Next Generation Hardware Speculation with a Technical Spin [post E3 2019, pre GDC 2020] [XBSX, PS5]

Status
Not open for further replies.
The part of the article you’re underlining is not a direct quote from Mark Cerny. It’s the interpretation of the interviewer. Again, I think it makes little to no sense to have a gpu that most likely has hardware ray tracing acceleration and then not use it for audio. All of AMDs audio work is in using the gpu to do real-time 3D audio.

The datastructure for raytracing acceleration will be in RAM, it can be use from a custom 3d audio unit and without interfering with graphical workload. An example Quaz51 when he was here on B3d meet Christophe Balestra in an IGDA meeting when he was at ND working on Uncharted. He asked a question about PS3 architecture and one of the only change he would talk in public it would have done to PS3 is to have a specliazed chip for audio and let more SPU power for graphical task. When Audio is done on GPU it fights with graphical task and graphics seems more important than audio.

mm4085-designing-a-game-audio-engine-for-hsa-by-laurent-betbeder-6-638.jpg


And now some of the default on CPU side disseapear, this is not a weak CPU anymore.
mm4085-designing-a-game-audio-engine-for-hsa-by-laurent-betbeder-8-638.jpg
 
Last edited:
@chris1515 PS3 is 14 years old. The hardware and development process on PS3 are not relevant anymore. And by AMDs own research the slides you are posting are outdated. AMD has a solution for the latency problem.

If you're doing audio processing on the gpu, and you're worried about the impact on graphics, you pick a faster gpu. Why bother making another chip? TrueAudio Next is AMD's current audio solution, and it is entirely a gpu solution with fallbacks to the cpu. I'm sure they could make a separate audio block that would look large likely a few compute units from the gpu, but why? Unless there's some specific technical reason why the gpu is not suited to this task, then I don't know why they'd re-invent the wheel (both hardware and software). My guess is the audio block is just two CUs from the GPU reserved for audio, and that's probably what Xbox will do as well.

From the navi whitepaper:

upload_2020-2-7_9-43-50.png
 
Exotic or custom hw in the sense of a seperate chip likely isnt going to happen, wont do any advantage to most games aside from some exclusives. It is most likely a solution from amd that can be found on navi/navi2, just like trueaudio in 2013 gpu’s (290x and onward).
 
@chris1515 PS3 is 14 years old. The hardware and development process on PS3 are not relevant anymore. And by AMDs own research the slides you are posting are outdated. AMD has a solution for the latency problem.

If you're doing audio processing on the gpu, and you're worried about the impact on graphics, you pick a faster gpu. Why bother making another chip? TrueAudio Next is AMD's current audio solution, and it is entirely a gpu solution with fallbacks to the cpu. I'm sure they could make a separate audio block that would look large likely a few compute units from the gpu, but why? Unless there's some specific technical reason why the gpu is not suited to this task, then I don't know why they'd re-invent the wheel (both hardware and software). My guess is the audio block is just two CUs from the GPU reserved for audio, and that's probably what Xbox will do as well.

From the navi whitepaper:

View attachment 3573

Because they can do it on CPU for example or a custom audio unit and it does not take compute from graphics.

SPU or CU it is the same it means audio can be sacrifice because graphics is more important.
 
@chris1515Why bother making another chip?
For better efficiency and better results? Is TrueAudio Next any good? This example from AMD isn't anything like high-quality spatial audio:


I've tried a few other demos on YT and not found anything that comes close to decent realistic 3D audio. If you can get proper immersive 3D audio on CPU or GPU efficiently, yeah, it makes sense over having custom hardware. But if you can't, custom hardware makes sense, and if the issue of next-gen sound was already a solved problem, why isn't it a case that games have great, realistic 3D audio?
 
Sony bought Audiokinetic, they will not use GPU compute like on PS4. This is the reason they say custom audio 3D chip.

https://www.audiokinetic.com/about/news/sony-interactive-entertainment-to-acquire-audiokinetic/

3622979-ps5.png
Has anyone else noticed that PS5 states "HARDWARE-BASED RAY TRACING" on their slide and not "HARDWARE ACCELERATED RAY TRACING"?. After doing some research myself I was unable to find a single instance of AMD using the term hardware-based ray tracing. AMD seems to always use the term hardware accelerated ray tracing, and MS uses this term as well. It's probably just a choice of verbiage by Sony but I found it interesting. I've notice a few people talking about this and I was curious if anyone had any thoughts on it. Appreciate anyone that has the time to respond. Also It is entirely possible AMD has used the term hardware-based ray tracing in the past but i just missed it.
 
