RSX Audio?

[Urian]

I have seen recently the DevStation agenda and it seems that Sony will give a keynote to the developers about how to manage audio in the RSX.

What type of sound coprocessor could be included on the RSX, DD 5.1 Decoder o DD HD Decoder?

 

[London Geezer]

I have seen recently the DevStation agenda and it seems that Sony will give a keynote to the developers about how to manage audio in the RSX.

What type of sound coprocessor could be included on the RSX, DD 5.1 Decoder o DD HD Decoder?

I think this has been mentioned already and people commented that Cell will handle sound - understandably - so this is probably about the fact that since HDMI carries digital sound, it has to go through RSX obviously.
I don't think anyone believes for a moment that they would let RSX handle sound when there are 7 SPEs on Cell to choose from, largely bored, which can handle sound just fine with both their hands tied behind their back.

 

[Carl B]

I have seen recently the DevStation agenda and it seems that Sony will give a keynote to the developers about how to manage audio in the RSX.

What type of sound coprocessor could be included on the RSX, DD 5.1 Decoder o DD HD Decoder?

RSX audio doesn't have anything to do with the RSX GPU. You'll find I provided a link to what RSX audio is in that very thread.

 

[Urian]

The SPE culture is just curious.

From my point of view is better for the programmer if a sound DSP exist and you can use one of the CPU to help it. But programming the system sound when you can get a good API+a good DSP is a lose of time.

 

[dizzyd]

Since we are talking about the PS3 and sound does anyone know if the PS3 will have an optical out jack like the PS2 did? I know I could probably look this up myself, but I'm lazy today.

 

[Synergy34]

I remember seeing the optical output on some of those backside pics taken from E3 or the TGS. Have to dig one up though to be sure.

EDIT: here's a shot of the back, it's there

http://media.ps3.ign.com/articles/614/614619/img_2786114.html

 

[booomups]

more importantly, will there be a console with analog surround channel output......
always having to use a decoder is not ideal.

 

[dizzyd]

I remember seeing the optical output on some of those backside pics taken from E3 or the TGS. Have to dig one up though to be sure.

EDIT: here's a shot of the back, it's there

http://media.ps3.ign.com/articles/614/614619/img_2786114.html

Thanks for the info.

 

[Guden Oden]

so this is probably about the fact that since HDMI carries digital sound, it has to go through RSX obviously.

Actually, the audio in a HDMI connection goes through separate pins compared to video, so audio would not have to be piped through the GPU...

 

[Fafalada]

From my point of view is better for the programmer if a sound DSP exist and you can use one of the CPU to help it.

If you want to get technical - by definition, every CPU is also a DSP. And these days most are considerably more powerfull then the dedicated sound DSPs anyway.

Anyhow, there are tasks that make sense being in dedicated hardware (such as decompression of samples), but the rest of the stuff that would require a programmable DSP to get the most out of it Anyhow - why would anyone settle for the extra cost and complication of yet another different processor when you already have a CPU that is practically built for tasks like these?

 

[DeanoC]

An SPU is good at audio work, its more limited by DMA bandwidth than computational power. So if you really wanted to do more audio work, it would not be a DSP you would want to add but even more high speed RAM...

The main issue we are facing is pyscho-acoustic decompression, they tend to use huge static tables, which are hard to fit into an SPU. I suspect we just need to invent/rediscover a new one that doesn't require the huge tables.

 

[Guden Oden]

Maybe a silly question, but why would one want to use a huge static anything when dealing with compressed stuff? Isn't the idea to save space, not blow it on tables and stuff?

 

[zidane1strife]

If you want to get technical - by definition, every CPU is also a DSP. And these days most are considerably more powerfull then the dedicated sound DSPs anyway.

Anyhow, there are tasks that make sense being in dedicated hardware (such as decompression of samples), but the rest of the stuff that would require a programmable DSP to get the most out of it Anyhow - why would anyone settle for the extra cost and complication of yet another different processor when you already have a CPU that is practically built for tasks like these?

But it shouldn't be any hard, I mean nvidia's probably got a few nforce designs around, and I'd guess such h/w is very small in terms of transistors usage and could easily be integrated into the RSX. I mean if they don't want to use it or want a better solution they can use the spes. Anyway, didn't the xbox 360 have some audio related h/w on the gpu?

PS
Hmmm, Deano, I thought sony'd provide the s/w for it. Do they not? Or do they and you're doing it because just want a few more unsupported audio features?

 

[Shifty Geezer]

But it shouldn't be any hard, I mean nvidia's probably got a few nforce designs around, and I'd guess such h/w is very small in terms of transistors usage and could easily be integrated into the RSX. I mean if they don't want to use it or want a better solution they can use the spes. Anyway, didn't the xbox 360 have some audio related h/w on the gpu?

XB360 has hardware audio decompression only. All the mixing and transforming is done on the CPU.

Having custom audio hardware is pushing the boundaries of what you have a CPU for. It's cheaper and more versatile to have a powerful progammable chip to set to the tasks you want, than a collection of specialised hardware components. Which would you prefer to work with?

a) A system of powerful and versatile programmable CPU and GPU
b) A system of seperate game-logic processor, audio processor, physics processor, graphics processor, image post-processor, hardware ray-caster, audio decompressor, video decompressor, and other seperate components with their own programming models that you may or may not use in your game

The idea of providing a separate processor to handle audio and thus relieve the Cell of the burden seems redundant when you think you've got power already available. There are those who feel with Cell running game, AI, physics, and audio, it'll get bogged down. You may be right, but no system has infinite power. If the designers feel the system can't cope with workload (though workload will always be scaled to system), they need to add more you're processing power, and if you're going to add silicon to do a task, you have to weigh up whether custom silicon will be more beneficial than versatile, programmable hardware. If you're going to add a DSP, why not add another SPE instead? And it's not just the silicon chip you have to worry about, but data pathways and interfacing too.

