations to compete even with first generation RTX hardware when the tech matures.

Ray Tracing calculations, at least in Cryengine is done on the GPU, not the CPU.

Crytek even supply a performance estimation in their developer documentation.

The performance depends on which GI settings are used. Usually on Xbox One it takes 4-5 ms of GPU time and on a good PC (GTX 780) it takes 2-3 ms (AO + Sun bounce, no point lights, low-spec mode). The fastest configuration is "AO only" mode; this provides large scale AO at a cost of about 2.5 ms on Xbox One

https://docs.cryengine.com/pages/viewpage.action?pageId=25535599
 
Again with the "but but... RTX cards superdooper....." no one is disputing their Ray Tracing prowess

You misunderstand Ray Tracing in this instance, its GPU and API agnostic, it doesn't need special RTX hardware or software. That's the point of it.

If by ray tracing you mostly mean cube maps with a little bit of raytracing on top then yeah sure, but people saying "oh look raytracing on a cheap card makes RTX overpriced cards stupid" are somewhat missing the point.
 
If by ray tracing you mostly mean cube maps with a little bit of raytracing on top then yeah sure, but people saying "oh look raytracing on a cheap card makes RTX overpriced cards stupid" are somewhat missing the point.

No, no cube maps, environment probes, just pure Ray Tracing.
 
If by ray tracing you mostly mean cube maps with a little bit of raytracing on top then yeah sure, but people saying "oh look raytracing on a cheap card makes RTX overpriced cards stupid" are somewhat missing the point.

In Cryengine a lot of it is using voxels (essentially representing the scene at a cruder resolution using the CPU) and some shader processing power:

  • First we prepare voxel representation of the scene geometry (at run-time, on CPU, asynchronously and incrementally).
  • Every frame on GPU we trace thousands of rays through the voxels (and shadow maps) in order to gather occlusion and indirect lighting.

It is a hybrid CPU/GPU solution but basically the same as other CPU attempts it runs into the realistic limits of the CPU or general GPU compute to do lots and lots of rays. This is largely used for indirect shadows and some global illuminations and is not a full unified ray tracing solution for the whole lighting model and has many limitations such as only really works properly with large static geometry and becomes increasingly less accurate with smaller and moving geometry and needs a lot of care to avoid artefacts like light leaking.

The demos that incorporate it also fake up other elements of ray tracing using cube maps, etc. hence why it isn't really comparable to a matured RTX solution even though that is also hybrid and uses a fair few render buffers, etc.
 
No, no cube maps, environment probes, just pure Ray Tracing.

From the article;
Furthermore, whenever possible we still use all the established techniques like environment probes or SSAO. These two factors help to minimize how much true mesh ray tracing we need and means we can achieve good performance on mainstream GPUs.

Environment probes are used to build cube maps, so again crytek say they are using cube maps and layering a small amount of voxel based ray tracing on top, not running on pure ray tracing.

Did you even read what they wrote?
 
"Wherever possible", IE where a static overlay can be used ^^^

In Cryengine a lot of it is using voxels (essentially representing the scene at a cruder resolution using the CPU) and some shader processing power:



It is a hybrid CPU/GPU solution but basically the same as other CPU attempts it runs into the realistic limits of the CPU or general GPU compute to do lots and lots of rays. This is largely used for indirect shadows and some global illuminations and is not a full unified ray tracing solution for the whole lighting model and has many limitations such as only really works properly with large static geometry and becomes increasingly less accurate with smaller and moving geometry and needs a lot of care to avoid artefacts like light leaking.

The demos that incorporate it also fake up other elements of ray tracing using cube maps, etc. hence why it isn't really comparable to a matured RTX solution even though that is also hybrid and uses a fair few render buffers, etc.

You have no idea what you are talking about, just watching the video should be obvious to you, if you had any idea what you are talking about and not just copying words you don't understand the context of into a sentience it would be immediately obvious to you that Cube Maps, or rather Environment Probes do not feature in that demo in any way shape or form (Edit at least not in the reflections), Environment Probes are a 2D static layer pasted over a reflective surface, as soon as you move out of its static alignment it's no longer aligned.

Watch it again and tell me anywhere where you don't see the reflection refresh its alignment with the camera motion in 3D space.


How many times do i need to post this?

Upto 28 seconds i'm using Environment Probes, beyond that's its pure Ray Tracing, see the difference.

 
Last edited:
From the article;


Environment probes are used to build cube maps, so again crytek say they are using cube maps and layering a small amount of voxel based ray tracing on top, not running on pure ray tracing.

Did you even read what they wrote?

It is obvious they aren't running on pure ray tracing if you think about it - full resolution reflections would require a minimum of 2 million rays per frame (realistically more) at 1080p and they are talking about a few thousand (which is fairly realistic for a CPU or hybrid CPU/GPU compute implementation) so they have to be carefully special casing and using a lot of environment probes, etc. to do it in the demos - you aren't limited to purely screen space with alternative methods you can render the scene from other positions into a buffer and prioritise for realtime the ones closest to what the player will notice - which is a common technique - other surfaces/objects might have reflections that update at a much slower rate.
 
Crytek Updates CryEngine Roadmap: Version 5.7 to Support DirectX 12, Vulkan and Ray Tracing

https://www.techpowerup.com/255385/...-to-support-directx-12-vulkan-and-ray-tracing

https://www.cryengine.com/roadmap

The next CryEngine version 5.7 due to launch in Spring 2020 SVOGI will have hardware ray tracing support through Vulkan ray tracing extension or DXR.

Interesting, it seemed Neon Noir looked very good on hardware agnostic ray tracing, it just a tech demo but it would be far too slow to use it in game without hardware ray tracing support so Crytek will add hardware ray tracing support in next engine version 5.7.

If SVOGI used software ray tracing point lights in Neon Noir tech demo then Vega 56 would ran Neon Noir like slideshow at 1080p 3 fps just like in original Star Wars Reflections tech demo last year ran like slideshow on GTX 1080 then Nvidia released toned down version last month which ran a bit better on GTX 1080.
 
It is obvious they aren't running on pure ray tracing if you think about it - full resolution reflections would require a minimum of 2 million rays per frame (realistically more) at 1080p and they are talking about a few thousand (which is fairly realistic for a CPU or hybrid CPU/GPU compute implementation) so they have to be carefully special casing and using a lot of environment probes, etc. to do it in the demos - you aren't limited to purely screen space with alternative methods you can render the scene from other positions into a buffer and prioritise for realtime the ones closest to what the player will notice - which is a common technique - other surfaces/objects might have reflections that update at a much slower rate.

That's called optimisation, mixing diffrent technologies to gether to get the same result with far less of a performance cost is exactly the right thing to do, having said that you're making assumptions based on something you don't understand, the Environment Probes are not in the reflections, they are in the static ambient lighting.
 
Back
Top Bottom