Digital Foundry Article Technical Discussion [2022]

[Cyan]

found this RT setting in the dx12user.settings file located in The Witcher 3 folder in the Documents folder. RTAOEnabled is active, and I wonder what's that AOSSSampling and AOPretrace. Those aren't present in the settings menu afaik and I just used RT Ultra and left it at that.

[Rendering/RT]
EnableAOSSSampling=true
EnableRT=true
Shadows=true
EnableRtRadiance=true
RTAOEnabled=true
EnableAOPretrace=true

 

[Lurkmass]

I think that @trinibwoy means It is easy from the perspective that ray tracing allows you to do it "elegantly" by doing something with the rays being shot when they hit the portal.
Morgan McGuire seems to be talking about how the fact that it can be done literally just breaks what PBR means since it is going above and beyond the conservation of energy.

I wouldn't exactly describe being able to accumulate an infinite amount of energy or radiance to be "elegant" but feel free to make your own interpretation ...

 

[hughJ]

How would a portal behave any differently than a waveguide or other optical path to shape light? You're doing an intersection test and then generating a new ray with a transform, just like you do with reflections, albeit here you would be applying a rotation and translation according to the difference between the entrance and exit. Anything behind the portal entrance is occluded, so you're not generating new energy.

 

[jlippo]

How would a portal behave any differently than a waveguide or other optical path to shape light? You're doing an intersection test and then generating a new ray with a transform, just like you do with reflections, albeit here you would be applying a rotation and translation according to the difference between the entrance and exit. Anything behind the portal entrance is occluded, so you're not generating new energy.

Yup.

Portal should just become 'hole' which the path of ray/light can take.

If each path ends in light or termination, no additional energy is generated.

I do wonder about the bounce amount though..
I wonder if they shooter additional rays out from path. (Well, they seem to do direct light sampling at least..)

 

[Shifty Geezer]

I wouldn't exactly describe being able to accumulate an infinite amount of energy or radiance to be "elegant" but feel free to make your own interpretation ...

How can it accumulate an infinite amount of light? It's no different to two walls reflecting light off each other even if the two walls are the same wall bouncing light onto itself with exponential decay of reflected light.

 

[trinibwoy]

How can it accumulate an infinite amount of light? It's no different to two walls reflecting light off each other even if the two walls are the same wall bouncing light onto itself with exponential decay of reflected light.

I’m interpreting his comments to mean that RT “allows” you to accumulate an infinite amount of light. I have no idea why that’s a bad thing though. It’s just math so of course it allows you to do that if you wanted to but it definitely doesn’t require or force you to do that.

 

[Shifty Geezer]

I don't see how, unless you have a transparent, emissive surface. Light comes out one side, through a portal, through the other portal back into the emissive subject and then some if transmitted through along with more emitted light.

 

[trinibwoy]

I don't see how, unless you have a transparent, emissive surface. Light comes out one side, through a portal, through the other portal back into the emissive subject and then some if transmitted through along with more emitted light.

The example McGuire gave is that an object can reflect light onto itself through a portal. So even if you attenuate the light contribution from each ray bounce the object itself keeps getting brighter as you look at it through multiple instances of the same portal

This could probably cause wacky stuff like the same object seen through a portal 4 layers deep being brighter than the original. This of course has nothing to do with RT being “bad” or “hard”. It’s just an interesting quirk of portals.

 

[hughJ]

The example McGuire gave is that an object can reflect light onto itself through a portal. So even if you attenuate the light contribution from each ray bounce the object itself keeps getting brighter as you look at it through multiple instances of the same portal

This could probably cause wacky stuff like the same object seen through a portal 4 layers deep being brighter than the original. This of course has nothing to do with RT being “bad” or “hard”. It’s just an interesting quirk of portals.

I don't think that phenomenon is any different than when you get hot and cold spots via caustics (or any collection of reflective or refractive surfaces.) You're reducing or occluding the light in one area which results in some other area inevitably getting an extra dose of energy. I would contend that you could produce the identical effect of portals with a set of lenses and mirrors (of course with the caveat that you're ignoring propagation time, which isn't being modelled here anyways.)

 

[see colon]

I don't see how, unless you have a transparent, emissive surface. Light comes out one side, through a portal, through the other portal back into the emissive subject and then some if transmitted through along with more emitted light.

