Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]

[iroboto]

If you don’t know the tier level support for DXR; you don’t know what it’s capable of.

that is what Matt was alluding to. He knows where DXR 1.0 sits in terms of performance. But he hasn’t seen DXR1.1 or 1.X which would be a variant specific to Xbox.

Lots of posts here about a lot of stuff in DXR that very few of us on this board are even able to qualify. If you haven’t written DXR pipeline and looked at those cases (outside of unreal for instance) you wouldn’t know.

as I see it so far; those challenges that I can see are specific to ray tracing in general and for obvious reasons.

 

[mpg1]

If you don’t know the tier level support for DXR; you don’t know what it’s capable of.

that is what Matt was alluding to. He knows where DXR 1.0 sits in terms of performance. But he hasn’t seen DXR1.1 or 1.X which would be a variant specific to Xbox.

Lots of posts here about a lot of stuff in DXR that very few of us on this board are even able to qualify. If you haven’t written DXR pipeline and looked at those cases (outside of unreal for instance) you wouldn’t know.

as I see it so far; those challenges that I can see are specific to ray tracing in general and for obvious reasons.

I agree, but his tweet seems to be referring to RDNA 2 architecture...not DXR.

 

[AlNom]

Good god I hope not. Or these pages will be filled by “How many bounces my platform can do” “How many gigarays” “how many inches is it tho” “does it do real time boob illumibounce” ad infinitum.

The answer lies solely in Team Ninja's sandy v-

...volleyball court. Which platforms are outselling Xbox and which platforms have a certain volleyball game? I rest my case.

 

[Scott_Arm]

"What efficiencies get unlocked from the perspective of end-to-end game production for 1P, MMO, and GaaS titles?"
???
The fuck does that even mean? Why is he talking about those 3 kinds of games in particular?

He's saying there's a lot more to providing a quality gaming platform and service than just offering a high-power GPU. You need to provide a good development platform, good infrastructure and services etc.

 

[MrFox]

DXR had to be a development between MS, AMD, and Nvidia. MS needs them to make compatible cards, and the gpu manufacturers need MS to fix the standard for PC gaming, and they need to get good benchmarks from all games. Any PC GPU will have to be using the resulting DXR method for ray tracing. For XBSX, MS is using the resulting api from that develpment cycle.

Sony doesn't have any of those business requirements for their RT implementation. Maybe it's the same, maybe it's completely different and was developped by amd and sony over the last 6 years. Maybe it's more performant but needs more silicon, maybe it's less performant but uses less silicon. We know nothing.

 

[iroboto]

I agree, but his tweet seems to be referring to RDNA 2 architecture...not DXR.

The other way I read it was the question whether it was solved. Not necessarily that they managed to solve it.

powerVR bless rys ; still use an acceleration structure to do intersection. They have an additional ray coherency engine on top of it.

People are acting like something else entirely is being developed. The solutions are not that far apart.

 

[AlNom]

The other way I read it was the question whether it was solved. Not necessarily that they managed to solve it.

powerVR bless rys ; still use an acceleration structure to do intersection. They have an additional ray coherency engine on top of it.

People are acting like something else entirely is being developed. The solutions are not that far apart.

https://www.imgtec.com/blog/coheren...racing-the-benefits-of-hardware-ray-tracking/

Microsoft’s DXR, demands an execution model that doesn’t really blend in with the way GPUs like to work, giving any GPU designer that needs to support it some serious potential headaches. That’s especially true if real-time ray tracing is something they haven’t been thinking about for the last decade or so and here at Imagination, we have been.

 

[iroboto]

https://www.imgtec.com/blog/coheren...racing-the-benefits-of-hardware-ray-tracking/

That changed with 1.1 though; the way that API can now interact with DXR via inline in a compute shader or via execute Indirect.
Ray coherency won’t be solved though via hardware. It could be done via compute

 

[Scott_Arm]

DXR had to be a development between MS, AMD, and Nvidia. MS needs them to make compatible cards, and the gpu manufacturers need MS to fix the standard for PC gaming, and they need to get good benchmarks from all games. Any PC GPU will have to be using the resulting DXR method for ray tracing. For XBSX, MS is using the resulting api from that develpment cycle.

Sony doesn't have any of those business requirements for their RT implementation. Maybe it's the same, maybe it's completely different and was developped by amd and sony over the last 6 years. Maybe it's more performant but needs more silicon, maybe it's less performant but uses less silicon. We know nothing.

It's possible the console api would have some extensions that would provide lower-level access than what would be available with DXR on PC. They've done the same with other parts of the directx api in the past.

 

[BRiT]

We know nothing.

We are Jon Snow.

 

[mpg1]

The other way I read it was the question whether it was solved. Not necessarily that they managed to solve it.

powerVR bless rys ; still use an acceleration structure to do intersection. They have an additional ray coherency engine on top of it.

