• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

State of Next-Gen Graphics So Far...

VFXVeteran

Banned
All,

I've spent the better part of 6 months playing some of the best games on the PC that has come out since the next-generation of consoles started. Here are some impressions that I've gathered about some of the games I've picked that covers pretty much every single tech that's out there including on the PS5/XSX exclusives.

1) Lighting

While we've come a long way with going towards RT for the foreseeable future, I'm still having a lot of complaints about lighting in general. We still see the old light probe GI solutions in most of the games that have come out since this generation has started. It's really really annoying to see that game companies are still satisfied with light probes for representing bounced light when objects are in shadow. I love the fact that games like Control, Metro, Crysis Remake and Cyberpunk has decided to enhance the experience by using RT for GI. However, Cyberpunk is suffering greatly with ambient occlusion. I'm still seeing the same old code from Witcher 3 in Cyberpunk and it's annoying:



The characters look horrible in the shade. Their clothes and skin literally look like constant ambient light has been put on them. CDPR wasted their resources on getting the RT ambient occlusion in for larger objects but completely abandoned casting rays on the characters clothes. Even their walking on the ground shows very light occlusion ultimately making the character look like they are floating on the surface.

2) Area Lights

Contrast that with what CDPR *did* get right and that's area lights. This is probably one of the most important forms of accurate rendering lights. The realworld has lights that are oddly shaped and not just spheres or arrows. Implementing area lights in a game is considered paramount to getting good local light sources.



3) NPCs, Fallbacks of SSAO

We are just now getting to the point where we can do full on subsurface scattering in gameplay and not worry about the massive hit in framerates. The hair shading still has a ways to go but adding a couple of specular terms in the hair, adding in the best PBR shading, higher res textures and you are closer to in-game cinematics - still aways off though. The other fallback is in SSAO. While it's great for still objects, it still rears it's ugly head. In this video, you can see it misses some items on the table and it disappears when an object is front of the view camera.



4) Baked lighting is still a good solution for now

Baked lightmaps and ambient occlusion is still a good solution for now when you set the scene up properly. You won't see the dynamic lighting but it will not suffer from the above mentions in games that want to utilize dynamic TOD, dynamic lights or dynamic objects in the scene. Demon Souls looks so good because they implemented this old tried and true system. There is an Indie game called Mortal Shell that is very similar to DS in both art direction and technology. Here is some gameplay that doesn't show any of the weaknesses in lighting that I was mentioning but it does use static objects, shadows, etc..



5) Better PBR without using RT

Microsoft's Flight Simulator has got me very excited for what's to come. They have hit all the nails on the heads in terms of graphics tech and it shows. This PBR shader usage is pretty much the best I've seen and I'd imagine that games like GT7 will look like this on their cars. If we could only get this quality for in-game characters.



6) Environment Lighting + FX

Again FS2020 is top dog here. I've studied several aspects of their lighting engine and I can't find any weaknesses at all. They are literally sampling the entire sky when they shade objects on the ground. I would love to read the paper on how they did this without ray-tracing. It reminds me of Lumen in the UE5 demo. They aren't using any tricks to do this like 99% of all games out there where the clouds aren't a part of the lighting equation or MIE scattering isn't being utilized.



7) Cloud FX Rendering

This has got to be a first for finally moving away from facing ratio sprite cards. FS2020's cloud rendering uses true 3D volume textures with literally no aliasing, light attenuation, and procedural noise shaders with implicit surfaces all deforming in realtime. The only fallback I see in this technique is there isn't enough layers of 3D texture. You can tell this when flying through them. This is obviously to limit bandwidth constraints. I would love to revisit 4-5yrs from now and see if we can adjust the resolution of these 3D volumes to maybe 64-128 z-levels to see how the GPU handles it.



8) Photogrammetry

We need more of this technique. It was done first in Star Wars battlefront I believe and those textures looked stellar. I know it's a lot of prep work to implement properly and would raise costs of the budget much higher, but you really can't beat true photography.



