-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
-
We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.
Without DLSS, Should devs really be chasing Ray Tracing on consoles?
- Thread starter Sems4arsenal
- Start date
I paid 250$ for my RTX 2060 card, i would say it was worth it after playing Metro, Quake 2 RTX and Control. And i have never seen lightning/reflections like with RT on.
God bless DLSS 2.0
This is circular. If you are using resources to improve performance, then you are freeing up resources. But I would bet that Microsoft definitely considered machine-learning for resolution upscaling when designing the new Xbox. I'd also bet Cerny and his team considered it as well.
DLSS 1.0 used all tensor cores and looked worse than regular upscaling. My point is its still "Deep learning super sampling" and didnt require any specialized hardware. We dont know exactly the changes between 1.9 and 2.0, how much could be achieved with or without tensor cores.
No. Horrible idea and it will hinder next gen until Ps5 pro and co that come with dedicated AI upscale.
The insane performance gain with all the bells and whistles is just unmatched. And UE5 will take care of dynamic lightning solutions without full RTX requirement as well.
(and if casuals can't tell the difference between 30 fps and 60 fps, they'll never see the difference between AI upscale and native)
The insane performance gain with all the bells and whistles is just unmatched. And UE5 will take care of dynamic lightning solutions without full RTX requirement as well.
(and if casuals can't tell the difference between 30 fps and 60 fps, they'll never see the difference between AI upscale and native)
Thank you for the explanations! I haven't see ray tracing in person yet, but they definitely made me question how the hell they were doing those reflections. It's going to be exciting what devs can do with it when it comes to next-gen.
What?
I do not think this is accurate? Any kind of RT will have a greater cost than vanilla GI lighting/reflections.
lol give me a break. The difference between the best screen-space reflections and ray-traced reflections is huge. It also cuts down on dev time because you don't have to strategically place cubemaps everywhere. We already saw a number of PS5 games running at native 4k with ray-traced reflections.
The improvements are also on the developer side. From DF videos it seems that the ability to place a light source and have it "work" as it would in the real world is a time savings for development. I am sure the real advantages won't be seen for a little while yet, but it is clear that ray tracing is the future of graphics.
It is accurate. There are several possibilities where using RT instead of rasterization would net the same performance or even be faster. There is no "vanilla" anything in modern rendering - even untextured polygons can be rendered differently these days.
People seem to not understand that RT isn't an "effect", it's a rendering method.
Already happening this gen with SSRs and dynamic lights, nothing new with RT here.
At least with the Battlefield V implementation, there was a distant limit to the reflections where everything just disappears, although I don't think that game was built from the ground up with raytracing in mind but something seemingly tacked on later.
I'd say we're already seeing some of the ways developers will be getting it to work:
- Don't run at native 4K (or 60FPS for that matter in some cases)
- Set thresholds on material properties or distances for when to use RT and when to fall back to a less-expensive method
- Don't do things like display reflections at full resolution
- Strategically accept a bit of noise in exchange for better performance
- Disable it on certain objects
- Avoid using it in non-optimal scenarios (accurate reflections on the River Thames in Watch Dogs for example would require somewhat diffuse reflections on a very large surface, which could be quite expensive for an open world game)
Probably more I can't think of or I'm not aware of. Ray Tracing (in this context) is early and the consoles aren't going to be magic, just have to make various compromises to get a result that works. If they manage a result that looks better and/or saves a bit of dev time then it probably was an overall success. It's not inherently unbearably slow, but we're a ways off from hardware that allows it to be casually thrown around everywhere.
- Don't run at native 4K (or 60FPS for that matter in some cases)
- Set thresholds on material properties or distances for when to use RT and when to fall back to a less-expensive method
- Don't do things like display reflections at full resolution
- Strategically accept a bit of noise in exchange for better performance
- Disable it on certain objects
- Avoid using it in non-optimal scenarios (accurate reflections on the River Thames in Watch Dogs for example would require somewhat diffuse reflections on a very large surface, which could be quite expensive for an open world game)
Probably more I can't think of or I'm not aware of. Ray Tracing (in this context) is early and the consoles aren't going to be magic, just have to make various compromises to get a result that works. If they manage a result that looks better and/or saves a bit of dev time then it probably was an overall success. It's not inherently unbearably slow, but we're a ways off from hardware that allows it to be casually thrown around everywhere.
Absolutely.
With a fixed hardware design to target, no reason you can't figure out how to spends the rays available to you to do things like fix glaring SSR artifacts, etc.
With a fixed hardware design to target, no reason you can't figure out how to spends the rays available to you to do things like fix glaring SSR artifacts, etc.
I'm sure 2 years from now there will be much cheaper software solution available and runs on all hardware. May not be the best but look at cryengine RT solution, pretty conviencing and runs on any hardware. I am surprised Sony didn't have a image scaling solution for ps5 at all seeing how good their other products has such great digital imaging capability and how muhh they talk about A. I and Ps4 pro was build with upscaling in mind. Don't want to say secret sauce but is there something more that they haven't talk about? They did listed machine learning in their earlier slide.
