Technical investigation into PS4 and XB1 audio solutions *spawn

NDA? Saved it for the AMD keynote?

And when has Sony ever went in depth in explaining their hardware before?

Whenever they had something good they have shown. It has not sense not to do it when all people rages about SHAPE. And these people read beyond3d, neogaf... and it seems they care about what people think there. I have tweeted Yoshida about the subject, let´s see if he answers.
 
The only thing Sony has been vocal about was the 8GB of GDDR5, and rightfully so.... that's pretty significant. The only people raging about SHAPE were the people on B3D. I doubt that's enough to force Sony to come out and give details on their audio chip. And like I said, maybe they couldn't talk about it because of NDA. I have reason to believe that the PS4 at least shares a lot of similarities with the PC tech. If it was as simple as decoding then I doubt they would go as far to say that it has TrueAudio technology and that even AMD themselves would be saying it at their keynote.
 
Last edited by a moderator:
imho it is just matter of writing the audio library driver that supports it.
maybe it was not ready for PS4 on launch time?
As long as the chip capabilities are the same, it is just matter of 'driver' things.
Or maybe Sony was on NDA with AMD... which would also explain it very nicely.
 
imho it is just matter of writing the audio library driver that supports it.
maybe it was not ready for PS4 on launch time?
As long as the chip capabilities are the same, it is just matter of 'driver' things.
Or maybe Sony was on NDA with AMD... which would also explain it very nicely.

The only thing i can think of is that is saved for the VR headset announcement so they can say it comes with stereo 7.1 sound.
 
The PS4 manual indicates that it has always-on voice recognition (which is a surprise).

It would be logical if part of this block is reserved for that purpose.
 
And when has Sony ever went in depth in explaining their hardware before? We usually get the basic stuff like cores, memory etc.
You're kidding, right? Sony have been very vocal about the incredible power of their hardware for over a decade. They gave us pretty clear descriptions of their hardware in PS4, but more importantly they have communciated with the developers using the hardware as to what's in it, and they didn't mention a DSP.

Now it could be the TrueAudio aspect is what's doing the mixing and decoding, in which case Sony have mentioned it as the functions it enables. In this case, the mention of TrueAudio would be what's throwing us, as we are associating that tech with a powerful hardware audio engine as described by AMD for PC. If the TrueAudio in PS4 is that same PC hardware though, that's a significant improvement in one area of PS4 and one Sony didn't mention, despite being clear about everything else in the system (two display planes, upscaler, 18 DX11+ CUs, 8 ACEs and 64 threads, 8 core Jag, HSA, Onion and Onion+ interfaces, ~2-GBs CPU to RAM BW, 8 GBs GDDR5, 172 GB/s BW or whatever it is, low-power ARM based background processor, video encoder/decoder...how much more info do you want if that doesn't count as 'in depth explanation of their hardware'?).

The nature of the hardware is significant in understanding PS4, and everything to date has been transparent. We've received an excellent understanding of the hardware. This DSP news goes completely against that, as simple mixing and decoding of audio could be accomplished by less than a programmable DSP.
 
You're kidding, right? Sony have been very vocal about the incredible power of their hardware for over a decade. They gave us pretty clear descriptions of their hardware in PS4, but more importantly they have communciated with the developers using the hardware as to what's in it, and they didn't mention a DSP.

Now it could be the TrueAudio aspect is what's doing the mixing and decoding, in which case Sony have mentioned it as the functions it enables. In this case, the mention of TrueAudio would be what's throwing us, as we are associating that tech with a powerful hardware audio engine as described by AMD for PC. If the TrueAudio in PS4 is that same PC hardware though, that's a significant improvement in one area of PS4 and one Sony didn't mention, despite being clear about everything else in the system (two display planes, upscaler, 18 DX11+ CUs, 8 ACEs and 64 threads, 8 core Jag, HSA, Onion and Onion+ interfaces, ~2-GBs CPU to RAM BW, 8 GBs GDDR5, 172 GB/s BW or whatever it is, low-power ARM based background processor, video encoder/decoder...how much more info do you want if that doesn't count as 'in depth explanation of their hardware'?).

The nature of the hardware is significant in understanding PS4, and everything to date has been transparent. We've received an excellent understanding of the hardware. This DSP news goes completely against that, as simple mixing and decoding of audio could be accomplished by less than a programmable DSP.

Sony is going to handle the coding for the DSP themselves so maybe that's why they haven't said much to the devs about it at the moment.
 
Yeah, I don´t buy PS4 has True Audio PC implementation. The "based on" wording sound suspicious, and so does not mentioning of stereo "binaural" like effects that is the greatest buying point of this IMHO ( why would you use GPGPU if with the True Audio hardware you can mix and simulate 7.1 channels in stereo? ). And the no more advertising from Sony of the audio hardware after all the noise made from XB1 "Shape" doesn´t help to believe PS4 has True Audio 3 DSPs, maybe only the one that makes the decoding.

Sony traditionally don't like to talk about other companies' technologies. They usually only hype their own brands.

I asked what's so special about TrueAudio. Anandtech kinda answered it. The ideas are not new, but the implementation is better.

In the presentation, the speaker did explain why you want to use the Jaguar cores or the GPGPU for audio sometimes. It depends on your needs. Setting up and processing take 10-20ms end-to-end minimally. This is sufficient for regular games but not music games. The developers will need to hide this latency somehow.

The GPGPU is very suitable for most audio processing except for a certain class of problems. The waiting in the queue may also require fine tuning to minimize latency.

The Jaguar cores are good at audio processing and flexible, but still lose out to the SPUs when the clock difference is normalized.

Since they are all fully programmable, it should be possible to achieve any effect the developers need in a game. The challenge is probably to provide a consistent, system-wide audio experience. Will be interesting to see how Sony tackle the sound domain on PS4. They had one or two interviews on audio technologies on PS3 and Vita separately.
 
The GPGPU is very suitable for most audio processing except for a certain class of problems. The waiting in the queue may also require fine tuning to minimize latency.

The GPGPU is only suitable for simple and feedforward systems (convolution/fir etc.) - not for branch heavy algorithms and feedback systems.
 
The GPGPU is only suitable for simple and feedforward systems (convolution/fir etc.) - not for branch heavy algorithms and feedback systems.

I might have missed it. Was he talking about decoding at that point ?

The presenter did mention heavy data dependency falls into that class of problems.

Edit:
In the last PS3 audio article, the interviewee speculated that nextgen audio is heading towards dynamic synthesis to minimize developers' work. Here, the speaker mentioned physics driven (low level) signals to trigger sound and effects.
 
The presenter called PS4 audio chip "ACP which is a specialized DSP" (09:40).
https://vts.inxpo.com/scripts/Server.nxp?LASCmd=AI:1;F:SF!42000&EventKey=115505

audio1-600x360.jpg

http://www.vgleaks.com/playstation-4-audio-processor-acp/
 
Some points from the presentation.

* The DSP/ACP will probably be rather locked regarding custom audio DSP lines because of the latency (minimum 2 frames) and fixed functions.

* The DSP/ACP will mostly be used for codecs and filterbanks for voice/echo cancellation processing (fixed functions on the Tensilica) etc.

* Even though he writes 'many' audio DSP functions can be delegated to the GPU - he could obviously only mention the two sets of workload suitable for GPU : simple (multiply = volume, addition = Mixing) and feedforward (convolution / fir / FFT). In the audio workstation environment - only 20% is related to these functions.

* The GPU is actual not suitable for most audio workloads because of the latency related to the compute queue. Only mid to high latency resistent audio functions could be delegated to the GPU - like a convolution reverb.

* The CPU is the prefered unit for audio DSP and the ONLY solution for low latency functions.

* The Jaguar cores are plenty fast - he do compare the performance to hand writen assembly SPU code, but was the x86 code vectorized CLANG/LLVM code or handwriten intrinsics?

*The gates are now open for high-end DSP functions from the workstation environment because of x86 - this is good news for us :)
 
Last edited by a moderator:
I'm starting to think that TrueAudio isn't really in PS4, and they just used the framework as a 'fixed-function' audio unit for OS processing off audio. So the TrueAudio, Tensillica core is just a decode+mix, and maybe a little voice work (that has more invovlement than we had heard up until recently). That's far more feasible than either 1) PS4 had a powerful audio DSP but Sony kept quiet about it, or 2) It's a last minute addition of the PC design. Not that anything is completely ruled out yet.
 
