GPU Ray Tracing Performance Comparisons [2021-2022]

AO doesn't require material shading and the rays can be bound in their traversal length, e.g. 50cm.

So AO is dramatically cheaper than other raytraced options.

Cheaper to implement or cheaper to run?


This page disagrees with you entirely regarding RE:V:

Resident Evil Village Ray Tracing Quality & Performance Impact of Different Settings | Page 4 of 4 | Hardware Times

So which data are you referring to?

That article doesn’t really tell us anything. The RTAO implementation has negligible performance impact and negligible IQ benefit. It doesn’t seem representative of a proper AO implementation.
 
Last edited:
Cheaper to implement or cheaper to run?
Both, but having a low performance impact is the more interesting aspect and is primarily why I compared it with other RT techniques.

That article doesn’t really tell us anything. The RTAO implementation has negligible performance impact and negligible IQ benefit. It doesn’t seem representative of a proper AO implementation.
I think you're going to have to qualify "proper AO implementation" in terms of the alternative forms of AO that the game provides. Without RT, is the game's AO "a proper AO implementation" in your opinion?

Separately, in terms of the performance cost of RTAO, I don't know of other data. It took me a while to find an article with numbers broken out for RTAO, so if anyone has better comparisons to discuss that would be cool.

I'm not sure which games have RTAO as an independently activated image quality option.
 
Both, but having a low performance impact is the more interesting aspect and is primarily why I compared it with other RT techniques.

Ok, I was suggesting that the incremental development effort for adding RT shadows for example is minimal if you’re already doing RTAO. I would like to see more uses of RT that show IQ benefit even if current hardware isn’t quite fast enough. Control and Cyberpunk are good examples of this.

I can’t find any info on RE’s GI implementation. It seems to be probe based with a very small radius around the player. Just guessing based on the “glitches” reported in the article.

I think you're going to have to qualify "proper AO implementation" in terms of the alternative forms of AO that the game provides. Without RT, is the game's AO "a proper AO implementation" in your opinion?

Separately, in terms of the performance cost of RTAO, I don't know of other data. It took me a while to find an article with numbers broken out for RTAO, so if anyone has better comparisons to discuss that would be cool.

I'm not sure which games have RTAO as an independently activated image quality option.

When I first read the article I thought they were saying RTAO had no visible impact vs having no AO.
 
I think you're going to have to qualify "proper AO implementation" in terms of the alternative forms of AO that the game provides. Without RT, is the game's AO "a proper AO implementation" in your opinion?

Separately, in terms of the performance cost of RTAO, I don't know of other data. It took me a while to find an article with numbers broken out for RTAO, so if anyone has better comparisons to discuss that would be cool.

I'm not sure which games have RTAO as an independently activated image quality option.

Deathloop has a seperated RTAO option. On Ampere the performance setting is 2% faster than the SSAO: https://www.computerbase.de/2021-09...t/2/#abschnitt_die_performance_von_raytracing
 
Ok, I was suggesting that the incremental development effort for adding RT shadows for example is minimal if you’re already doing RTAO.
CoD:MW shadows required intensive development effort:

https://www.activision.com/cdn/research/Raytraced_Shadows_in_Call_of_Duty_Modern_Warfare.pdf

  • For the start of our DXR adventure, we wanted to keep things simple.
  • We felt that sticking to visibility only algorithms will save us a lot of time, since we don't have to care about shading.
  • This left 3 main candidates: ambient occlusion, sun shadows and local light shadows.
  • We picked local lights, since we thought using per-light acceleration structures will help us deliver better performance, by keeping acceleration structures small, and any problems will be easier to contain. As it later turned out, this greatly helped with achieving high raytrace performance
So that doesn't appear as if it would fall into the category of "incremental".

I don't think devs should be aiming merely for RTAO, as I think we're now past the point where gamers will accept such an apparently minor improvement to IQ versus traditional hacks. But it does seem to have been a missed opportunity this past 3+ years, since AO hacks really harm the look of games in my opinion.

I should have searched earlier and then I would have stumbled upon this article:

The Dark Side Of Ray-Traced Ambient Occlusion (RTAO) | The Gamedev Guru

Most likely, ray-traced AO is something you can afford in your game.

Just keep the sample count low, adjust the layer mask and be frugal with the ray length. Oh, and don’t push your players to go all-in on screen resolution.

