GPU Ray Tracing Performance Comparisons [2021-2022]

[chris1515]

Source for that? the game is years away!


Again, source for that? I haven't seen any of the developers of these games saying they won't use HW-RT.

The trailer of Silent Hill 2 is not using HW-RT this is sure at 100%. There is some light leaking not seen using HW-RT like in Matric demo. And it is maybe running on PC. If Tekken 8 use Nanite it is 60 fps on consoles and don't use HW-RT. They just said they use Unreal Engine 5 but no details about Nanite and Lumen for Tekken 8.

Nanite landscape?

Is it a good idea to convert landscape to static mesh and then enable nanite, so that you could remesh it and discplace with the texture of the landscape?

forums.unrealengine.com

If you want to use Nanite, you can't use HW-RT. For terrain this is Nanite or another technology.

 

[DavidGraham]

They didn't specify that they were using Nanite for foliage in the video ...

Right in the beginning of the forest scene he states ray traced foliage, volumetric fog shadows .. with thousands of foliage actors. In fact, the HW-RT runs slightly faster than raster with a much higher increase in quality, here time stamped for you.

The trailer of Silent Hill 2 is not using HW-RT this is sure at 100%. There is so light leaking not seen using HW-RT. If Tekken 8 use Nanite it is 60 fps on consoles and don't use HW-RT. They just said they use Unreal Engine 5 but no details about Nanite and Lumen for Tekken 8.

These are early trailers, HW-RT support may come later, also how do you know that the PC version won't support HW-RT?

 

[chris1515]

Right in the beginning of the forest scene he states ray traced foliage, volumetric fog shadows .. with thousands of foliage actors. In fact, the HW-RT runs slightly faster than raster with a much higher increase in quality, here time stamped for you.

These are early trailers, HW-RT support may come later, also how do you know that the PC version won't support HW-RT?

First we don't know if it is a console footage maybe this is PC. For example in Tekken 8, there is realtime trailer running on PS5 annotation. This is not Nanite foliage...

 

[Dampf]

These are early trailers, HW-RT support may come later, also how do you know that the PC version won't support HW-RT?

It's literally just a toggle between HW and SW-RT. If the game were to use HW-RT, the trailer would show that. Why would they not show the game in its best form?

 

[chris1515]

And I don't think UE 5 will kill HW-RT but I think we will see probably until next generation of consoles less HW-RT games. And it will maybe help HW-RT to evolve to something more flexible too.

 

[troyan]

And I don't think UE 5 will kill HW-RT but I think we will see probably until next generation of consoles less HW-RT games. And it will maybe help HW-RT to evolve to something more flexible too.

Na, UE5 will be dying very fast without meaningfull support for HW-RT. No reflections, no direct lightning. The Matrix demo looks just bad at night. Cyberpunk is years ahead and this is a cross plattform game.

 

[chris1515]

Na, UE5 will be dying very fast without meaningfull support for HW-RT. No reflections, no direct lightning. The Matrix demo looks just bad at night. Cyberpunk is years ahead and this is a cross plattform game.

Did I say it will not support HW-RT? Don't you know not all game are open-world inside a city? I say that games out of city will probably not use HW-RT. And some game inside a city like Silent 2 remake will maybe not use HW-RT.

 

[DavidGraham]

PCGamesHardware posted some new RT benchmarks covering recent hardware releases. Benchmarked at 4K max settings.

Cyberpunk 2077:
6950XT: 100%
3090Ti: 178%
4080: 225%
4090: 322%

The 4080 is 25% faster than 3090Ti and 40% faster than 3090 and 2.25x times faster than 6950XT.

Doom Eternal:
6950XT: 100%
3090Ti: 166%
4080: 200%
4090: 275%

The 4080 is 20% faster than 3090Ti and 30% faster than 3090 and 2x times faster than 6950XT.

Dying Light 2:
6950XT: 100%
3090Ti: 186%
4080: 235%
4090: 340%

The 4080 is 25% faster than 3090Ti and 40% faster than 3090 and 2.35x times faster than 6950XT.

Guardians of the Galaxy:
6950XT: 100%
3090Ti: 168%
4080: 190%
4090: 245%

The 4080 is 13% faster than 3090Ti and 22% faster than 3090 and 90% faster than 6950XT.

Lego:
6950XT: 100%
3090Ti: 164%
4080: 220%
4090: 310%

The 4080 is 35% faster than 3090Ti and 45% faster than 3090 and 2.2x times faster than 6950XT.

Metro Exodus: (Medium RT)
6950XT: 100%
3090Ti: 155%
4080: 180%
4090: 250%

The 4080 is 15% faster than 3090Ti and 25% faster than 3090 and 80% faster than 6950XT.