Raytracing is great for the games that support it. DLSS makes it worthwhile on the current generation of RTX cards. The real problem right now isn't the hardware, it's just the low number of titles that take advantage of proper raytracing as well as DLSS 2.0.
I mean Microsoft at least has been experimenting with DirectML. I am not sure how well it works performance wise to DLSS but I am guessing it will help performance. I think Ratchent and Clank looked amazing with the raytracing they showed in their gameplay demo and that game was running in native 4K.
Image of DirectML working via overlock3d.net
Link to article: https://www.overclock3d.net/news/so...on_game-changer_that_nobody_s_talking_about/1
Image of DirectML working via overlock3d.net
Link to article: https://www.overclock3d.net/news/so...on_game-changer_that_nobody_s_talking_about/1
DLSS 1.0 did not use the Tensor cores it used shaders.
And we do know the changes between DLSS 1.9 and 2.0.
for the Devs, it also means spending less time on lighting as, once you've decided the lighting model, the hardware does it for you. Thus, easy to integrate into editors and no need for spending owners manually baking lighting to make it look good enough.
this isn't even getting into things that can only be done with Raytraced lighting such as physically accurate reflections of off-screen geometry.
OP
OP
I really doubt any AAA will use path trace lighting anytime soon.
Massive performance cost.
this forum is actually pretty enhanced if you just imagine all the posters are just cats lounging around like in neko atsume
hmm I disagree. I think each AAA game will be strategic especially those with in-house engines. They will find which element of Ray-tracing makes the game pop best while hitting their performance target (this will be 30fps [not locked]) for many titles with a goal to have at least one title for each Ray-traced feature to ensure their engines are forward compatible for the day when it's possible to use all.
The DLSS2.0 process occurs in series with the rest of the rendering pipeline, not in parallel. The dedicated hardware doesnt allow you to simultaneously use other resources elsewhere. The GPU process the scene, the tensor cores do the upscaling work, then the GPU does its post work.
MS' solution will have the same work flow, but will a bit slower.
Last edited:
I mean Microsoft at least has been experimenting with DirectML. I am not sure how well it works performance wise to DLSS but I am guessing it will help performance. I think Ratchent and Clank looked amazing with the raytracing they showed in their gameplay demo and that game was running in native 4K.
Image of DirectML working via overlock3d.net
Link to article: https://www.overclock3d.net/news/so...on_game-changer_that_nobody_s_talking_about/1
DirectML would just be API that manufactuers can implment. It says image source is Nvidia, so most likely DLSS via DirectML.
Yup me too.
Give me faked/non-RT reflections, so I can have better frame rate, longer draw distance, better physics, better AI, more density, etc.
DLSS 1.0 used the tensor cores.
DLSS 1.9 was a single version of it that was a proto 2.0 running on normal ALUs without any machine learning at all, it was a hand tailored thing.
DLSS 2.0 uses tensor cores but is a different approach entirely to DLSS 1.0. It is also cheaper.
I hope both platforms are looking into it seriously. So far my main concern is that both consoles are being sold with barely ANY RT feature talk other than "we support RT." I only know sony is attemping it so far, hope MS shows if off at their 2nd event. I just don't think Nvidia alone with RTX and DLSS can push RT alone.
Like anything in real-time rendering, it's all about compromise to reach what appears to be a better visual output, modern day ray-tracing is no exception.
Ray-tracing as we see it in games is true ray-tracing like what we see in path-traced renderers like V-Ray, it's calculating lights hitting surfaces according to how it happens in the real world based off of physical values, but even my RTX 2080 Ti doesn't have the power to do this at an unlimited amount of samples or bounce rate.
Typically ray-traced reflections for a real-time renderer like Unreal Engine 4 would only do a ray tracing calculation for a pixel if the material of the surface had a roughness of something like 0.7 or higher, all pixels below that would fall over to screen space reflections or prebaked reflection probes. Another way to improve speed would be to test if the pixel reflection is present in the screen space reflection map already calculated, using the SSR result instead of calculating an RT reflection for that pixel.
Now let's presume that we need to do an RT calculation for that pixel, how many reflection bounces should we do if the surface we hit in the reflection also has a roughness value higher than 0.7? The more bounces we need to do, the more compute power we need.
Now it's also really computationally heavy to do this for every pixel on screen, so what if we only calculate a limited amount of RT pixels every frame and temporally reconstruct them over 3-5 frames to make it even faster at the cost of ghosting artifacts if the viewport changes significantly?
However developers chose to implement RT features, it will always be about compromising until they reach the best performance to visual output ratio, upscaling algorithms like DLSS is just another way of compromising, RT can be implemented in many many ways.
Ray-tracing as we see it in games is true ray-tracing like what we see in path-traced renderers like V-Ray, it's calculating lights hitting surfaces according to how it happens in the real world based off of physical values, but even my RTX 2080 Ti doesn't have the power to do this at an unlimited amount of samples or bounce rate.
Typically ray-traced reflections for a real-time renderer like Unreal Engine 4 would only do a ray tracing calculation for a pixel if the material of the surface had a roughness of something like 0.7 or higher, all pixels below that would fall over to screen space reflections or prebaked reflection probes. Another way to improve speed would be to test if the pixel reflection is present in the screen space reflection map already calculated, using the SSR result instead of calculating an RT reflection for that pixel.
Now let's presume that we need to do an RT calculation for that pixel, how many reflection bounces should we do if the surface we hit in the reflection also has a roughness value higher than 0.7? The more bounces we need to do, the more compute power we need.
Now it's also really computationally heavy to do this for every pixel on screen, so what if we only calculate a limited amount of RT pixels every frame and temporally reconstruct them over 3-5 frames to make it even faster at the cost of ghosting artifacts if the viewport changes significantly?
However developers chose to implement RT features, it will always be about compromising until they reach the best performance to visual output ratio, upscaling algorithms like DLSS is just another way of compromising, RT can be implemented in many many ways.
You are right although the article states it is from "This talk, which is entitled "Deep Learning for Real-Time Rendering: Accelerating GPU Inferencing with DirectML and DirectX 12" showcases Nvidia hardware upscaling Playground Games' Forza Horizon 3 from 1080p to 4K using DirectML in real-time. " And it doesn't directly mention DLSS.
Last edited:
You realize that BVH is a way to skip brute forcing RT, right?
I'd say even something like 2.0 would be possible. MS has INT8/16 support so you could probably simulate Tensor cores (slowly) using INT16? I think a something that is similar in quality/perfrmance to DLSS2.0 will be possible in the next few yeras.
It's *possible* but not necessarily useful. The thing about Tensor Cores is they do a full 4x4 matrix FMA (d = a * b + c) in one go. The Pascal implementation needs to do it step by step which requires 12x as much math to be performed because they can only operate on one row at a time.
RTX Voice for instance, seems like a simple task for ML. On my Pascal 1080 Ti it takes up like 10% of the card's compute power. Just to apply noise reduction to audio through ML.
I've said this numerous times I'm glad some of you weren't aroud for the late 90's or early part of the start of century.
AA would've gotten nowhere with this attitiude.
I say devs can do what they want if the framepacing above 30fps.
I don't care if the game looks like quake 1, rdr2, or tlou2 its the content that ultimately matters.
I want ray tracing after see so many decent global illimination hacks, minecraft rtx, quake 2 rtx, and metro I'd rather not go back to flat non dyanmic lighting to say the least. Devs will find a sweet spot.
AA would've gotten nowhere with this attitiude.
I say devs can do what they want if the framepacing above 30fps.
I don't care if the game looks like quake 1, rdr2, or tlou2 its the content that ultimately matters.
I want ray tracing after see so many decent global illimination hacks, minecraft rtx, quake 2 rtx, and metro I'd rather not go back to flat non dyanmic lighting to say the least. Devs will find a sweet spot.
RT reflections are not that demanding at all if done right, 1080 Ti can run RT BFV reflections at decent performance. It's RT GI like in Control, Quake 2, Minecraft & Exodus that is demanding as all fuck, we probably won't see that much this gen anyway, it's hard to fake good real time reflections but you can fake good lighting.
Some people around here have a really weird hangup with ray tracing. Yes it's extremely noticeable, yes it can use a lot of resources if you're doing crazy stuff, yes that will change as we continue through the generation. But there's more to it than full on path tracing and there's more to visuals than RAW POWER. Ray traced sound will be massive too and I'm sure Cerny's sound focus will show that as well.
Ding ding
Same thing happens every gen, people are a bit underwhelmed but year 3 is usually when things are in full swing. Of course there's still gems that come out before then.
That is something that every person will individually decide. In my case ray tracing is very worth it. What developers cannot do right now is not give the option to disable it.
GP104 only has one Vec2 FP16*2 per 128 FP32 ALUs. So while you can run them using INT16 or INT8, it's all promoted to FP32 and only works on FP32 pipes with no packed math speed up. The only reason GP104 even has a Vec2 FP16*2 ALU per SM is for running FP16 code which *must* be run natively and not emulated by promoting to FP32.
Turing on the other hand has Tensor Cores which do the work for them.
I don't think the current cost is worth it.
Devs can, of course, find a cheaper and more efficient way to do it. For example, Anti-aliasing was such a hindrance during the PS360 era, but now everything looks so freaking sharp with Temporal AA tecnhiques and it is way less demanding than MSAA, as far as I know.
Devs can, of course, find a cheaper and more efficient way to do it. For example, Anti-aliasing was such a hindrance during the PS360 era, but now everything looks so freaking sharp with Temporal AA tecnhiques and it is way less demanding than MSAA, as far as I know.
Right but I thought MS mentioned for RDNA2 they have INT16/8 packed?
Maybe on RDNA2 but OP was specifically talking about 10 series cards.
