Next gen lighting technologies - voxelised, traced, and everything else *spawn*

[troyan]

Ray casting on turing for a 1080p image with 1spp sounds like it takes slightly less than 1ms, which is significant, especially as you push to higher frame rates. At 120fps you're using 12% of your frame time just casting 1 ray per pixel. But with Turing shading is still the primary cost of ray-tracing, especially because non-coherent rays are probably not friendly to gpu cache. I'd like to see gpu power measurements before and after the patch. People said the gpus drew less power with ray-tracing on before the patch. I wonder how much closer the power gap is now, as kind of a proxy for gpu utilization.

But there isnt 1spp in Battlefield 5. "Low" has only 13% rays per resolution. And yet the performance impact is only slighty better than with Ultra (40%)...

 

[DavidGraham]

I'm pretty sure vegetation was raytraced before in most scenarios, though there were buggy places too where it never appeared in reflections

Big ones yes (tree tops and big bushes), small ones no (grass, leaves). You can see that in the TechReport video and HardwareUnboxed. PCGH stated the same too. And now ComputerBase as well.

https://www.computerbase.de/2018-12/battlefield-v-raytracing-dxr-patch-benchmark/2/

I don't remember seeing them in under the ray tracing section in the bug report. They were excluded on purpose.

Question is are the falling leaves now reflected

Yes they are, but reflected in SSR.

 

[Scott_Arm]

But there isnt 1spp in Battlefield 5. "Low" has only 13% rays per resolution. And yet the performance impact is only slighty better than with Ultra (40%)...

Which means the bottleneck is not ray casting, but somewhere in the shading. They're not hitting good utilization rates of the other parts of the gpu when ray tracing reflections.

 

[eloic]

Big ones yes (tree tops and big bushes), small ones no (grass, leaves). You can see that in the TechReport video and HardwareUnboxed. PCGH stated the same too. And now ComputerBase as well.

https://www.computerbase.de/2018-12/battlefield-v-raytracing-dxr-patch-benchmark/2/

I don't remember seeing them in under the ray tracing section in the bug report. They were excluded on purpose.

Yes they are, but reflected in SSR.

Wasn't the point of RTRT to get the ultimate, perfect reflections? If we still have to mix "perfect" but incomplete reflections with SSR or other tricks, I don't know... I think I'd prefer they use the resources for real time GI + shadows, which would be great for destructible stuff. It's just that this "look, these are perfect reflections, but they are not, actually" seems a bit absurd. Or it could be just me being a little bit grumpy, now.

 

[DavidGraham]

Wasn't the point of RTRT to get the ultimate, perfect reflections?

You are still getting those for the vast majority of objects, except small foliage (grass and flying leaves). So dynamic material reflections are still there, off screen reflections are still there, artifact free reflections are still there. Again except for the small foliage part.

Or it could be just me being a little bit grumpy, now.

You are not grumpy, I think this is a middle ground solution from DICE until they implement the hybrid ray march system, which swaps SSR with RT when SSR fails.

 

[Shifty Geezer]

Game graphics have always been faked. Baked GI is GI, but not really. Would you pull it out? 'Volumetric' particle smoke is thick clouds, but not really. Should we get rid of smoke altogether then? You could go on and on. At the end of the day, is the result better than using for somethihg else, which, as you say, may well be better in lighting. However, these 'perfect' reflections are true reflections so a notable compromised step up.

 

[JoeJ]

Wasn't the point of RTRT to get the ultimate, perfect reflections?

I guess we have to lower expections a bit, because if we had perfect reflections, we would already have perfect GI too.

But the worse limitation here is not the combination with SSR (which is interesting: You could ignore expensive foliage during RT and get that from screen space instead?).
I think it's more the fallback to static cube maps for rough materials. So we still see unoccluded reflections with baked GI, and that's the main artifact of PBS in current games IMO.
It would be nice to use RT AO to reduce this, but BFV is too much of an outdoor game i guess and the artifact shows only indoors usually.

I think I'd prefer they use the resources for real time GI

I'd like to know how Metro / Enlisted do this. If anyone has some guesses...?

 

[eloic]

You are still getting those for the vast majority of objects, except small foliage (grass and flying leaves). So dynamic material reflections are still there, off screen reflections are still there, artifact free reflections are still there. Again except for the small foliage part.

You are not grumpy, I think this is a middle ground solution from DICE until they implement the hybrid ray march system, which swaps SSR with RT when SSR fails.

Yes, yes, I know all the other benefits, which I like. I'm just consciously pushing my demand a bit in this... Maybe to make a point so that this could be discussed.

Game graphics have always been faked. Baked GI is GI, but not really. Would you pull it out? 'Volumetric' particle smoke is thick clouds, but not really. Should we get rid of smoke altogether then? You could go on and on. At the end of the day, is the result better than using for somethihg else, which, as you say, may well be better in lighting. However, these 'perfect' reflections are true reflections so a notable compromised step up.

I've been supporting Dice and Nvidia for their move with RTX and I've shown my enthusiasm and questioned the excessive criticism towards it in several threads in this forum, so I think it's clear that I'm (mostly) fine with RTRT. It's just that I find it a bit of a contradiction in the specific aspect I mentioned, that's all. And by no means my opinion is "everything or nothing", so no, I don't think we should get rid of everything that is not 100% perfect. I actually agree with you and I understand that it's a proper reply to my previous post, which may seem a bit contradictory with my previous speech.