Riftbreaker:
6950XT: 100%
3090Ti: 158%
4080: 196%
4090: 250%

The 4080 is 25% faster than 3090Ti and 35% faster than 3090 and 96% faster than 6950XT.

Ghostwire Tokyo:
6950XT: 100%
3090Ti: 146%
4080: 188%
4090: 255%

The 4080 is 30% faster than 3090Ti and 40% faster than 3090 and 88% faster than 6950XT.

Nvidia Geforce RTX 4080 16GB im Test: Ein fast perfektes Gesamtpaket - wenn der Preis nicht wäre [Update]

Nvidias Geforce RTX 4080 im Spieletest mit zahlreichen Raytracing-Benchmarks in vier Auflösungen.

www.pcgameshardware.de

All in all, the 4080 is up to 35% faster than 3090Ti in heavy RT titles, on average it's ~25% faster though than 3090Ti, and ~35% faster than 3090, while being 2X faster than 6950XT. The relevancy of this is that the 4080 will easily best the upcoming 7900XTX in heavy ray tracing, by a margin that is anywhere from 20% to 40%.


Edit: corrected summary percentages ..

 

[Subtlesnake]

According to the summary graphs, the 3080 is 28% faster than the 6800 XT on average. If the 7900 XTX is 50% faster than the 6950 XT, the 4080 will be 18% faster on average, for a 20% higher RRP.

 

[Putas]

Cyberpunk 2077:
6950XT: 100%
3090Ti: 178%
4080: 225%
4090: 322%

The 4080 is 25% faster than 3090Ti and 40% faster than 3090 and 225% faster than 6950XT.

It is pretty spicy to see someone accusing others of irrational vendor hate to correctly state differences between GeForces but then add 100 % to their advantage over the Radeon.

 

[Kaotik]

According to the summary graphs, the 3080 is 28% faster than the 6800 XT on average. If the 7900 XTX is 50% faster than the 6950 XT, the 4080 will be 18% faster on average, for a 20% higher RRP.

Official number is "up to 1.8x in heavy RT loads" (in their examples there's one with 2x performance without FSR, though (and ~1.85x with FSR))

 

[Henry swagger]

I just love how ray tracing scales beautifully with teraflops compute power..

 

[Deleted member 2197]

PCGamesHardware posted some new RT benchmarks covering recent hardware releases. Benchmarked at 4K max settings.

Cyberpunk 2077:
6950XT: 100%
3090Ti: 178%
4080: 225%
4090: 322%

The 4080 is 25% faster than 3090Ti and 40% faster than 3090 and 2.25x times faster than 6950XT.

Doom Eternal:
6950XT: 100%
3090Ti: 166%
4080: 200%
4090: 275%

The 4080 is 20% faster than 3090Ti and 30% faster than 3090 and 2x times faster than 6950XT.

Dying Light 2:
6950XT: 100%
3090Ti: 186%
4080: 235%
4090: 340%

The 4080 is 25% faster than 3090Ti and 40% faster than 3090 and 2.35x times faster than 6950XT.

Guardians of the Galaxy:
6950XT: 100%
3090Ti: 168%
4080: 190%
4090: 245%

The 4080 is 13% faster than 3090Ti and 22% faster than 3090 and 90% faster than 6950XT.

Lego:
6950XT: 100%
3090Ti: 164%
4080: 220%
4090: 310%

The 4080 is 35% faster than 3090Ti and 45% faster than 3090 and 2.2x times faster than 6950XT.

Metro Exodus:
6950XT: 100%
3090Ti: 155%
4080: 180%
4090: 250%

The 4080 is 15% faster than 3090Ti and 25% faster than 3090 and 80% faster than 6950XT.

Riftbreaker:
6950XT: 100%
3090Ti: 158%
4080: 196%
4090: 250%

The 4080 is 25% faster than 3090Ti and 35% faster than 3090 and 96% faster than 6950XT.

Ghostwire Tokyo:
6950XT: 100%
3090Ti: 146%
4080: 188%
4090: 255%

The 4080 is 30% faster than 3090Ti and 40% faster than 3090 and 88% faster than 6950XT.

Nvidia Geforce RTX 4080 16GB im Test: Ein fast perfektes Gesamtpaket - wenn der Preis nicht wäre [Update]

Nvidias Geforce RTX 4080 im Spieletest mit zahlreichen Raytracing-Benchmarks in vier Auflösungen.

www.pcgameshardware.de

All in all, the 4080 is up to 35% faster than 3090Ti in heavy RT titles, on average it's ~25% faster though than 3090Ti, and ~35% faster than 3090, while being 2X faster than 6950XT. The relevancy of this is that the 4080 will easily best the upcoming 7900XTX in heavy ray tracing, by a margin that is anywhere from 20% to 40%.