I'm excited for what's ahead but at the same time I'm very cautious. I don't expect all games to start having these things correct out the gate this generation. MFS2020 is clearly designed with future hardware in mind and RT is a bandwidth hog. We might see only a handful of games doing a lot of these features at once. But anything is better than nothing.
 

mansoor1980

Member
played a bit of COD BO cold war campaign on xsx and it is extremely impressive , the detail on the characters skin for example is great and at 60 fps , only nitpick i have with it is the hair textures , next gen graphics are only gonna get better for sure
 

Hobbygaming

has been asked to post in 'Grounded' mode.
I'm impressed with those games along with the visuals in these

Crimson Desert
Black Myth Wukong
Demon Souls remake
Guilty Gear Strive
Ratchet and Clank Rift Apart
Project Awakening
Kena
Returnal
Spider-man Miles Morales
 

Hendrick's

If only my penis was as big as my GamerScore!
Microsoft's Flight Simulator has got me very excited for what's to come. They have hit all the nails on the heads in terms of graphics tech and it shows. This PBR shader usage is pretty much the best I've seen and I'd imagine that games like GT7 will look like this on their cars. If we could only get this quality for in-game characters.
Don't you mean Forza? I mean, MSFS runs on the Forza engine.
 
Cheers to you, VFXVeteran VFXVeteran

CP77 (with RT Ultra) and Flight Sim 2020 on my 3080 are two of the best looking games I have ever played. And yes, I feel like both of these games will be used as main benchmarks for the generation and beyond.

Really curious to see how both fare on NG consoles (when CP77 NG update goes live for XsX and PS5, and when Flight Sim makes it to Xbox)
 
Last edited:

Omni_Manhatten

Neo Member
How many next gen engines exist right now? We know UE5 is coming. Most all engines are cross gen right now with at least 2 years before they move on. We even seen Jim Ryan and Phil Spencer advise they will support crossgen for at least a couple years. This may end up a long console gen. Both may even have reiterations at some point.
 

VFXVeteran

Banned
How many next gen engines exist right now? We know UE5 is coming. Most all engines are cross gen right now with at least 2 years before they move on. We even seen Jim Ryan and Phil Spencer advise they will support crossgen for at least a couple years. This may end up a long console gen. Both may even have reiterations at some point.
I wouldn't label a game engine as "next-gen" as most of them already have a lot of things to push visuals. Take Miles Morales for example. The RT and fast loading SSD was incorporated into the engine without having to overhaul the entire thing. UE5 is actually a derivative of the latest UE 4.2X.X. I've seen all the RT code in there ready to use even though there isn't a UE game using the RT yet.
 

OverHeat

« generous god »
jQ2PaV3.jpg

VRAM is already a problem on consoles...
 
I wouldn't label a game engine as "next-gen" as most of them already have a lot of things to push visuals. Take Miles Morales for example. The RT and fast loading SSD was incorporated into the engine without having to overhaul the entire thing. UE5 is actually a derivative of the latest UE 4.2X.X. I've seen all the RT code in there ready to use even though there isn't a UE game using the RT yet.
Not to spark a console war but I'm curious what you think of the PS5 vs XSX hardware. I've sort of had it in the back of my mind XSX will come into its own when RT and something akin to DLSS comes into play as more next gen games and dev experience arrives. What's your take on the RT side of the hardware differences?
 

VFXVeteran

Banned
Not to spark a console war but I'm curious what you think of the PS5 vs XSX hardware. I've sort of had it in the back of my mind XSX will come into its own when RT and something akin to DLSS comes into play as more next gen games and dev experience arrives. What's your take on the RT side of the hardware differences?

