GPU Ray Tracing Performance Comparisons [2021-2022]

[Silent_Buddha]

Metro Exodus's real time RTGI wasn't compelling for him but suddenly became so in EE? Why I wonder?

To be fair Metro: Exodus's original RT implementation was absolutely horrible. Mixing RT lighting with traditional rasterization tricks made the whole game look far worse than if they had just stuck with non-RT lighting throught. It's incredibly jarring in game when the two are mixed and you constantly transition between RT and non-RT lighting. It basically ends up looking like crap and is, IMO, the worst RT showcase of the first RT implementations. Uniformity and consistency in features is far more important unless it's done well and almost seemlessly when viewed. That wasn't the case in Metro Exodus. The game would have looked significantly better, IMO, if it was purely non-RT lighting and then they used the RT for something else instead of that aborted attempt to mix RT with non-RT lighting. Especially when their non-RT lighting was so subpar compared to other games on the market that did a much better job with non-RT lighting.

Metro: Exodus EE fixed that with finally making all lighting RT based. While there were still some drawbacks due to this, the lighting looks significantly better and the whole looks far more pleasing than the frankenstein that was Metro: Exodus prior to EE. So, while I agree that Control was a superb implementation of RT, Metro: Exodus was horrible until EE fixed it.

Even had I had a GPU that could run Metro: Exodus at 120+ FPS with RT on at the time, I would have instantly turned off RT just because of how bad it made the game look due to the inconsistent lighting. OTOH - I would have gladly left it on in Control and Metro: Exodus EE (although you obviously can't disable it in EE) if I had hardware that could run it at an acceptable to me framerate and resolution.

So, I fully get where he's coming from WRT Metro: Exodus and Metro: Exodus EE.

Regards,
SB

 

[Davros]

And I honestly fail to see how exactly he has arrived to that conclusion.

I think he means he doesnt recommend anyone buy a gfx card at the moment if you have to have a gpu to game on and want to actually be able to buy one without paying way more than the card is worth this is your best option to tide you over until cards stop selling for silly money (even though the 6600 is still overpriced, it's less overpriced than other cards)
eg: the cheapest nv 30 series card (aka the budget option) is close to £600
anyway that was my take

 

[troyan]

To be fair Metro: Exodus's original RT implementation was absolutely horrible. Mixing RT lighting with traditional rasterization tricks made the whole game look far worse than if they had just stuck with non-RT lighting throught.

They didnt mix anything. The RTGI replaced the voxel GI in Metro.

The game would have looked significantly better, IMO, if it was purely non-RT lighting and then they used the RT for something else instead of that aborted attempt to mix RT with non-RT lighting. Especially when their non-RT lighting was so subpar compared to other games on the market that did a much better job with non-RT lighting.

4A Games uses state of the art voxel GI. Metro has a full day and night cycle btw...

Even had I had a GPU that could run Metro: Exodus at 120+ FPS with RT on at the time, I would have instantly turned off RT just because of how bad it made the game look due to the inconsistent lighting. OTOH - I would have gladly left it on in Control and Metro: Exodus EE (although you obviously can't disable it in EE) if I had hardware that could run it at an acceptable to me framerate and resolution.

The voxel GI is worse and it isnt even near the quality of the RTGI. Here is an example from the game: https://imgsli.com/NTk1OTY
The difference between both GIs implementation is huge and RTGI is more consistent.

 

[Davros]

Metro has a full day and night cycle btw..

ooh, is it in real time, and is the time of day linked to the time of day in the real world?

 

[troyan]

Didnt say it is in real time.

 

[Dictator]

Far Cry 6 has visibly lower resolution RT reflections on AMD GPUs compared to NVIDIA GPUs.

https://imgur.com/IKnZBJR

https://imgur.com/uGSaZ5P

https://imgur.com/a/ukGxJLK

What the hell?! Now I have to check this out, holy kaboodle. I never even thought to do a test like this as settings between vendors, by standard, are always the same!!!

edit: can someone rehost at another site? Those images do not load at all.

 

[Deleted member 2197]

What the hell?! Now I have to check this out, holy kaboodle. I never even thought to do a test like this as settings between vendors, by standard, are always the same!!!
edit: can someone rehost at another site? Those images do not load at all.

Far Cry 6 DXR reflections resolution: AMD vs NVIDIA : Amd (reddit.com)
Hope you do a segment on this if you find any differences. The original video they is using RT + FSR for both vendors, but the difference is also obvious from later screenshots taken by people.

Edit: It would be hilarious if runtime vendor checks reduced RT reflections for performance gains!

 

[PSman1700]

I must admit I did find Hardware Unboxed conclusion on the RX6600 funny

Yes but those (RX6600/XT) are low (er) end gpu's around the 10TF in performance, someone opting for a gpu in that class shouldnt expect so much more.

 

[iroboto]

Far Cry 6 DXR reflections resolution: AMD vs NVIDIA : Amd (reddit.com)
Hope you do a segment on this if you find any differences. The original video they is using RT + FSR for both vendors, but the difference is also obvious from later screenshots taken by people.

Edit: It would be hilarious if runtime vendor checks reduced RT reflections for performance gains!

indeed, quite noticeable even with blurry youtube. There wasn't a time where I could notice that the two were equivalent. Should be investigated further, curious to see what Alex digs up.

 

[Davros]

Depends how you look at it, If you see it as a £400+ card then maybe you should expect more

 

[iroboto]

Depends how you look at it, If you see it as a £400+ card then maybe you should expect more

I think what needs to be investigated is whether it can be replicated properly. The results aren't confirmed yet. There may well be situations discovered that settings won't change unless you restart the game etc. Youtubers/home hobbyists aren't looking for these types of gotchas that journalists are looking for.

Even then, if it's true and they are both using ultra settings, for instance but the game engine is variably decreasing the RT load by reading a vendor card, that's not the norm. It pretty much means, AMD users will never see ultra. I know it's not the same as 'pay more expect more'. But you don't want to be a in a situation where you get say a 7800XT which may have very powerful RT and suddenly be vendor locked to this situation.

 

[Deleted member 13524]

You asked for an example of a game with a great RT implementation that runs well on accessible hardware and I provided one.

We'll disagree about what a "great implementation" is.
Control is IMO a pretty standard 8th-gen looking game with prettier reflections.It's light-years away from Demon's Souls 2020 which BTW has no raytracing.
I actually played Control at around the same time I played Division 2 and I found the later to be better looking in similar interior zones. The Snowdrop engine is a beast.
I get that the RT implementation in Control is a nice academic exercise for some, but it's not an especially pretty or more photoreallistic game. It's a nice curiosity to being able to zoom into the main character's eye and see the whole room around it, but it's a rather useless capability for actual gameplay.

Minecraft and Quake II RTX are great implementations of raytracing, you really feel like that's a level of lighting we can't find anywhere else. It's just that those games have the geometry and texture detail of >20 year-old games and people aren't willing to regress that much just for the prettier lighting and reflections. I guess Lego is a bit better in that regard, but it's got the scope of a Petri dish.

People are just used to it because they've never seen the alternative. And I don't blame them, they've never seen what they've never seen.

I see this "people are just ignorant because they haven't seen it yet" argument a lot, but side-by-side blind tests show that this is not the case:

Check timestamp or ~11m30s.

Is this meant to be a joke?

https://images.nvidia.com/geforce-c...ons-interactive-comparison-001-on-vs-off.html
https://images.nvidia.com/geforce-c...e-comparison-002-all-rt-on-vs-all-rt-off.html

No joke.. Control isn't a great looking game. What you're linking to us is Control with maxed out RT reflections showing a perfect reflection of a non great looking game.
Control is also the game where characters look like this:

The only real disagreement here is where exactly we are on the timeline of adoption of RT as the standard light transport mechanism.

Exactly. No one is arguing the "if", only the "when".

