• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

PS5 AMD GPU will support RT

Soldato
Joined
6 Feb 2019
Posts
17,464
Indeed, AMD's hardware / shader approach is more flexible and requires far less dedicated hardware compared to Nvidias.

however does that mean that AMD is using shader cores that would otherwise do raster operations? And would that not mean that having Ray Tracing on takes away performance that would be used for higher resolution, higher other settings and higher fps etc.

With Nvidias method we know that at least in theory - if they added enough RT cores they could achieve Ray Tracing performance exactly the same as having RT off right. But AMDs approach always sacrifices raster performance to have RT on?

I have so many questions like this, can't wait till AMD unveils the tech
 
Man of Honour
Joined
13 Oct 2006
Posts
90,805
however does that mean that AMD is using shader cores that would otherwise do raster operations? And would that not mean that having Ray Tracing on takes away performance that would be used for higher resolution, higher other settings and higher fps etc.

With Nvidias method we know that at least in theory - if they added enough RT cores they could achieve Ray Tracing performance exactly the same as having RT off right. But AMDs approach always sacrifices raster performance to have RT on?

I have so many questions like this, can't wait till AMD unveils the tech

From what I can see it is possible to process a lot of your shader bulk along side ray tracing work - to some degree it will allow you to leverage idle shader time (and one advantage AMD has there is that it traditionally hard to fully utilise the full width of their shader pipeline partly due to a vision of async shader utilisation that hasn't really come to reality) but ultimately you will be taking away some performance from your raster and other operations and even with a flexible shader design and improvements in terms of memory storage and access for ray tracing results it won't compete with dedicated hardware as ray tracing like physics (and hence why there is talk of physics and sound, etc. making use of ray tracing hardware) is one area that really benefits from hardware specialisation for most efficient throughput.

Ultimately AMD seems to be angling at an easier to implement hybrid approach which has slightly better backwards compatibility performance wise but ultimately is as much a dead-end long term as traditional techniques.

EDIT: I suspect with AMD's approach a lot of reliance will be on building baked lighting using ray tracing acceleration with a small amount updated in real time where it would be most noticeable which might look good with less problems trying to fix noise artefacts, etc. but when you spend more time around an engine where every single reflection is fully dynamic, etc. you start to notice the difference - kind of like 60Hz v 120+Hz once you get used to it most people are less forgiving of 60Hz going back. Something like the Godot engine people were talking about https://www.gdcvault.com/play/1026182/
 
Last edited:
Soldato
Joined
20 Apr 2004
Posts
4,365
Location
Oxford
.

EDIT: I suspect with AMD's approach a lot of reliance will be on building baked lighting using ray tracing acceleration with a small amount updated in real time where it would be most noticeable which might look good with less problems trying to fix noise artefacts, etc. but when you spend more time around an engine where every single reflection is fully dynamic, etc. you start to notice the difference - kind of like 60Hz v 120+Hz once you get used to it most people are less forgiving of 60Hz going back. Something like the Godot engine people were talking about https://www.gdcvault.com/play/1026182/

Using Ray tracing to to build baked lighting ? thats a oxymoron, what would be the point in that. You have the same limitations of baked lighting but gobbling up GPU resources as instead of it being generated by a game edoit when the map is "built" the user GPU would do it.

You can fully reflect everything now, problem is with rasteration, baked lighting and SSR have to used for it to be displayed and only in the players view port and not everything can reflect on every surface.
 
Man of Honour
Joined
13 Oct 2006
Posts
90,805
Using Ray tracing to to build baked lighting ? thats a oxymoron, what would be the point in that. You have the same limitations of baked lighting but gobbling up GPU resources as instead of it being generated by a game edoit when the map is "built" the user GPU would do it.

Still a lot cheaper than real time ray tracing - by using acceleration you could update baked GI for instance to take into account changes to lighting conditions but it would still lack interaction with dynamic objects in the scene.
 
Soldato
Joined
20 Apr 2004
Posts
4,365
Location
Oxford
Most of the current RTX uses have been hybrid, bar I think Quake 2 RTX and mine craft, and RTX is based off DXR afterall.

Also don't get what you mean about using RT hardware to generate bake lighting, what would be the point. It already gets pre-build as part of map data

As for reflections, you are limited to the like of SSR and other Rasterization "techniques" Which limits what can be displayed before you even talk about scalability
Still a lot cheaper than real time ray tracing - by using acceleration you could update baked GI for instance to take into account changes to lighting conditions but it would still lack interaction with dynamic objects in the scene.

You basically talking about Light maps, quality would be unchanged as current light sources are used generate the data for the light map. You still have the same limitations you just be wasting GPU resources generating then rendering a static light map. The more time the wanted to generate a new light map the more resources that be wasted. That's why they are pre compiled and you don't have many different ones for a scene or object and why at most you get a limited set for time of day cycles.
 
Soldato
Joined
7 Feb 2015
Posts
2,864
Location
South West
From what I can see it is possible to process a lot of your shader bulk along side ray tracing work - to some degree it will allow you to leverage idle shader time (and one advantage AMD has there is that it traditionally hard to fully utilise the full width of their shader pipeline partly due to a vision of async shader utilisation that hasn't really come to reality) but ultimately you will be taking away some performance from your raster and other operations and even with a flexible shader design and improvements in terms of memory storage

The biggest issue with 3D workloads and CU utilisation with GCN is the 64bit wavefront and 4 cycle issue simd16 processing. Which is why RDNA went to single cycle issue simd32 which is much more in line with how Nvidia does things. It also suits 3D workloads better.
 
