• 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.

Sony files new patent for ‘Accelerated Ray Tracing’ - speculation that it could be more performant

cormack12

Gold Member
Source: https://www.psu.com/news/sony-files-new-patent-that-mentions-accelerated-ray-tracing-on-ps5/
Patnet: https://data.epo.org/publication-server/document?iDocId=6759906&iFormat=0


5.3
Publication of the European patent application
5.3.001
The European patent application is published as soon as possible after the expiry of eighteen months from the date of filing or the earliest priority date. You may however request that it be published earlier.
The publication contains the description, the claims and any drawings, all as filed, plus the abstract. If the European search report is available in time, it is annexed (A1 publication); if not, it is published separately (A3 publication). If the European patent application was not filed in English, French or German, its translation will be published.
All European patent applications, European search reports and European patent specifications are published in electronic form only, on the EPO's publication server. The publication server is accessible via the EPO website (www.epo.org).






 

Panajev2001a

GAF's Pleasant Genius
Source: https://www.psu.com/news/sony-files-new-patent-that-mentions-accelerated-ray-tracing-on-ps5/
Patnet: https://data.epo.org/publication-server/document?iDocId=6759906&iFormat=0


5.3
Publication of the European patent application
5.3.001
The European patent application is published as soon as possible after the expiry of eighteen months from the date of filing or the earliest priority date. You may however request that it be published earlier.
The publication contains the description, the claims and any drawings, all as filed, plus the abstract. If the European search report is available in time, it is annexed (A1 publication); if not, it is published separately (A3 publication). If the European patent application was not filed in English, French or German, its translation will be published.
All European patent applications, European search reports and European patent specifications are published in electronic form only, on the EPO's publication server. The publication server is accessible via the EPO website (www.epo.org).







Interesting, was not sure what was in the RT units inside/alongside the TMU’s on RDNA2 (doing more in HW than just ray intersection tests).
 
Can someone explain? What is it?
This is a software? (or driver)?
If this can be done on current hardware means it could run on any compatible hardware? (incl. xbox or nvidia?)
This is RDNA2 or even PS5 exclusive optimization?
I
 

hlm666

Member
reading patents really is a pain.

the drawings don't tell you anything other than the usual concept of high and low level bounding volume traversal and hardware data transfer in rdna2. claims are just in the text.
It looks like alot of nvidia patents about ray traversal, talking about using a dedicated core for RT traversal (nvidia patents are worded like bvh traversal on RT cores etc.) which AMD apparently doesn't have yet so i'm not sure where this is going. Is it a clue AMD's next line of gpus will have dedicated RT cores? and Sony expect to use these in a ps6 or pro down the line?

"A graphics processing unit (GPU) includes one
or more processor cores adapted to execute a soft-
ware-implemented shader program, and one or more
hardware-implemented ray tracing units (RTU) adapted
to traverse an acceleration structure to calculate inter-
sections of rays with bounding volumes and graphics
primitives asynchronously with shader operation. The
RTU implements traversal logic to traverse the acceler-
ation structure including transformation of rays as need-
ed"
 

assurdum

Banned
Can someone explain? What is it?
This is a software? (or driver)?
If this can be done on current hardware means it could run on any compatible hardware? (incl. xbox or nvidia?)
This is RDNA2 or even PS5 exclusive optimization?
I
From what I have understood it accelerates the raytracing process sparing operations via async job. It something you can do at low level of API access I guess.
 
Last edited:

Dr Bass

Member
and what would that di exactly? is this weird believe that Insomniac's RayTracing implementation is magically better than others still alive? because, news flash... it isn't... they just made the correct decisions on how to implement it, but technically it's not in any way special
Couple things.

How do you know? Have you ever written any ray tracing code, real time or offline? I have! 😁 And yet I have no opinion on this because I have zero insight into how their code works. But apparently you know it was “the right way but nothing special.” Which would imply that most others are doing it “the wrong way.”

Given that software development, and especially games development, is a generally difficult task, it would stand to reason that making games requires some smarts. But you’re suggesting most graphics programmers here have a blind spot with ray tracing since most of them are “doing it the wrong way.” Either that or they are stupid, but that can’t really be the case if they are making decent games otherwise, correct?

I don’t think anyone said Insomiacs ray tracing was “magically” better than anyones. It’s better through hard work and smart programming. At least on the console side, where RT will be limited in general.