 

[Scott_Arm]

I guess we have to lower expections a bit, because if we had perfect reflections, we would already have perfect GI too.

But the worse limitation here is not the combination with SSR (which is interesting: You could ignore expensive foliage during RT and get that from screen space instead?).
I think it's more the fallback to static cube maps for rough materials. So we still see unoccluded reflections with baked GI, and that's the main artifact of PBS in current games IMO.
It would be nice to use RT AO to reduce this, but BFV is too much of an outdoor game i guess and the artifact shows only indoors usually.



I'd like to know how Metro / Enlisted do this. If anyone has some guesses...?

This vid goes into some reasonable detail about implementation.

 

[OCASM]

Wasn't the point of RTRT to get the ultimate, perfect reflections? If we still have to mix "perfect" but incomplete reflections with SSR or other tricks, I don't know... I think I'd prefer they use the resources for real time GI + shadows, which would be great for destructible stuff. It's just that this "look, these are perfect reflections, but they are not, actually" seems a bit absurd. Or it could be just me being a little bit grumpy, now.

This is just the beginning of the hybrid rasterization-raytracing era. Perfect RT in AAA games is ways off still.

 

[milk]

I'm sure shadowing will prove a much more efficient use of ray tracing resources, because it requires no further shading. As soon as the ray has gotten a hit, you have all the info you need. It might be the case Ray tracing, when hardware accelerated, proves to be no more expensive than using shadowmaps. I'm eager to see the results.

 

[eloic]

I'm sure shadowing will prove a much more efficient use of ray tracing resources, because it requires no further shading. As soon as the ray has gotten a hit, you have all the info you need. It might be the case Ray tracing, when hardware accelerated, proves to be no more expensive than using shadowmaps. I'm eager to see the results.

Thankfully, we'll see that soon with Metro: Exodus.

 

[JoeJ]

This vid goes into some reasonable detail about implementation.

Thanks, sounds like a first application of spatiotemporal variance filtering for GI.
I guess instead path tracing they shade the hit points with direct light and shadow maps.

 

[Dictator]

PCGH did their testing confirming the huge performance uplift.

EDIT: for further calrifications.

So before the patch, small foliage like grass and flying leaves were not reflected at all with DXR, but after the patch, DICE decided to make these objects be reflected but through SSR, on top of the ray traced reflections.

Old Post:
PCGH had several notes regarding the hybrid ray tracing methods, it looks like this is now used to make previously non-reflected objects, become visibly reflected. So some foliage (grass, flying leaves) and fine particles were previously not visible with the DXR reflections, after the patch they are now reflected through SSR, while the rest of the environment is reflected through ray tracing, large foliage like tree tops and branches are also ray traced.

So the patch increases the image quality in this particular aspect, but introduces some limited SSR artifacts into the mix.

http://www.pcgameshardware.de/Battl...5-Day-1-Patch-Direct-X-12-Nvidia-RTX-1269296/

Grass and leaves were not at all a part of DXR reflections last patch, now with hybrid ray marching they are included as an SSR overlay onto the RT reflections along with decals and a couple other things.

Hybrid system though only workks for some pixels and not all and has some specific requirements to actually apply as far as I understand it, but here is a screen where you can see objects not in the BVH being included due to it:

Wasn't the point of RTRT to get the ultimate, perfect reflections? If we still have to mix "perfect" but incomplete reflections with SSR or other tricks, I don't know... I think I'd prefer they use the resources for real time GI + shadows, which would be great for destructible stuff. It's just that this "look, these are perfect reflections, but they are not, actually" seems a bit absurd. Or it could be just me being a little bit grumpy, now.

It all depends how much performance people are willing to be OK with - also, somethings may not actually be able to be seen in RT reflections anyway no matter what due to when they are applied in the pipeline. Decals for example in frostbite would never show up no matter what unless they used the hybrid ray marching.

 

[Scott_Arm]

 

[DavidGraham]

So they shaved off an entire 1ms of rendering time from DXR with the parallel execution optimization, the higher the resolution (and hence the higher the cost of a frame) the bigger this impacts performance.

 

[jlippo]

http://on-demand.gputechconf.com/gtc-eu/2018/video/e8527/
The realtime updated lightmap in the end looks quite nice.

 

[Scott_Arm]

I time stamped to when they start demoing rtx.

 

[DavidGraham]

HOCP did an analysis with DX12 vs DX11 performance in Battlefield 1 on a RTX 2070, they found out that DX12 implementation is so bad that it slashes performance by as much as 22% before any ray tracing is activated. If DX12 was an equal performer to DX11, or a better one (as it's supposed to be), RTX performance would be higher.

https://www.hardocp.com/article/2018/12/17/battlefield_v_nvidia_ray_tracing_rtx_2070_performance/9

 

[Kaotik]

HOCP did an analysis with DX12 vs DX11 performance in Battlefield 1 on a RTX 2070, they found out that DX12 implementation is so bad that it slashes performance by as much as 22% before any ray tracing is activated. If DX12 was an equal performer to DX11, or a better one (as it's supposed to be), RTX performance would be higher.https://www.hardocp.com/article/2018/12/17/battlefield_v_nvidia_ray_tracing_rtx_2070_performance/9

This would only be true if the performance with raytracing enabled would be dependent on the rest of the rendering performance and not raytracing performance, and I'm pretty sure everything suggests it's the raytracing performance that's the limiting factor, which would be just as slow no matter which API you're using.