Edit: corrected summary percentages ..

A bit unusual RT benchmark since the results reflect including FSR 1.0 Ultra Quality for the highest performing AMD GPU.

Knowing that the current Geforce generation has a significant raytracing performance advantage, we staked out the basic hurdles usingRadeon RX 6900 XT Liquid Coolingand activated FSR upscaling if possible.

 

[Kaotik]

A bit unusual RT benchmark since the results reflect including FSR 1.0 Ultra Quality for the highest performing AMD GPU.

It must be some kind of copy/pasta error from old article or something, the numbers clearly don't reflect 6900XT LC having FSR on anywhere (just compare it to 6900XT air, or just look at the numbers)

 

[Deleted member 2197]

It must be some kind of copy/pasta error from old article or something, the numbers clearly don't reflect 6900XT LC having FSR on anywhere (just compare it to 6900XT air, or just look at the numbers)

Possibly though they specifically pointed out the 6900XT LC as having FSR 1.0 applied and no others in the article. Or perhaps FSR 1.0 was applied to all results?

The validation of the current settings took place in spring on the basis of the fastest Radeon graphics card. Knowing that the current Geforce generation has a significant raytracing performance advantage, we staked out the basic hurdles usingRadeon RX 6900 XT Liquid Coolingand activated FSR upscaling if possible. In the following benchmark you will find all results determined so far.

 

[Subtlesnake]

It is pretty spicy to see someone accusing others of irrational vendor hate to correctly state differences between GeForces but then add 100 % to their advantage over the Radeon.

The percentages are directly taken from the article. They only look strange because the 6950 XT starts at 100%.

 

[DegustatoR]

According to the summary graphs, the 3080 is 28% faster than the 6800 XT on average. If the 7900 XTX is 50% faster than the 6950 XT, the 4080 will be 18% faster on average, for a 20% higher RRP.

These results should be taken with their context.
The only reason why 3080/10 there is "just" 28% faster than 6800XT is because of the result in FC6 with HD textures pack - which was pretty much engineered specifically to kill off any card without at least 12GBs of VRAM -- 6800XT there is 123% faster than 3080/10 with RT!
If we remove this result and re-do an average we'd get 38,2 fps for 6800XT and 54,3 for 3080/10 - which in percents is 100 vs 142 - so a +42% in favor of 3080.
This is a much more relevant number if we're talking about future GPUs performance - unless you're expecting some of them to also have less than 12GBs of VRAM of course.

I'm also wondering which RT implementation should serve as a better blueprint for the future here as this would likely affect the actual positioning of new GPUs too.
In addition to FC6 there are a couple of games there which RT implementation is what I'd call "lite" - F1 2021, Riftbreaker, Ghostwire. In these 3080's advantage tend to be in a 30-40% range.
But if we're looking at games with "heavy" usage of RT 3080's advantage grows and hits +60% in titles like CP2077 and DL2.
While it is impossible to predict which "type" of RT usage will be more widespread in the nearest couple of years I'd wager that the percentage of games with "heavy" RT usage will increase. This is already happening in case of F1 series for example where 2022 iteration is a lot more heavy on RT than the 2021 used in this benchmark.

Btw I've used the 4080 benchmark blow out to check how many games with RT are being used in benchmarking suites of the most popular websites.
The percentage is - surprisingly maybe? - between 60 and 70. As in 60-70% of games used in benchmarking by the likes of TPU, CB.de and HUB have RT support / options in them.
And yet none of them are including RT benchmarks results into their "overall averages".

 

[pjbliverpool]

In addition to FC6 there are a couple of games there which RT implementation is what I'd call "lite" - F1 2021, Riftbreaker, Ghostwire. In these 3080's advantage tend to be in a 30-40% range.
But if we're looking at games with "heavy" usage of RT 3080's advantage grows and hits +60% in titles like CP2077 and DL2.
While it is impossible to predict which "type" of RT usage will be more widespread in the nearest couple of years I'd wager that the percentage of games with "heavy" RT usage will increase. This is already happening in case of F1 series for example where 2022 iteration is a lot more heavy on RT than the 2021 used in this benchmark.

I think the important thing to note here is that it's usually the games with heavy RT implementations where you need the extra performance.

 

[DegustatoR]

I think the important thing to note here is that it's usually the games with heavy RT implementations where you need the extra performance.

Yes and no. These "lite RT" games aren't exactly flying in 4K either. F1 2021 from the benchmark for example while showing "only" +27% advantage for 3080 runs at 44/56 fps on the cards respectively which isn't much.

 

[troyan]

There arent any "lite RT" implementations outside of Resident Evil and Farcry 6.

I think we are at a point in which "RT and DLSS" is the new normal. Miles Morales will run with 160FPS on a 4080 with DLSS and FG: