Ahahaha GT7 only has RT during replays and menu garage

[Remij 4,670]

40 minutes ago, Voidler said:

If they were competent it would be. Eg Miles Morales

No, that's wrong.  Do you really want a long winded explanation why?  

[Voidler 1,675]

13 minutes ago, Remij said:

No, that's wrong.  Do you really want a long winded explanation why?  

Go for it

 

They are both cross gen games, one has an RT upgrade and the other doesn't 

[Remij 4,670]

Just now, Voidler said:

Go for it

 

They are both cross gen games, one has an RT upgrade and the other doesn't 

Ok it'll have to wait until I get a chance.. I'm working ATM lol

[Quad Damage 1,717]

what modes will halo infinite have ray tracing at launch?

[MalaXmaS 646]

4 hours ago, Voidler said:

If they were competent it would be. Eg Miles Morales

This is what happens when you get into a thread about a dev that you don't know shit about, avoidler.

I already know how ass backwards Polyphony Digital is, with the inclusion of features.

I know how they start game development, their priorities and how they handle criticism.

[Substatic 420]

6 hours ago, Quad Damage said:

what modes will halo infinite have ray tracing at launch?

 

The best FPS of all time mode. 

[Quad Damage 1,717]

14 minutes ago, Substatic said:

 

The best FPS of all time mode. 

 

so none?

[Remij 4,670]

10 hours ago, Voidler said:

Go for it

 

They are both cross gen games, one has an RT upgrade and the other doesn't 

 

Ok 

 

So basically to understand what is happening in the most basic form  For ray tracing, the GPU calls a shader to dispatch a ray, calculates if it hits any geometry, and depending on if it does or not, will return a result and shade a pixel.  It may also possibly pass through, or stop, as well as potentially spawn secondary bounce rays for lighting, or reflections in reflections.

So that's what it does... but to make this process efficient, they create something called a BVH, which is a Bounding Volume Hierarchy structure.  What that does is orders everything into a tree-like structure using bounding boxes (because squares are simple) which contain the objects and geometry in the game world.  That structure, allows the GPU to very easily know in a general sense where objects are.. and reduces the amount of non-essential rays being cast.

 

On top of that, objects in games can have very high polygon counts.. so if they were to cast a ray for every pixel and test if any of those rays intersect with a very high poly object inside of a bounding box, it dramatically increases the calculations required.  So developers usually create proxy meshes which are more simplified low poly versions of the high poly objects, and test the rays against those instead..  and they can then use that information to determine how the corresponding pixels should be shaded.

 

So, simplified tree-like structures and simplified object meshes allow the developers to use less rays, and essentially know where to cast them.  That's absolutely great... but there's some caveats... and this is the big one which makes the difference.  Creating, and more importantly, updating and storing that BVH can become very costly on the CPU very quickly if things get out of hand, so developers try to keep it as simple as possible.  The CPU is responsible for updating and storing the BVH every time objects inside the game world move around inside it and traverse it and/or change state.  So for a scene relatively static which doesn't change often, it requires less updating.. which means less stress on the CPU.

So applying that concept to a game like Spider-man...  The game essentially consists of a city grid.. conveniently mostly square-like in its structure.  City blocks are squarish, the buildings all over the place are also essentially rectangles.  This allows the team to divide the city up into boxes, and allows the boxes to essentially fit nice and tight to the geometry as possible.  A square building inside of a square box inside of a square city block.. which fits very closely to it.  Nice, tight and compact.  Exactly what you want.  That allows them to keep track of exactly where the building is in the BVH structure, and keeps the calculations of the rays cast on the building much simpler, reducing the amount of unnecessary rays cast into the scene which wouldn't hit anything.  The less rays cast the better.. and wasting processing power casting rays which wont hit anything isn't smart.

 

So Spider-man, with it's RT reflections, mostly has a very simple structure.  You have Spider-man, which is the main high poly object, as well as some low poly pedestrians, and a general reflection of the buildings surroundings.. which is essentially more buildings which are less detailed and simplified.  That makes it quite simple as far as the entire chain goes.  It's basically a best case scenario.  And that's not to talk lightly about Insomniacs results.. because even despite that, RT is very demanding.. and they have a world class implementation for sure.  It's very high resolution and looks great.  Their algorithms are top notch.. to be getting those results on hardware that really isn't THAT accelerated when it comes to RT.

 

However, in the case of Gran Turismo 7.. you don't have any obvious square like pattern of the objects to help simplify the structure for you.  The creation of the BVH becomes much more of an issue because now you have to make many more sub structures because the boxes are going to be smaller and smaller to fit around the different shapes of the objects.  Trees and other environmental details with lots of tiny protrusions may not work well without getting in way over your head and obliterating the CPU, depending on how it's done.  Then.. you have the cars.  Ok, a box around a car seems easy enough.. but when you consider how high poly count these cars are, and how you really wouldn't want the reflection of it to look extremely basic.. the amount of rays/poly intersect testing shoots up through the roof.  So now not only is building the BVH structure much more costly for the CPU, but ALSO the ray calculations for the GPU are most costly as well.  Consider that in Spider-man, they're doing mirror reflections only, and it's all on flat surfaces.. ie building windows.  In GT7, cars are round and curved.. and making those reflections work properly and look correct is much more challenging to begin with.

 

It very easily spirals out of control.. because then you take that car.... and multiply it by 16.. or more..   That's like having 16 Spider-man quality characters needing to be reflected.. and since those cars are all moving independently of each other, and reflecting in each other the CPU needs to update that structure constantly.. and you're also travelling extremely fast.  It's that CPU work which I believe is the limiting factor.  It's a TON more calculations in all honesty.

 

So I hope you can kinda see what I mean by that.  GT7 will have it in the garage/menus which is straight forward enough... a static environment and only a handful of cars on screen at most (I'm guessing)... and then also replay mode.  If you think of replay mode, and ask why they can do it there, but not gameplay.. is likely because in replay, there's no car AI running, as well as the fact that perhaps they will limit RT in replays to 30fps?  Maybe not, they might hit the full 60fps depending on other factors like resolution, or only limiting the RT to only the player car and the next closest car.  Something like that.  It sounds like they're still working on it though.. but for gameplay it's probably just a bit too much at this point for the PS5 and same with Forza Horizon on Series X.  But in that game's defense.. the environment is insane, and destructible, and massive..so not having it in gameplay isn't really surprising at this point.. which is why I was laughing when Jehurey was acting like it was pathetic that FH5 didn't have RT during gameplay when he believed that GT7 would.  I think GT7 could see a PS5 Pro update which enables it for gameplay in the future.  Seems logical there, and a nice upgrade path for them.
 

 

Edited September 17, 2021 by Remij

[Voidler 1,675]

1 hour ago, Remij said:

 

Ok 

 

So basically to understand what is happening in the most basic form  For ray tracing, the GPU calls a shader to dispatch a ray, calculates if it hits any geometry, and depending on if it does or not, will return a result and shade a pixel.  It may also possibly pass through, or stop, as well as potentially spawn secondary bounce rays for lighting, or reflections in reflections.

So that's what it does... but to make this process efficient, they create something called a BVH, which is a Bounding Volume Hierarchy structure.  What that does is orders everything into a tree-like structure using bounding boxes (because squares are simple) which contain the objects and geometry in the game world.  That structure, allows the GPU to very easily know in a general sense where objects are.. and reduces the amount of non-essential rays being cast.

 

On top of that, objects in games can have very high polygon counts.. so if they were to cast a ray for every pixel and test if any of those rays intersect with a very high poly object inside of a bounding box, it dramatically increases the calculations required.  So developers usually create proxy meshes which are more simplified low poly versions of the high poly objects, and test the rays against those instead..  and they can then use that information to determine how the corresponding pixels should be shaded.

 

So, simplified tree-like structures and simplified object meshes allow the developers to use less rays, and essentially know where to cast them.  That's absolutely great... but there's some caveats... and this is the big one which makes the difference.  Creating, and more importantly, updating and storing that BVH can become very costly on the CPU very quickly if things get out of hand, so developers try to keep it as simple as possible.  The CPU is responsible for updating and storing the BVH every time objects inside the game world move around inside it and traverse it and/or change state.  So for a scene relatively static which doesn't change often, it requires less updating.. which means less stress on the CPU.

So applying that concept to a game like Spider-man...  The game essentially consists of a city grid.. conveniently mostly square-like in its structure.  City blocks are squarish, the buildings all over the place are also essentially rectangles.  This allows the team to divide the city up into boxes, and allows the boxes to essentially fit nice and tight to the geometry as possible.  A square building inside of a square box inside of a square city block.. which fits very closely to it.  Nice, tight and compact.  Exactly what you want.  That allows them to keep track of exactly where the building is in the BVH structure, and keeps the calculations of the rays cast on the building much simpler, reducing the amount of unnecessary rays cast into the scene which wouldn't hit anything.  The less rays cast the better.. and wasting processing power casting rays which wont hit anything isn't smart.

 

So Spider-man, with it's RT reflections, mostly has a very simple structure.  You have Spider-man, which is the main high poly object, as well as some low poly pedestrians, and a general reflection of the buildings surroundings.. which is essentially more buildings which are less detailed and simplified.  That makes it quite simple as far as the entire chain goes.  It's basically a best case scenario.  And that's not to talk lightly about Insomniacs results.. because even despite that, RT is very demanding.. and they have a world class implementation for sure.  It's very high resolution and looks great.  Their algorithms are top notch.. to be getting those results on hardware that really isn't THAT accelerated when it comes to RT.

 

However, in the case of Gran Turismo 7.. you don't have any obvious square like pattern of the objects to help simplify the structure for you.  The creation of the BVH becomes much more of an issue because now you have to make many more sub structures because the boxes are going to be smaller and smaller to fit around the different shapes of the objects.  Trees and other environmental details with lots of tiny protrusions may not work well without getting in way over your head and obliterating the CPU, depending on how it's done.  Then.. you have the cars.  Ok, a box around a car seems easy enough.. but when you consider how high poly count these cars are, and how you really wouldn't want the reflection of it to look extremely basic.. the amount of rays/poly intersect testing shoots up through the roof.  So now not only is building the BVH structure much more costly for the CPU, but ALSO the ray calculations for the GPU are most costly as well.  Consider that in Spider-man, they're doing mirror reflections only, and it's all on flat surfaces.. ie building windows.  In GT7, cars are round and curved.. and making those reflections work properly and look correct is much more challenging to begin with.

 

It very easily spirals out of control.. because then you take that car.... and multiply it by 16.. or more..   That's like having 16 Spider-man quality characters needing to be reflected.. and since those cars are all moving independently of each other, and reflecting in each other the CPU needs to update that structure constantly.. and you're also travelling extremely fast.  It's that CPU work which I believe is the limiting factor.  It's a TON more calculations in all honesty.

 

So I hope you can kinda see what I mean by that.  GT7 will have it in the garage/menus which is straight forward enough... a static environment and only a handful of cars on screen at most (I'm guessing)... and then also replay mode.  If you think of replay mode, and ask why they can do it there, but not gameplay.. is likely because in replay, there's no car AI running, as well as the fact that perhaps they will limit RT in replays to 30fps?  Maybe not, they might hit the full 60fps depending on other factors like resolution, or only limiting the RT to only the player car and the next closest car.  Something like that.  It sounds like they're still working on it though.. but for gameplay it's probably just a bit too much at this point for the PS5 and same with Forza Horizon on Series X.  But in that game's defense.. the environment is insane, and destructible, and massive..so not having it in gameplay isn't really surprising at this point.. which is why I was laughing when Jehurey was acting like it was pathetic that FH5 didn't have RT during gameplay when he believed that GT7 would.  I think GT7 could see a PS5 Pro update which enables it for gameplay in the future.  Seems logical there, and a nice upgrade path for them.
 

 

Yeah but car reflections are usually oblique and distorted to begin with and so I doubt many would notice if they are reflecting much lower poly meshes. Same with the environment, I feel like they could get away with a lot more on car reflections. They don't even really need ray tracing on the surrounding cars, but finding a solution to it on the car you are in should be there. And finding a balance between SSR and ray tracing seems to be the best compromise. Obviously you probably don't want ray tracing on a rainy track for example

 

I know where you're coming from but I say they are incompetent because they are unable to find a satisfactory compromise. If it were Insomniac doing the tech you know they would get it right. Even the Dirt 5 devs managed to get RT in their game

 

And it just reflects the overall frustration everyone has with Polyphony. Its the car damage models all over again, and their old argument that there is no real damage model because they can't do it perfectly. Or the old 1000 cars for GT4 and 60% of them ended up being PS1 models. They are kings of over promising under delivering

 

Maybe they are still working on some ancient engine which wouldn't surprise me. You never hear of Polyphony when it comes to the Sony first party collaborations. Kojima was able to use Guerrilla's engine for example. Imagine how much that could help their last gen looking environments. Or even using Unreal Engine.  Imagine how much they could benefit the car models using nanite technology 

 

They should have named Driveclub as Gran Turismo Driveclub and I bet that franchise would have over taken the main one by now like Forza Horizon did to the original

 

 

 

 

[JonDnD 2,607]

This gen consoles arent powerful enough for RT. RT will be a PC thing if you have a top card. Once NG kicks off for real dont expect much RT if you want 60fps on console 

[DynamiteCop 2,085]

17 minutes ago, JonbX said:

This gen consoles arent powerful enough for RT. RT will be a PC thing if you have a top card. Once NG kicks off for real dont expect much RT if you want 60fps on console 

The Ray tracing capability of these consoles is like that of a 2060 super. It's there and it can be done for certain, but it will only really work overtime as a nuanced feature. It's not something you can push.

[Remij 4,670]

1 hour ago, Voidler said:

Yeah but car reflections are usually oblique and distorted to begin with and so I doubt many would notice if they are reflecting much lower poly meshes. Same with the environment, I feel like they could get away with a lot more on car reflections. They don't even really need ray tracing on the surrounding cars, but finding a solution to it on the car you are in should be there. And finding a balance between SSR and ray tracing seems to be the best compromise. Obviously you probably don't want ray tracing on a rainy track for example

 

I know where you're coming from but I say they are incompetent because they are unable to find a satisfactory compromise. If it were Insomniac doing the tech you know they would get it right. Even the Dirt 5 devs managed to get RT in their game

 

And it just reflects the overall frustration everyone has with Polyphony. Its the car damage models all over again, and their old argument that there is no real damage model because they can't do it perfectly. Or the old 1000 cars for GT4 and 60% of them ended up being PS1 models. They are kings of over promising under delivering

 

Maybe they are still working on some ancient engine which wouldn't surprise me. You never hear of Polyphony when it comes to the Sony first party collaborations. Kojima was able to use Guerrilla's engine for example. Imagine how much that could help their last gen looking environments. Or even using Unreal Engine.  Imagine how much they could benefit the car models using nanite technology 

 

They should have named Driveclub as Gran Turismo Driveclub and I bet that franchise would have over taken the main one by now like Forza Horizon did to the original

 

 

 

 

Like I say though, it's mostly the constantly updating of the BVH structure which kills the CPU that's likely causing them grief at this point.  Forget even about the fact that reflections are some of the more simple calculations.. it's the traversal of the BVH which isn't accelerated on RDNA2 GPUs as it is on Nvidia's RT cores.  The RDNA2 RT hardware accelerates ray/triangle intersect calculations, but uses the shader cores to do BVH traversal.  What that offers is added flexibility over Nvidia's solution because it's more programmable... but it's also a lot slower than Nvidia's fixed function units.

 

So there's trade offs there.

 

I'm sure they'll get around to it eventually.  It's very well a mix of a lot of things.  The engine is becoming dated compared to others out there, and as already stated PD take forever and move at a snails pace.  And it's also very likely the fact that Yamauchi wants to meet a certain quality with it... something the team likely can't quite hit at this point at the framerate they want to ship at.

 

We'll see though.