But since you obviously know “the right way” to do these things perhaps you can offer your consulting services to the gaming studios?

Next issue …

Software patents suck.
 

assurdum

Banned
It looks like alot of nvidia patents about ray traversal, talking about using a dedicated core for RT traversal (nvidia patents are worded like bvh traversal on RT cores etc.) which AMD apparently doesn't have yet so i'm not sure where this is going. Is it a clue AMD's next line of gpus will have dedicated RT cores? and Sony expect to use these in a ps6 or pro down the line?

"A graphics processing unit (GPU) includes one
or more processor cores adapted to execute a soft-
ware-implemented shader program, and one or more
hardware-implemented ray tracing units (RTU) adapted
to traverse an acceleration structure to calculate inter-
sections of rays with bounding volumes and graphics
primitives asynchronously with shader operation. The
RTU implements traversal logic to traverse the acceler-
ation structure including transformation of rays as need-
ed"
CUs are the dedicated cores for raytracing in the AMD GPU. It's knew from awhile.
 

assurdum

Banned
Couple things.

How do you know? Have you ever written any ray tracing code, real time or offline? I have! 😁 And yet I have no opinion on this because I have zero insight into how their code works. But apparently you know it was “the right way but nothing special.” Which would imply that most others are doing it “the wrong way.”

Given that software development, and especially games development, is a generally difficult task, it would stand to reason that making games requires some smarts. But you’re suggesting most graphics programmers here have a blind spot with ray tracing since most of them are “doing it the wrong way.” Either that or they are stupid, but that can’t really be the case if they are making decent games otherwise, correct?

I don’t think anyone said Insomiacs ray tracing was “magically” better than anyones. It’s better through hard work and smart programming. At least on the console side, where RT will be limited in general.

But since you obviously know “the right way” to do these things perhaps you can offer your consulting services to the gaming studios?

Next issue …

Software patents suck.
Insomniac raytracing was magically better because showed how incredibly "flexible" is the raytracing, using the low level access of ps5 API. You have an access to the rays in a way an higher level of abstraction can't give (as direct X).
 
Last edited:
From what I have understood it accelerates the raytracing process sparing operations via async job. It something you can do at low level of API access I guess.
So, because PS5 has some unique hardware inside and/or have access to low-level?
I don't understand why Mark Cenry and why it's PS5-only as everyone uses same base HW.
 
Can someone explain? What is it?
This is a software? (or driver)?
If this can be done on current hardware means it could run on any compatible hardware? (incl. xbox or nvidia?)
This is RDNA2 or even PS5 exclusive optimization?
I


Looks like a method of enabling an extra step in the raytracing stack that uses shader processors to shorten rays, probably as a way to increase RT performance.
All the hardware capabilities I see are RDNA2, so this is for the PS5 but I don't see a reason why the same method couldn't be applied on other RDNA2 architectures down the road.

RDNA2 accelerates less stages on raytracing processing than Ampere and even Turing but OTOH it's reportedly more flexible. This seems to be making use of that.
 

assurdum

Banned
So, because PS5 has some unique hardware inside and/or have access to low-level?
I don't understand why Mark Cenry and why it's PS5-only as everyone uses same base HW.
No it's not about an unique hardware. It's the coding method of the ps5 which give some advantages.
 

assurdum

Banned
Looks like a method of enabling an extra step in the raytracing stack that uses shader processors to shorten rays, probably as a way to increase RT performance.
All the hardware capabilities I see are RDNA2, so this is for the PS5 but I don't see a reason why the same method couldn't be applied on other RDNA2 architectures down the road.

RDNA2 accelerates less stages on raytracing processing than Ampere and even Turing but OTOH it's reportedly more flexible. This seems to be making use of that.
Depends to the direct X. From what I heard the direct X not allow an access to the raytracing as the low level API of ps5.
 
Last edited:
It looks like alot of nvidia patents about ray traversal, talking about using a dedicated core for RT traversal (nvidia patents are worded like bvh traversal on RT cores etc.) which AMD apparently doesn't have yet so i'm not sure where this is going. Is it a clue AMD's next line of gpus will have dedicated RT cores? and Sony expect to use these in a ps6 or pro down the line?