They are the same. AMD lagged on the tech to incorporate actual chipset hardware for A.I. and deep learning on their GPUs which is perfect for DLSS-like hardware acceleration. I'm not convinced their software solution will alleviate bandwidth constrained features. As a consequence, I fear neither console will be able to afford anything more than 1 RT affect with render targets already being constrained to sub-4k render targets.
 
They are the same. AMD lagged on the tech to incorporate actual chipset hardware for A.I. and deep learning on their GPUs which is perfect for DLSS-like hardware acceleration. I'm not convinced their software solution will alleviate bandwidth constrained features. As a consequence, I fear neither console will be able to afford anything more than 1 RT affect with render targets already being constrained to sub-4k render targets.
Interesting it's an inherent issue for both consoles. I'm a little surprised Sony/Xbox didn't push back for that sort of hardware to be included given the length of the last generation.
 

ZywyPL

Banned
Microsoft's Flight Simulator has got me very excited for what's to come. They have hit all the nails on the heads in terms of graphics tech and it shows. This PBR shader usage is pretty much the best I've seen and I'd imagine that games like GT7 will look like this on their cars. If we could only get this quality for in-game characters.

FS2020 in VR is the most realistic thing I have ever experienced, the combination of details, materials and the lightning really make you feel like you are there, in a plane, and like you said, all that without a slightest usage of RT
 
Last edited:

perkelson

Member
1st point is wrong. It is not ambient occlusion.

CB2077 uses real time GI that is changing every once in while so it is not realtime dynamic but dynamic which also means that sometimes it can "break" and you get those weird ambient lighting on stuff like characters. But that happens very rarely.

Moreover we are talking about games here. No game ever outside of some actually want real dark because it would make nights unplayable which means they have to have minimal amount of ambient lighting regardless of conditions.
 

TheGejsza

Member
Interesting it's an inherent issue for both consoles. I'm a little surprised Sony/Xbox didn't push back for that sort of hardware to be included given the length of the last generation.
The PRO/X revision of PS5/XSX will be RayTracing-focused ones I guess - 4k 60 FPS with Full RT. It would be pretty nice upgrade for enthustiasts and engines would be ready by the time PRO/X release.

On the other note I am glad we are hitting resolutions caps (I mean going beyond 4k is pointless) so the resources can be put into lightnint, textures and SFX once reaching stable 4k will be normal. And I have to admit althoug 4k vs 1440p it is obvious that 4k is superior - I am still not convinced to 4k. In still-shots you can clearly see the resolution diffrence especialy in foliage/trees etc. But most LCD/QLED/OLED panels are still having pretty significant smudge and in motion the diffrences between 4k and 1440p is becoming... insignificant. I spent a few hours comparing Demon's Soul's and SpiderMan: Miles Morales graphics modes and 4K is a bit of a meme outside of screenshots.

I also played a lot of GOT and this game is glorious in 4k (art-design) but you can clearly see that bumping resolution without increasing textures fidelity is not the way to go. Sony showed in Demon's Souls one of the best textures I've ever seen and can't wait to see what GE and Nanite will bring in the future.
 

VFXVeteran

Banned
1st point is wrong. It is not ambient occlusion.

Ambient occlusion is a part of the global illumination rendering equation. Look it up.

CB2077 uses real time GI that is changing every once in while so it is not realtime dynamic but dynamic which also means that sometimes it can "break" and you get those weird ambient lighting on stuff like characters. But that happens very rarely.
All the GI light probes updating isn't exactly real-time. We all know this. It's an assumption to make. There are several things in the rendering pipe that update several frames instead of 1x/frame (like animation of NPCs). It's harmless to call it "realtime".

Moreover we are talking about games here. No game ever outside of some actually want real dark because it would make nights unplayable which means they have to have minimal amount of ambient lighting regardless of conditions.
There are many games that do AO properly and it works. FS2020 is a perfect example of "getting it right" under any lighting conditions and they still use SSAO.
 