It's just like how your character can accelerate to insane speeds using 2 lined up portals. If you have a light casting through a portal both into a second portal and onto a reflective object that reflects into both portal 1 and portal 2, that means you are casting rays into a portal onto an object and into the portal again and onto an object and into a portal again and onto an object.... We know it's 4 bounces, right? So if bounce 1 is off the object and through the portal, then that ray by itself would hit the object 4x, meaning that the object is 4x brighter than it otherwise would be. In the 2 walls reflecting each other example you gave, each wall would count as a bounce, so at max you would have a ray hit each wall 2x. Plus, because of portal placement, you could get a situation where a ray may not touch the object through the first portal, but does in the second. Which would mean that a ray may touch only in the 3rd. So on and so forth.

 

[Silent_Buddha]

The example McGuire gave is that an object can reflect light onto itself through a portal. So even if you attenuate the light contribution from each ray bounce the object itself keeps getting brighter as you look at it through multiple instances of the same portal

This could probably cause wacky stuff like the same object seen through a portal 4 layers deep being brighter than the original. This of course has nothing to do with RT being “bad” or “hard”. It’s just an interesting quirk of portals.

Portal lasers (turning light into a deadly beam via constant amplification)! Be careful which portal you use.

Fun with light going into portal which sends the light back into the portal. Infinite light loop where the energy builds constantly and nearly infinitely (infinitely in a vacuum).

Regards,
SB

 

[jlippo]

It's just like how your character can accelerate to insane speeds using 2 lined up portals. If you have a light casting through a portal both into a second portal and onto a reflective object that reflects into both portal 1 and portal 2, that means you are casting rays into a portal onto an object and into the portal again and onto an object and into a portal again and onto an object.... We know it's 4 bounces, right? So if bounce 1 is off the object and through the portal, then that ray by itself would hit the object 4x, meaning that the object is 4x brighter than it otherwise would be. In the 2 walls reflecting each other example you gave, each wall would count as a bounce, so at max you would have a ray hit each wall 2x. Plus, because of portal placement, you could get a situation where a ray may not touch the object through the first portal, but does in the second. Which would mean that a ray may touch only in the 3rd. So on and so forth.

Shouldn't calculating brdf at each hitpoint during path to light take care of this.

 

[hughJ]

Portal lasers (turning light into a deadly beam via constant amplification)! Be careful which portal you use.

Fun with light going into portal which sends the light back into the portal. Infinite light loop where the energy builds constantly and nearly infinitely (infinitely in a vacuum).

Regards,
SB

Why would shining a laser recursively through portals additively increase the power ? If that were true then shining it into opposing perfect mirrors would also add power, which is clearly not the case because that would imply that even imperfect mirrors could also have some non-zero accumulation.

 

[Silent_Buddha]

Why would shining a laser recursively through portals additively increase the power ? If that were true then shining it into opposing perfect mirrors would also add power, which is clearly not the case because that would imply that even imperfect mirrors could also have some non-zero accumulation.

Sorry, it's a bit of an obscure joke referring back to when abstract diagrams of how a LASER worked used the concept of light bouncing between 2 mirrors (one fully reflective and one 99.99% reflective).

IE a simplified version based on something like what is shown on this page.

LASER Full Form - javatpoint

LASER full form, LASER stands for, meaning, what is LASER, description, example, explanation, acronym for, abbreviation, definitions, full name

www.javatpoint.com

With the resulting "LASER beam" being the acummulated energy from all those reflected light "bounces" that eventually went through the non-reflective part of the mostly reflective end of the chamber.

The picture you'd often see back in the 70's and 80's was such a simplified version of that illustration that it spread lots of inaccurate interpretations of how LASERs actually worked. For the average person on the street that saw something along those lines the picture was the only thing they could understand ... thus the joke.

We used to joke about it back then when those simplified illustrations were making the rounds.

Regards,
SB

 

[PSman1700]

 

[Shifty Geezer]

It's just like how your character can accelerate to insane speeds using 2 lined up portals.

You'd reach terminal veolcity caused by air resistance. If that's not modelled, it's a fault of the simulation and not an innate thing wtih portals.