"A graphics processing unit (GPU) includes one
or more processor cores adapted to execute a soft-
ware-implemented shader program, and one or more
hardware-implemented ray tracing units (RTU) adapted
to traverse an acceleration structure to calculate inter-
sections of rays with bounding volumes and graphics
primitives asynchronously with shader operation. The
RTU implements traversal logic to traverse the acceler-
ation structure including transformation of rays as need-
ed"

this patent claims (from what i can gather) a software methodoligy to avoid / improve barriers / cache pressure through a-synchronicity of RA<->Shader interactions.

I think you have some major misconceptions about the actual differences between nvidias RT cores and amd RAs.
 

hlm666

Member
It's talking about async operation of the shader program with the RT traversal on a dedicated core, can AMD do this? I thought the CU could do one or the other? Would they have enough control to maybe use some CU's for RT and throw the RT work at them while leaving other cores to work async on the rest of the shader code?
 

Sosokrates

Gold Member
Maybe something that will be implemented in a PS5 PRO.

Edit: ok, its for PS5 or potentially other gpus.
 
Last edited:

assurdum

Banned
It's talking about async operation of the shader program with the RT traversal on a dedicated core, can AMD do this? I thought the CU could do one or the other? Would they have enough control to maybe use some CU's for RT and throw the RT work at them while leaving other cores to work async on the rest of the shader code?
Who knows? Not impossible if ps5 can do it (it's the same hardware) but for now the only way to use raytracing on AMD it's just with Direct X and it's far behind to the Nvidia tech and the ps5 API uses (looking to Insomniac games) hearing some developers.
 
Last edited:
Depends to the direct X. From what I heard the direct X not allow an access to the raytracing as the low level API of ps5.
This looks like it could be implemented using DXR 1.1 Inline Raytracing, but as a high-level implementation I don't know if it's performant enough.



The problem with DXR 1.0 is that it was written by or with Nvidia, as AMD came later to the party. I'm sure Microsoft will eventually update DXR to be more RDNA2 friendly considering all their 1st party studios need to launch games that look good on their RDNA2 consoles.


Maybe something that will be implemented in a PS5 PRO.
No, this is for current hardware.
 
Last edited:

assurdum

Banned
This looks like it could be implemented using DXR 1.1 Inline Raytracing, but as a high-level implementation I don't know if it's performant enough.
The problem of the high level of abstraction is you can't handle the raytracing method in a flexible/straight way as you can with the low level of API access. It's the disadvantage of the virtual abstraction in the coding.
 
Last edited:
It's talking about async operation of the shader program with the RT traversal on a dedicated core, can AMD do this? I thought the CU could do one or the other? Would they have enough control to maybe use some CU's for RT and throw the RT work at them while leaving other cores to work async on the rest of the shader code?
One CU has 64 shader processors and a dedicated raytracing unit next to the texture filtering and texture mapping units (TMU). The shader processors can work in parallel with the RT unit, though it looks like the RT unit can't work in parallel with the TMU.

RDNA 2 deep-dive: What's inside AMD's Radeon RX 6000 graphics cards -  OfficeJo Computer & Printer Shop


Using shader processors in parallel with RT units has always been used in the modern hybrid raster/ray-tracing architectures. That is not the novelty that the patent mentions.
 
Last edited:

hlm666

Member
this patent claims (from what i can gather) a software methodoligy to avoid / improve barriers / cache pressure through a-synchronicity of RA<->Shader interactions.

I think you have some major misconceptions about the actual differences between nvidias RT cores and amd RAs.
You are probably right.
 
It's talking about async operation of the shader program with the RT traversal on a dedicated core, can AMD do this? I thought the CU could do one or the other? Would they have enough control to maybe use some CU's for RT and throw the RT work at them while leaving other cores to work async on the rest of the shader code?

this patent claims (from what i can gather) a software methodoligy to avoid / improve barriers / cache pressure through a-synchronicity of RA<->Shader interactions.

I think you have some major misconceptions about the actual differences between nvidias RT cores and amd RAs.

I made you a GPU profile of WD:L running on a 6700XT to show you the problem with rdna2's RT (RT pipeline in red). i think its probably not "unaccelerated ray traversal" (which makes maybe 10% of the RT frametime) but that because of shared registries and logic interdependencies RT occupancy [in this case] is only at 25% for long portions of the frame.



i think that is the general problem this patent is trying to improve on.
 
Last edited:
Top Bottom