Useful conclusion:

And remember: while real-time ray-tracing is now cheaper than ever, you will still have to make room in your performance budget for it.

Yep, there's such a thing as a performance budget. Each RT technique eats into that budget. RTAO eats the least.

On the other hand, as developers climb the RT learning curve, their "second generation" ray tracing implementations should be much more compelling.

I think it's fair to say that 4A games really showed the power of the "second generation" with ME:EE and we can only assume that IQ and performance will enjoy another step up in their next project.
 
CoD:MW shadows required intensive development effort:

https://www.activision.com/cdn/research/Raytraced_Shadows_in_Call_of_Duty_Modern_Warfare.pdf

So that doesn't appear as if it would fall into the category of "incremental".

It's more work of course but they invited additional complexity by building a BLAS per light and then having to merge them. They also had issues due to differences between the geometry in the BLAS and the geometry that was rasterized. These aren't universal problems in all engines.

Naive raytraced shadows require just 3 steps:

1. For each pixel choose a light (many engines already do this, Modern Warfare chooses up to 4 lights)
2. Cast 1 ray per pixel through the scene toward the chosen lights (Modern Warfare targets 1/2 ray per pixel per light)
3. Denoise
 
Question : how do you treat the sun in raytracing?
Do you treat it as a small light source close by or as a massive (millions of times the size of the earth) light source far away where the rays would be emitted in parallel and you'd have to have a ray coming from every pixel of the sky(edit: in that case you could ignore the sun) ?
 
Question : how do you treat the sun in raytracing?
Do you treat it as a small light source close by or as a massive (millions of times the size of the earth) light source far away where the rays would be emitted in parallel and you'd have to have a ray coming from every pixel of the sky(edit: in that case you could ignore the sun) ?

The direction of the sun would matter, so I'd guess it would be a large area light.
 
Question : how do you treat the sun in raytracing?
Do you treat it as a small light source close by or as a massive (millions of times the size of the earth) light source far away where the rays would be emitted in parallel and you'd have to have a ray coming from every pixel of the sky(edit: in that case you could ignore the sun) ?

It’s the latter I think. Just like rasterization you would treat the sun as an infinitely large directional light that’s infinitely far away so all the incoming light is parallel to the sun angle.
 
The sun is usually treated as an area light in ray tracing otherwise you don't get soft shadows - this is frequently seen in non-ray-traced rendering hacks with the name "contact shadows".

The darkest part of the shadow is called the umbra and the "soft" portion corresponds with the penumbra.
 
Isnt raytracing only enabled in the gallery/photo mode? Performance probably doesn’t matter that much.

Unless that benchmark run is going into gallery mode, RT performance isn't influencing anything at all.

Furthermore, those AMD GPU results don't make a lot of sense, especially the Navi 10 ones.

I could try these out on my PC when I get home (Microsoft is offering those 3 months for 1€ again, so why not?) to see if I also get those numbers, but then again this has little to do with RT performance.
 
Doesn't it use ray tracing to effect the audio?
Forza Horizon 5 uses ray tracing – but not how you think | TechRadar
It turns out that Forza Horizon 5 will use ray tracing to affect the game’s audio, which is used throughout the world. The game will send out ray-traced audio paths that can detect walls, buildings and ceilings as well. This means you’ll hear the roar of your car’s engine sounds bouncing off of all the buildings around you, and that will change depending on the environment.
...
“As soon as we turned it on, it grounded the world in reality,” Strachan said. “It made the world really feel alive. If you’re listening with spatial audio, something like Dolby Atmos with your headphones on, you’ll be able to hear your car bouncing off the roofs as well.”

“Every material in the game has been set up with a different absorption coefficient, which is kind of an acoustic term which essentially says if it’s concrete it will bounce off more than foliage, which actually absorbs the sound, so it’s really dynamic to the world around you.”


Ray-traced audio will also affect multiplayer modes by using occlusion, as the game can detect where every single car is in relation to your car. If a car’s behind a building, for example, you’ll know not to head directly there as their engine will sound muffled. This means you could choose to cut them off instead, as you’ll be more of their location.
 
It is a cross-generation game, so certain limitations around that. With that said, it's the best Forza Horizon title by far.
 
Back
Top