Also on Jaguar performance:
*On Killzone SF: 300 voices [probably a rough number to give an idea] with ton of DSP's, auxillary effects, mastering effects, all that stuff runs in less than 50% of 1 jaguar core.
*Ultra low latency
*A quarter slower than SPU's (no surprises there) at same clock, lacks fused multiply accumulate (which was present on Bulldozer, and SPU's)

I'm starting to think that TrueAudio isn't really in PS4, and they just used the framework as a 'fixed-function' audio unit for OS processing off audio.

This seems to be the case. The ACP mentioned in the HSA for Audio presentation is not a good fit for "modular content driven DSP pipelines", but is good for decoding stuff and maintaining OS functions.

So the revelations of TrueAudio presence in PS4 is either wrong information (to the point that I wouldn't understand why it is being mentioned in the keynote) or maybe the TrueAudio is supported through api's that work on CPU/GPU.

The HSA for Audio presentation doesn't make it sound like the ACP is something to write home about.
 
Last edited by a moderator:
I'm starting to think that TrueAudio isn't really in PS4, and they just used the framework as a 'fixed-function' audio unit for OS processing off audio. So the TrueAudio, Tensillica core is just a decode+mix, and maybe a little voice work (that has more invovlement than we had heard up until recently). That's far more feasible than either 1) PS4 had a powerful audio DSP but Sony kept quiet about it, or 2) It's a last minute addition of the PC design. Not that anything is completely ruled out yet.

The unit should be programmable by Sony. It is not programmable by the game developers. There are security and latency concerns, especially when the OS uses it in parallel.

This seems to be the case. The ACP mentioned in the HSA for Audio presentation is not a good fit for "modular content driven DSP pipelines", but is good for decoding stuff and maintaining OS functions.

So the revelations of TrueAudio presence in PS4 is either wrong information (to the point that I wouldn't understand why it is being mentioned in the keynote) or maybe the TrueAudio is supported through api's that work on CPU/GPU.

The HSA for Audio presentation doesn't make it sound like the ACP is something to write home about.

What's missing is the software framework for making the CPU, GPU and the ACP work together in a performant way (while keeping the GPU and CPU occupied). He merely speculated at using physics signals to "thread" through the end-to-end audio pipelines.
 
Also on Jaguar performance:
*On Killzone SF: 300 voices [probably a rough number to give an idea] with ton of DSP's, auxillary effects, mastering effects, all that stuff runs in less than 50% of 1 jaguar core.
*Ultra low latency
*A quarter slower than SPU's (no surprises there) at same clock, lacks fused multiply accumulate (which was present on Bulldozer, and SPU's)

Are there any transcripts around? If they're basing this performance comparison on SIMD throughput which the comment about FMA alludes to then that raises more questions.
 
I'm starting to think that TrueAudio isn't really in PS4, and they just used the framework as a 'fixed-function' audio unit for OS processing off audio. So the TrueAudio, Tensillica core is just a decode+mix, and maybe a little voice work (that has more invovlement than we had heard up until recently). That's far more feasible than either 1) PS4 had a powerful audio DSP but Sony kept quiet about it, or 2) It's a last minute addition of the PC design. Not that anything is completely ruled out yet.

Maybe the TrueAudio API is supported but on the backend the ACP is leveraging a combination of CPU/GPU and the hardware decoder/voice chip to emulate what would be in a full PC hardware implementation. Not the processing savings initially thought but still a win to gain adoption by virtue of having installed base on every PS4.
 
Maybe the TrueAudio API is supported but on the backend the ACP is leveraging a combination of CPU/GPU and the hardware decoder/voice chip to emulate what would be in a full PC hardware implementation. Not the processing savings initially thought but still a win to gain adoption by virtue of having installed base on every PS4.

Nope. It would introduce too much latency.

If they say they use TrueAudio, some variant of the h/w should be there.

TrueAudio is not an API right ? It's a DSP tailored for audio work.
 
Nope. It would introduce too much latency.

If they say they use TrueAudio, some variant of the h/w should be there.

TrueAudio is not an API right ? It's a DSP.

Latency-sensitive functions would go to the CPU, non-senstive to the GPU, supported functions to the hardware chip, etc
 
Back
Top