Has anyone else noticed that PS5 states "HARDWARE-BASED RAY TRACING" on their slide and not "HARDWARE ACCELERATED RAY TRACING"?. After doing some research myself I was unable to find a single instance of AMD using the term hardware-based ray tracing. AMD seems to always use the term hardware accelerated ray tracing, and MS uses this term as well. It's probably just a choice of verbiage by Sony but I found it interesting. I've notice a few people talking about this and I was curious if anyone had any thoughts on it. Appreciate anyone that has the time to respond. Also It is entirely possible AMD has used the term hardware-based ray tracing in the past but i just missed it.
somewhere between pages 5-50, and 60-80, and 80-100 this becomes a topic of discussion. It has been resolved by Cerny saying that the PS5 supports hardware accelerated ray tracing. That was just the verbiage they went with at the time.
 
noice. Love this type of discussion. Creative Labs and Aureal is gonna make a comeback

Still unbeaten to this day, advancement in audio hasn't been a priority so far. I think the OG xbox (2001) had support for that in hardware too? Or is it HRTF im confused with?
 
somewhere between pages 5-50, and 60-80, and 80-100 this becomes a topic of discussion. It has been resolved by Cerny saying that the PS5 supports hardware accelerated ray tracing. That was just the verbiage they went with at the time.
Thank you for the response. I figured that would be the case where hardware-based implies it's not a software solution. It will be interesting to see if both the PS5 and XBSX use the same solution. I'm guessing they will ,and that's probably the best outcome to ensure developers use it to it's fullest capabilities.
 
Thank you for the response. I figured that would be the case where hardware-based implies it's not a software solution.
'Hardware based' and 'hardware accelerated' both mean not using software. If there is a difference in degree, 'hardware based' would imply more dedicated silicon than 'hardware accelerated' in the context of accelerated being shared between software and hardware (RTX) versus a solution based on custom RT hardware, whatever the heck that might look like. Some sort of processor that builds and traverses RT structures and performs RT tests.

In all likelihood, it's just hardware accelerated, maybe called 'hardware based' because some people are willing to imagine 'hardware accelerated' can mean 'running on the normal GPU hardware' and someone thought it important to make note that there's actually hardware designed for RT in there.
 
There are two parts of object based audio processing. Ray tracing the source to listener with all reflections, and the actual audio processing that can be very intensive if you start doing the convolutions for really good reverb.

They might use the ray tracing hardware for sound propagation and reflections, and have some dsps for the actual stream processing.

Was trueaudio just a bunch of dsps?
 
Last edited:
There are two parts of object based audio processing. Ray tracing the source to listener with all reflections, and the actual audio processing that can be very intensive if you start doing the convolutions for really good reverb.

They might use the ray tracing for sound propagation and reflections, and have some dsps for the actual stream processing.

Was trueaudio just a bunch of dsps?

TrueAudio was DSP based, but TrueAudio Next ditched the DSPs for a pure GPU compute solution.
 
Has anyone else noticed that PS5 states "HARDWARE-BASED RAY TRACING" on their slide and not "HARDWARE ACCELERATED RAY TRACING"?.
Good catch. There was speculation in the baseless thread: Maybe Sony had RT plans much earlier than thought, and DXR was a reaction also to this, not only because NV was ready at the time.
Personally i now think Sony has a different and custom RT solution than XBSX + AMD, hinted by those points:
DXR came totally surprising out of nothing. NV exposing this just by API extensions would have been expected, but full integration into DirectX although there was only one HW vendor with support, without any announcement before is just suspicious and seems rushed.
The Github leak mentions AMD RT for XBSX but not for the PS5 chip.
Using completely a different term, 'based' instead 'accelerated' maybe also hints Sony had RT in mind before RTX came up.
 
For better efficiency and better results? Is TrueAudio Next any good? This example from AMD isn't anything like high-quality spatial audio:


I've tried a few other demos on YT and not found anything that comes close to decent realistic 3D audio. If you can get proper immersive 3D audio on CPU or GPU efficiently, yeah, it makes sense over having custom hardware. But if you can't, custom hardware makes sense, and if the issue of next-gen sound was already a solved problem, why isn't it a case that games have great, realistic 3D audio?

The fact that it's not a solved problem is exactly why you want to do it on the GPU. It's incredibly flexible and programmable compute power. Especially now that ray tracing is going to be so heavily involved. Occlusion and reflections will be a key component.
 
'Hardware based' and 'hardware accelerated' both mean not using software. If there is a difference in degree, 'hardware based' would imply more dedicated silicon than 'hardware accelerated' in the context of accelerated being shared between software and hardware (RTX) versus a solution based on custom RT hardware, whatever the heck that might look like. Some sort of processor that builds and traverses RT structures and performs RT tests.

In all likelihood, it's just hardware accelerated, maybe called 'hardware based' because some people are willing to imagine 'hardware accelerated' can mean 'running on the normal GPU hardware' and someone thought it important to make note that there's actually hardware designed for RT in there.

This to me looks like a deep rabbit hole of semantic parsing. I honestly have no idea how you could distinguish between hardware accelerated and hardware based and which one implies more dedicated silicon.They could mean the same thing, they could mean different things. We have no idea what the context is. Is there some absolutist way to say whether Nvidia's RTX is "hardware based" or "hardware accelerated" and be consistent with a different product?
 
Status
Not open for further replies.
Back
Top