Man of Honour
Joined
13 Oct 2006
Posts
90,805
You basically talking about Light maps, quality would be unchanged as current light sources are used generate the data for the light map. You still have the same limitations you just be wasting GPU resources generating then rendering a static light map. The more time the wanted to generate a new light map the more resources that be wasted. That's why they are pre compiled and you don't have many different ones for a scene or object and why at most you get a limited set for time of day cycles.

I'm trying to illustrate a point indicatively rather than get into specifics games currently use a variety of techniques in both real time and design time to build shadows and indirect lighting, etc. a lot of which, other than direct shadows, are still purely static or use very approximate voxel/probe techniques even in today's engines while a hybrid system could generate new ones over a few frames in response to light changes (but wouldn't take into account dynamic scene objects) to give the impression of a more dynamic system without going to the full load of ray tracing.

Trying to be very general here as it is too easy to get bogged down in specifics and different engines doing things differently again like Cryengine, etc.
 
Soldato
Joined
20 Apr 2004
Posts
4,365
Location
Oxford
I'm trying to illustrate a point indicatively rather than get into specifics games currently use a variety of techniques in both real time and design time to build shadows and indirect lighting, etc. a lot of which, other than direct shadows, are still purely static or use very approximate voxel/probe techniques even in today's engines while a hybrid system could generate new ones over a few frames in response to light changes (but wouldn't take into account dynamic scene objects) to give the impression of a more dynamic system without going to the full load of ray tracing.

Trying to be very general here as it is too easy to get bogged down in specifics and different engines doing things differently again like Cryengine, etc.

sorry but you are changing the goal posts and making no sense whats so ever now,
 
Soldato
Joined
6 Feb 2019
Posts
17,464
Digital Foundry think that AMD's ray tracing approach will result in a technically lower quality image output than Nvidia's RT cores. However they can think of smart ways to hide the lower image quality - or maybe it doesn't even matter. Many people on these forums say they don't think the different between RT on and RT off to be big enough, so if you lower the quality of RT on, does it even matter? Maybe not - especially for console gamers

 
Permabanned
Joined
28 Nov 2006
Posts
5,750
Location
N Ireland
Digital Foundry think that AMD's ray tracing approach will result in a technically lower quality image output than Nvidia's RT cores. However they can think of smart ways to hide the lower image quality - or maybe it doesn't even matter. Many people on these forums say they don't think the different between RT on and RT off to be big enough, so if you lower the quality of RT on, does it even matter? Maybe not - especially for console gamers


That is why Nvidia pushed it out premature, So they can catch AMD out and claim a monopoly. Claims like that are also pure gold for Nvidia if this is true expect Nvidia to really milk it. They will already be pushing developers to add RT that heavily skews fps in the favour of Nvidia. I am not sure how they would do this but i am guessing for now on AAA Nvidia titles will have so much unnessecary reflective surfaces it will be crazy.
 
Soldato
Joined
7 Feb 2015
Posts
2,864
Location
South West
however does that mean that AMD is using shader cores that would otherwise do raster operations? And would that not mean that having Ray Tracing on takes away performance that would be used for higher resolution, higher other settings and higher fps etc.

With Nvidias method we know that at least in theory - if they added enough RT cores they could achieve Ray Tracing performance exactly the same as having RT off right. But AMDs approach always sacrifices raster performance to have RT on?

I have so many questions like this, can't wait till AMD unveils the tech

The RT cores only accelerate part of the ray casting process, it is a computationaly heavy chunk of it, but shaders still need to be running to launch and handle the ray casting in unison with the RT cores.

image20-1-1024x604.jpg
 
Man of Honour
Joined
13 Oct 2006
Posts
90,805
I am not sure how they would do this but i am guessing for now on AAA Nvidia titles will have so much unnessecary reflective surfaces it will be crazy.

The way to load it up would be for the game to either enforce lots of rays per pixel or to do a lot of bounce passes for complex radiosity, etc. with path tracing reflective surfaces are relatively tame performance wise once you've got some kind of ray tracing system working in the first place - it is one of the things that is a bit annoying with BF5 actually in that the hit from the reflection system is already a big slice of doing more than just reflections with ray tracing.

The RT cores only accelerate part of the ray casting process, it is a computationaly heavy chunk of it, but shaders still need to be running to launch and handle the ray casting in unison with the RT cores.

With Turing you can do quite a bit of your bulk shader work concurrent with the ray tracing process (kind of ironically in a similar manner to async compute/shaders).
 
Soldato
Joined
7 Feb 2015
Posts
2,864
Location
South West
With Turing you can do quite a bit of your bulk shader work concurrent with the ray tracing process (kind of ironically in a similar manner to async compute/shaders).

Indeed, the RT cores will perform BVH and Ray intersection checks concurrently with shaders doing other things, it isn't a blocking process unless a shader is waiting for the RT core.

But the shaders are still used with some parts of the ray calculations. So it's not like they are completely free from being used by RT.
 
Last edited:
Man of Honour
Joined
13 Oct 2006
Posts
90,805
Indeed, the RT cores will perform BVH and Ray intersection checks concurrently with shaders doing other things, it isn't a blocking process unless a shader is waiting for the RT core.

But the shaders are still used with some parts of the ray calculations. So it's not like they are completely free from being used by RT.

Tensors core are also involved for optimising BVH stuff - where things seem to fall down is if DLSS is in play where you end up waiting on the Tensor cores :s
 
Back
Top Bottom