Custom processors aren;t always the best solution. The Amiga was a multi-chip solution with lots of custom processors, but after a while it's performance fell behind that of a the monolithic processing power of the x86. You have to balance out cost, versatility, ease of development, ease of system integration and such to determine if a piece of hardware is worth including. And sooner or later you have to draw the line when to stop adding features and driving up the cost, even if a feature would be nice.

 

[London Geezer]

Well according to Faf, all sound processing could run on an SPE and leave a lot of room for other things too, on that same SPE.

Maybe DeanoC could clarify, if he can...

 

[DeanoC]

Hmmm, Deano, I thought sony'd provide the s/w for it. Do they not? Or do they and you're doing it because just want a few more unsupported audio features?

Yep Sony provides some software, but it still has to be written and hard things remain hard things, who-ever writes them...

 

[zidane1strife]

XB360 has hardware audio decompression only. All the mixing and transforming is done on the CPU.

Having custom audio hardware is pushing the boundaries of what you have a CPU for. It's cheaper and more versatile to have a powerful progammable chip to set to the tasks you want, than a collection of specialised hardware components. Which would you prefer to work with?

a) A system of powerful and versatile programmable CPU and GPU
b) A system of seperate game-logic processor, audio processor, physics processor, graphics processor, image post-processor, hardware ray-caster, audio decompressor, video decompressor, and other seperate components with their own programming models that you may or may not use in your game

The idea of providing a separate processor to handle audio and thus relieve the Cell of the burden seems redundant when you think you've got power already available. There are those who feel with Cell running game, AI, physics, and audio, it'll get bogged down. You may be right, but no system has infinite power. If the designers feel the system can't cope with workload (though workload will always be scaled to system), they need to add more you're processing power, and if you're going to add silicon to do a task, you have to weigh up whether custom silicon will be more beneficial than versatile, programmable hardware. If you're going to add a DSP, why not add another SPE instead? And it's not just the silicon chip you have to worry about, but data pathways and interfacing too.

Honestly if we're just gonna keep throwing more and more cores in the coming future, I don't see a problem with using some of that space to throw a few small and maybe even old pieces of dedicated h/w. I'm not talking adding large beefy chips, they don't even have to do the full deal(a portion like the 360 stuff). How large is the ps2 sound processor? 2-3M transistors, or something like that? a small upgrade and it'd still be small and could probably do your average surround sound. That'd easily fit in the gpu or along with the cell. If dev.s wanted more they could use the cell to improve audio performance, ala EE in ps2, if not the audio'd be taken care of.

I'm just a bit nervous that shuffling audio along with other stuff might not allow one of the spus to show its maximum potential. Especially with all this talk of space limitations, and knowing how that's said to be the spus achilles' heel.

Custom processors aren;t always the best solution. The Amiga was a multi-chip solution with lots of custom processors, but after a while it's performance fell behind that of a the monolithic processing power of the x86. You have to balance out cost, versatility, ease of development, ease of system integration and such to determine if a piece of hardware is worth including. And sooner or later you have to draw the line when to stop adding features and driving up the cost, even if a feature would be nice.

It seems nowadays the days of ever bigger single core procs are all but gone. Either such designs are simply impractical/unviable, or their design is too complex for human minds to achieve results as good as multi-core. If we're gonna have so much space that we're going to be throwing 10s if not 100s of cores in, might as well throw in a few small pieces of dedicated h/w.

For example rather than throwing 100s of mini cores, if future gpus become versatile enough, it may be better to simple throw the 1-4 main cpu cores and mix it in with future gpuish h/w rather than 100s of minicores.

 

[TrungGap]

Custom processors aren;t always the best solution. The Amiga was a multi-chip solution with lots of custom processors, but after a while it's performance fell behind that of a the monolithic processing power of the x86. You have to balance out cost, versatility, ease of development, ease of system integration and such to determine if a piece of hardware is worth including. And sooner or later you have to draw the line when to stop adding features and driving up the cost, even if a feature would be nice.

Off topic: The Amiga hardware fell behind not because of the technology, but because of the business direction. Having multiple chips to handle different things allows manufacture to quickly get a product out of the door (especially if they pick off the shelf products) and reduce development risk. But as any form of cutting cost, integration will reduce the cost of manufacturing.

What is this audio stuff they're putting in RSX? Is it just a DAC or does it actually do audio processing? The question is why RSX and not Cell? I imagine the Cell have more than enough juice to do audio processing right?

 

[psurge]

DeanoC - have you guys looked at the fmod stuff at all? They claim to be developping a version for Cell on their website...

Serge

 

[Edge]

Well according to Faf, all sound processing could run on an SPE and leave a lot of room for other things too, on that same SPE.

I've touched on this before also, and it's not hard to see that a 3.2 GHz processor (the SPE) with 128 x 128-bit registers, and 256 KB of localized memory is overkill for just sound processing in a game. One quarter (one quarter of the processing, registers, and localized memory) of a SPE's total resources is plenty of power for sound processing, and the rest of the SPE could be used for something else.

When I've discussed this before, some people brought up some unrealistic objections, like excessive amounts of sound channels, etc., but my observation is in regards to a typical game using 5.1 Dolby surround sound with dozens of internal sound sources being multiplexed togeather onto those channels. That would be a realistic work load for a typical game level.

A typical game level would have no more than 5 to 20 MB of sound data, and the SPE would have no problems dealing with such an amount of data, and processing it.