Last edited:

regawdless

Banned
The first and - for me - most important step is to get rid of weirdly glowing, wrongly lit objects. Lighting plays a huge factor and Cyberpunk with all that RT lighting looks incredible at times. But as OP has pointed out, the characters stick out.

Realistic global GI is the shit. But expensive and I don't know if the consoles will manage.
 

JimboJones

Member
Interesting it's an inherent issue for both consoles. I'm a little surprised Sony/Xbox didn't push back for that sort of hardware to be included given the length of the last generation.
A lot of people very sceptical of nvidia for dedicating hardware to tensor cores and incorporating Ray tracing when the 20xx series rolled out and it didn't help the first implementation of DLSS sucked.
It was only kind of recently DLSS 2.0 started showing what it could really achieve.
AMD got kind of blind sided in that regard.
 

perkelson

Member
Ambient occlusion is a part of the global illumination rendering equation. Look it up.

But the problem is not in ambient occlusion but in ambient lighting. Those are two different things. Which is why i posted further on that.

Simply put GI probe was not updated and characters were lit with old value or something like that. You can see it on consoles often where probe is much rarer which creates overlit characters from time to time.
 
Last edited:

Krappadizzle

Gold Member
VFXVeteran VFXVeteran Wasn't first Photogrammetry used in The Vanishing of Ethan Carter?

Was just about to say this. The game looked great as a result. The environments were stunning at release.
 

DeaconOfTheDank

Gold Member
Very informative post. It's clear that you did quite a bit of research and put a ton of effort into finding good examples.

VFXVeteran VFXVeteran I'm curious to hear about your thoughts on tessellation and Turing mesh shaders (AMD's equivalent seems to be primitive shaders). I was blown away by the use of tessellation in Demon's Souls - it seemed like a very scalable technique to bring out raw geometric detail while requiring very little from the artist (uneven brickwork is a good example of this). Plus, it actually perturbs the surface of a mesh rather that manipulating the lighting of surface like what you get with normal maps. Looking at Nvidia's Asteroids demo, I would think that tessellation and mesh shaders would go hand in hand.




Additionally, do you believe that temporal rendering techniques will see more heavy adoption by developers? Lumens showed very impressive GI that was achieved by building up GI data over many frames (hence, the "temporal" aspect). I believe that future implementation of temporal techniques will start to open up rendering techniques that were very expensive in the past, by amortizing the cost over many frames. While shifting the focus away from raw shader teraflops, I imagine that this will put quite a strain on VRAM usage.

I'm quoting myself here, but I want to hear your take on this:
The way I see it, Nvidia is investing in two different approaches to rendering:
  1. Heavy reliance on AI upscaling to get to native 4K/8K and more developers targeting lower resolutions (e.g.1440p) such that VRAM isn't a problem. Cyberpunk can be seen as an example of this because, frankly, running at native 4K without DLSS is a fool's errand. Many techniques also scale and are less expensive at lower resolutions, thus reducing number of shader teraflops needed and instead placing emphasis on tensor teraflops for upscaling. Additionally, by allowing for very realistic real-time lighting, less lighting is baked into materials and again brings down the amount of VRAM needed. This comes at the cost of placing emphasis on RT teraflops for intersection tests and the CPU for building BVHs. Nvidia can try to steer developers in this direction by artificially enforcing a VRAM limit (e.g. 10 GB) to where native 4K is very difficult to achieve while adopting modern graphics rendering techniques...
  2. Targeting native resolution which requires large amounts of VRAM, shader teraflops, and RT teraflops. Going down this path will probably see an emergence of many temporal techniques that build up data and amortize render cost over many frames. Unreal Engine 5 demo was a good example of this concerning expensive techniques like Global Illumination. By making the cards that support this approach very expensive (MSRP $1499 for RTX 3090), more developers are inclined to look at cheaper cards and hopefully head in the direction that Nvidia is best prepared for.
Finally, with mainstream adoption of SSDs and reliance on storage as an extra pool of pseudo-VRAM primarily for streaming, developers can minimize GPU VRAM usage to assets that are directly in the player's view (e.g. next 5 seconds of gameplay) rather than holding all assets that player might interact with in the near future (e.g. next 30 seconds of gameplay).

Sounds like they're just covering their bases.
 
Last edited:

mansoor1980

Member
while adding new rendering tech is nice developers should also look to get rid of certain flaws which are immersion breaking for example object/scenery/shadow pop in, if UE5 engine games eliminate or hide that well then it will be the GOD LIKE game engine this gen
 

VFXVeteran

Banned
But the problem is not in ambient occlusion but in ambient lighting. Those are two different things. Which is why i posted further on that.
There is no such thing as ambient lighting in the realworld. Since the games are using ambient lighting and not direct light sources, they have to compute a shadow somehow. That's where ambient occlusion comes in to give the object depth. There is no ambient occlusion being used on the individual characters. That is, there is no rays being shot out into the scene on the coat or shirt or shoes, for example, to test occlusion.

Simply put GI probe was not updated and characters were lit with old value or something like that. You can see it on consoles often where probe is much rarer which creates overlit characters from time to time.
It happens because the shaders aren't using AO on the individual characters. This is akin to trees not showing self-occlusion and appearing flat from the ambient lighting (which is basically a constant color). As an exercise, go into a room that's not hit by direct sunlight. Take your hand and spread your fingers out away from each other. Then slowly start to bring your fingers in close to each other. You will see them cast shadows on each other. That's ambient occlusion. Clothes in the scene of Cyberpunk needs to cast shadows on themselves and the body parts. People's belt should cast shadow on their shirt/pants. Their arm pits need to cast shadows underneath their arms. Eyeglasses need to cast shadows on their face, etc.. etc..
 
Last edited:

VFXVeteran

Banned
Very informative post. It's clear that you did quite a bit of research and put a ton of effort into finding good examples.

VFXVeteran VFXVeteran I'm curious to hear about your thoughts on tessellation and Turing mesh shaders (AMD's equivalent seems to be primitive shaders). I was blown away by the use of tessellation in Demon's Souls - it seemed like a very scalable technique to bring out raw geometric detail while requiring very little from the artist (uneven brickwork is a good example of this). Plus, it actually perturbs the surface of a mesh rather that manipulating the lighting of surface like what you get with normal maps. Looking at Nvidia's Asteroids demo, I would think that tessellation and mesh shaders would go hand in hand.
Polygon throughput never seems to be a problem these days. Consoles have been able to process large amounts of triangles with ease tbh. I think as long as we aren't trying to make handles on faucets in a bathroom round, we should be good with that. The terrain in FS2020 literally has no polygonal edges at all. We are starting to see more and more assets being completely sharp-edge free. And then add normal mapping on top of that to handle the micro details and you are looking at some really nice geometry. The problem has always been in the lighting/shading pipeline. While the geometry throughput is good enough, the shader throughput is lacking.
Additionally, do you believe that temporal rendering techniques will see more heavy adoption by developers? Lumens showed very impressive GI that was achieved by building up GI data over many frames (hence, the "temporal" aspect). I believe that future implementation of temporal techniques will start to open up rendering techniques that were very expensive in the past, by amortizing the cost over many frames. While shifting the focus away from raw shader teraflops, I imagine that this will put quite a strain on VRAM usage.

I'm quoting myself here, but I want to hear your take on this:
Yes I do. This is definitely worth investigating. Even the RT for certain rendering pipes like ambient occlusion and GI has a slight delay in it when it happens. You can even see this in the UE5 demo when the ceiling breaks and the lighting shows through and bounces around in the scene. As long as the latency isn't too bad, I think it's a great way to go forward. And you are right, native 4k is just too much - even for a 3090 right now. We'll have to pull back on the bandwidth constraints and go with amoritization techniques to make things look good... for now anyway.
 
Top Bottom