If you have a light casting through a portal both into a second portal and onto a reflective object that reflects into both portal 1 and portal 2, that means you are casting rays into a portal onto an object and into the portal again and onto an object and into a portal again and onto an object.... We know it's 4 bounces, right? So if bounce 1 is off the object and through the portal, then that ray by itself would hit the object 4x, meaning that the object is 4x brighter than it otherwise would be. In the 2 walls reflecting each other example you gave, each wall would count as a bounce, so at max you would have a ray hit each wall 2x. Plus, because of portal placement, you could get a situation where a ray may not touch the object through the first portal, but does in the second. Which would mean that a ray may touch only in the 3rd. So on and so forth.

What arrangement do you you have for portals and lights so this is happening? Where's the light source, what's the angle of incidence, and where's the observer? What's the nature of the final surface? If it's a perfect mirror, it'll accumulate zero light as it's all reflected and just bounces around the circuit. Otherwise it'll absorb some light and the rest will be bounced around, for another piece to be absorbed. That's no different to light bouncing back and forth between surfaces, such as the pages of a book. You'll get a highlight in the crease where the light that illuminates the page is concentrated between pages before reflecting into your eye, but it can't amplify the light.

Without a diagram showing the arrangement that causes this light energy amplification illustrating what I'm missing, I can't see any case where it can happen if the light transmission is modelled accurately.

 

[davis.anthony]

I would love to see the speed runs of the RTX version vs standard to see if anything has changed or broken as a result of the RT or RTX Remix.

 

[PSman1700]

 

[Dampf]

@Dictator I fully agree with you that non-scaling options for RTGI is hurting the low- to mid end PC GPUs and even higher end GPUs as well.

If the current RT implementation is indeed intended performance, then they should add some scaling options for it, seriously. It must be A LOT more demanding than on consoles.

The game cannot be played with raytracing on my 2060 laptop, at all. Even with DLSS Performance at 1080p and at way reduced settings, I cannot get a stable 30 FPS under any circumstances, which left me very disappointed. Even DLSS Ultra Performance is not helping (so there must be some kind of issue in the pipeline that hard locks it to under 30).
I expected 720p to 1440p at 30 FPS with only RTGI on (DLSS Performance), which is definately reasonable. A real shame, because RTGI adds a lot to the game.

3080 does not get a stable 60 fps experience as well, even with reduced settings at 1440p, far from it. This needs patching. After holidays, it would be nice if you could provide them some feedback when you get in contact with them again.

 

[Cyan]

@Dictator I fully agree with you that non-scaling options for RTGI is hurting the low- to mid end PC GPUs and even higher end GPUs as well.

If the current RT implementation is indeed intended performance, then they should add some scaling options for it, seriously. It must be A LOT more demanding than on consoles.

The game cannot be played with raytracing on my 2060 laptop, at all. Even with DLSS Performance at 1080p and at way reduced settings, I cannot get a stable 30 FPS under any circumstances, which left me very disappointed. Even DLSS Ultra Performance is not helping (so there must be some kind of issue in the pipeline that hard locks it to under 30).
I expected 720p to 1440p at 30 FPS with only RTGI on (DLSS Performance), which is definately reasonable. A real shame, because RTGI adds a lot to the game.

3080 does not get a stable 60 fps experience as well, even with reduced settings at 1440p, far from it. This needs patching. After holidays, it would be nice if you could provide them some feedback when you get in contact with them again.

there is a new patch -more info in the TW3 thread-. Tried but is bedtime right now. Wattage dramatically decreased, and I mean 60W-70W average less wattage, GPU went from 180W power consumption to 105-110W. All of this with RT Ultra on -which means Ultra+ and all RT options enabled. The performance has been increased -although they are still investigating certain issues and MSi Afterburner shows no numbers for me as of now-. It looks like 60, 60 something, I can't exactly say -though there are slight stutters at times-.

DLSS Auto (a hack I performed lol). The GPU is now actually resting on its laurels, 'cos I enabled Unlimited framerate and power consumption never goes above 128W-131W.

edit: false alarm, the game now launches the DX11 version by default -I checked and DLSS Auto was enabled but RT wasn't-, that's why it felt much smoother now, like 60+fps. This is how menus appear for me now.

edit2: Something is not okay, the version I am running is actually the DX12 version, but RT is unavailable.