 

[iroboto]

I see this "people are just ignorant because they haven't seen it yet" argument a lot, but side-by-side blind tests show that this is not the case:

lol yea. I totally believe the result there, it's not surprising to be honest. You'd need some really dramatically dynamic movie lighting sequences for people to notice it. (like fighting with swinging lights above your head) Which I assume, we are due to see in some time in the future. People's focus is elsewhere in games, usually it's pretty zero'd in on playing the game. Eventually you're just going to forget any of it unless lighting become part of the gameplay. The issue can be further exasperated when you play outdoors, the likelihood they (heck anyone but Alex) notice any RT is probably zero. You almost have to be looking for it for people to notice, and even then when you do find a scene with dramatic difference, people are not likely to notice. Without a side by side comparison, I have strong doubts many people will know whether RT is on or not especially when outdoors.

It's not that people are ignorant, but there's enough in traditional baked lighting to sufficiently suspend disbelief that you don't care if you only want to succeed in the game. And when it comes to actually being ignorant or not, I think we all are in some ways. We haven't really seen an AAA RT game in which dynamic lighting is a major part of the gameplay. Not yet, we haven't seen it yet. I'm curious to see what that will be like. Minecraft RT is closest, but only if you play with light puzzles in the dark. But yea, outside of that, you've really got to take some massive advantages of real time shadows and lighting. Clouds moving, fog, etc.

 

[Putas]

You know I actually prefer the look of the original

How toxic of you.
Now seriously, I wonder if us old farts prefer the original look because our memory keeps remind us how it was special back then. Maybe age is an factor in RT appreciation.

 

[troyan]

I see this "people are just ignorant because they haven't seen it yet" argument a lot, but side-by-side blind tests show that this is not the case:

Check timestamp or ~11m30s.

You have to be blind not to see the differences in Control, Wolfenstein or "Deliver us the Moon". But i guess using Tomb Raider in 2021 is okay...

 

[iroboto]

You have to be blind not to see the differences in Control, Wolfenstein or "Deliver us the Moon". But i guess using Tomb Raider in 2021 is okay...

If you play them side by side, it's fairly easy to tell; if you're playing them one after the other, that's a real challenge. The issue with games is that you don't necessarily know whether or not the developer intended or not to render something. You need an actual comparison to show you where you're at a deficit.

It's good at least wrt that video, they tested all scenarios, and caught people falling into placebo effect. It's not a very large sample size (obviously 30 people would be ideal to really gain insight about a population), but it's reasonable given the type of channel they are.

 

[troyan]

In "Deliver us the Moon" the character and enviroment gets reflected in the every window. That has a huge visual impact and makes this game more immersive. You cannot miss it in the space station. Oh and there arent any SSRs with Raytracing so all these SSR artefacts are not disturbing...

 

[DegustatoR]

If you play them side by side, it's fairly easy to tell; if you're playing them one after the other, that's a real challenge.

Not really. RT reflections are the easiest RT use case to spot no matter how you're playing.
Shadows are next.
The most difficult is GI actually since the difference despite being rather major is of "well, that looks like proper lighting" vs "well, that looks like any game I've been playing for the last 10 years" variety.

 

[iroboto]

Not really. RT reflections are the easiest RT use case to spot no matter how you're playing.
Shadows are next.
The most difficult is GI actually since the difference despite being rather major is of "well, that looks like proper lighting" vs "well, that looks like any game I've been playing for the last 10 years" variety.

yep, I agree. So it really comes down to which scenes you're showing a viewer in a side by side comparison.

There are obvious ways we could bias a survey for sure. Just never show reflections or the opposite, only showing scenes with reflection etc.

 

[trinibwoy]

Control isn't a great looking game. What you're linking to us is Control with maxed out RT reflections showing a perfect reflection of a non great looking game.

You're bending over backwards to avoid a simple question. Does Control implement highly visible and valuable graphical effects that are not possible without raytracing? The answer is yes. Do games implement similar effects without RT? The answer is no. Your subjective opinion on the art style or gameplay of Demon Souls isn't relevant.