People are acting like something else entirely is being developed. The solutions are not that far apart.

Well I look at this 2 ways:

1. He is out of the loop with regards to AMD's RT because they are using something different (though I am still skeptical of this because of that Cerny quote stating the "There is ray-tracing acceleration in the GPU hardware")

2. A much worse case scenario.... AMD RT is far behind in development/not working well enough that even the engineers at the console manufactures don't know it's technical capabilities yet....

 

[iroboto]

Well I look at this 2 ways:

1. He is out of the loop with regards to AMD's RT because they are using something different (though I am still skeptical of this because of that Cerny quote stating the "There is ray-tracing acceleration in the GPU hardware")

2. A much worse case scenario.... AMD RT is far behind in development/not working well enough that even the engineers at the console manufactures don't know it's technical capabilities yet....

MS wouldn’t have burned their chip if the hardware wasn’t ready. It’s clear that it meets spec. So it can’t be option 2.
Whatever MS has planned for DXR and DXR wrt Xbox they are done. And MS has been chugging along quietly and confidently.

 

[DavidGraham]

Microsoft’s DXR, demands an execution model that doesn’t really blend in with the way GPUs like to work, giving any GPU designer that needs to support it some serious potential headaches. That’s especially true if real-time ray tracing is something they haven’t been thinking about for the last decade or so and here at Imagination, we have been.

https://www.imgtec.com/blog/coheren...racing-the-benefits-of-hardware-ray-tracking

So PowerVR is advocating an even more reliance on hardware acceleration for RTm, what do you think about that? @JoeJ

 

[Deleted member 13524]

1. He is out of the loop with regards to AMD's RT because they are using something different (though I am still skeptical of this because of that Cerny quote stating the "There is ray-tracing acceleration in the GPU hardware")

Is there a reason to assume if it isn't AMD's RT blocks then it can't be integrated in the SoC, next to the iGPU?

Moreover, if we're talking about something off-SoC then it could still be called part of the GPU hardware.

 

[Scott_Arm]

Is there a reason to assume if it isn't AMD's RT blocks then it can't be integrated in the SoC, next to the iGPU?

Moreover, if we're talking about something off-SoC then it could still be called part of the GPU hardware.

Maybe wanting to have the ray tracing blocks tightly-coupled to CUs to share caches and have CUs dispatch RT work directly to a local RT block? The Nvidia GPUs have RT cores very tightly coupled to SMs.

 

[dobwal]

https://www.imgtec.com/blog/coheren...racing-the-benefits-of-hardware-ray-tracking/

I’m not sure how you interpreted that but I interpreted this as DXR uses an execution model that doesn’t fit well with current GPUs. Imagination overcomes this dilemma with changes to the hardware.

Inevitably, the problem isn’t DXR but that current hardware design needs to be addressed. It doesn’t seem like Rys is trying to disparage DXR in anyway. But rather implies that DXR is going to be a bitch to use because GPUs in general will force devs to overcome raytracing limitations that current GPUs present.

 

[Proelite]

Maybe wanting to have the ray tracing blocks tightly-coupled to CUs to share caches and have CUs dispatch RT work directly to a local RT block? The Nvidia GPUs have RT cores very tightly coupled to SMs.

Like this?

 

[iroboto]

I’m not sure how you interpreted that but I interpreted this as DXR uses an execution model that doesn’t fit well with current GPUs. Imagination overcomes this dilemma with changes to the hardware.

Inevitably, the problem isn’t DXR but that current hardware design needs to be addressed. It doesn’t seem like Rys is trying to disparage DXR in anyway. But rather implies that DXR is going to be a bitch to use because GPUs in general will force devs to overcome raytracing limitations that current GPUs present.

if I understood things correctly, Rys is right. DXR 1.0 required you to make a ray intersection call and from there call the compute shaders to do the work.
It's an entirely new path.

1.1 allows you to inline DXR commands or to execute indirect them, meaning you can take your existing compute shader code and inline DXR into it. This makes for enhancing the existing pipeline if that is the process you want to go without having to throw away traditional render paths.

 

[dobwal]

if I understood things correctly, Rys is right. DXR 1.0 required you to make a ray intersection call and from there call the compute shaders to do the work.
It's an entirely new path.

1.1 allows you to inline DXR commands or to execute indirect them, meaning you can take your existing compute shader code and inline DXR into it. This makes for enhancing the existing pipeline if that is the process you want to go without having to throw away traditional render paths.

I’m getting that regardless of DXR, devs are going to be forced to handle coherency gathering in software while Imagination’s solution does it transparently in hardware.

 

[iroboto]

I’m getting that regardless of DXR, devs are going to be forced to handle coherency gathering in software while Imagination’s solution does it transparently in hardware.

Yea that also is a good take