Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Johnny B]

Nathan,

Well you mention some of the people working in digital, some of whom disagree with one another on certain issues IIRC. I'm not certain any of those you mention have much input into the chip makers. Most established companies with good policies in place often turn to the field for input, they turn to the end users, in fact, many will admit that their best ideas come out of the field from their end users.

And while it is true that one may want to give greater weight to the heavy hitters input, usually they will also give some value to the not-so-heavy-hitters as well. From what I understand this is based in part from a desire to serve a broad market and make as many in the targeted end user community as happy as possible. It is in this way that all input from end users can be seen to have value...including input about the end users' complaints and their problems.

Some will go so far as to say things like: "End user complaints, we love 'em. Gives us something to work on."

Further, it is not unheard of that some end user will come up with an off-the-wall solution to a problem that eventually finds its way into a more refined product.




     

[crm0922]

Johnny, what is your fascination with 64-bit CPUs and their relation to audio quality?

A 64bit CPU does not change sampling theory.  It is simple the ability to utilize a larger wordlength for processing purposes.

This might let you run more plugins, address more memory, etc., but 64-bit native isn't a help anywhere except in cycle-to-cycle performance.

It will not be responsible for digital audio undergoing a revolution in "sonics".

Chris

[Johnny B]

Chris,

I'm not so sure, we do not currently have the sorts of things that one might envision which could be helped by 64-bits. I sense we will see it whether it helps or not, but maybe thinking about how DSP, truncation, latency, and those sorts of things might be helped could be a useful exercise.

Right off the top you get twice the number of registers, that helps with software samplers doing disk streaming, and you get gobs more memory access...

I sense there may be some places to get rid of bottlenecks as well, I could be wrong, I often am.

 

[PookyNMR]

Johnny B wrote on Tue, 04 October 2005 22:39

And while it is true that one may want to give greater weight to the heavy hitters input, usually they will also give some value to the not-so-heavy-hitters as well.

I can easily agree that all companies want to listen to their customers.  If they don't they won't have customers!  

I do, however, believe that 99% of their attention will be given to heavy hitters who can provide them with specific requests and specific data to back up their reqeusts.  Part of it is marketing politics, the other part is that these fellows are heavy hitters because they have the ears and technical skills to get the repeat high paying work.   This gives a certain legitimacy to their technical complaints.  

Another part is that the design engineers know the technical terms and the science behind the devices so they can effectively communicate specifically what improvement they are looking for towards the parts manufacturers.

When us little guys who do little indy label projects complain that something doesn't sound like we want it to and we don't have the technical chops to put it into proper scientific terms, our comments are most likely taken with a grain of salt.  

[Jon Hodgson]

Johnny B wrote on Wed, 05 October 2005 17:02

Chris, I'm not so sure, we do not currently have the sorts of things that one might envision which could be helped by 64-bits. I sense we will see it whether it helps or not, but maybe thinking about how DSP, truncation, latency, and those sorts of things might be helped could be a useful exercise. Right off the top you get twice the number of registers, that helps with software samplers doing disk streaming, and you get gobs more memory access... I sense there may be some places to get rid of bottlenecks as well, I coiuld be wrong, I often am.

Wo! Careful there...

You have to be careful about what is the result of 64 bit processing, and what is the result of other changes occuring at the same time.

The doubling of registers is simply a result of the AMD processor design (those extensions now also being adopted by Intel), there's nothing inherent in 64 bit processing that gives you that, it's just that when they added one, they also added the other, saying going to 64 bits gives you double the registers is a bit like saying that getting Magnesium Alloys on your car also gets you a Leather steering wheel because the car of your choice happens to have both on the sports model.

So there's no doubling of registers going from 32 bit to 64 bit in the PowerPC range for example.

DSP processors are a different situation, if the Tiger SHARC is anything to go by, they aren't going to 64 bit processing.. but they are getting longer acumulators which will give much of the same advantage in a number of algorithms.

As for memory addressing space, well not many musical applications need over 4 gigabytes of memory space, so that's not so important.

Well ok if you're using your PC as a sampler for huge sample sets

Most of the time 32bit or 64bit will be completely imperceptable, but there are certain algorithms (some recursive filters for example) where you either need to process at higher than 32 bit resolution, or jump through hoops to do it.

Note also that if you're talking about floats, and you're talking about PCs and Macs, then hardware support for 64 bit operations has been there as standard for 10 years, so there's been nothing to stop programmers from using it apart from performance considerations (twice as much data to move around equals half the speed, unless your system resources are more than you need then you don't do it unless you need to)

As far as audio processing goes, the new Intel and PowerPC processors can't do anything new, they're just faster at doing the same things (as every other generation has been), this will however probably lead to some programmers using greater precision in their algorithms (outside of academia it is a commercial decision, it doesn't matter how good it sounds, if it won't run fast enough on the customers' machines, you get no customers).

Basically quality will continue to improve incrementally, but there are only so many places where the greater precision makes a difference (it makes zero difference on a final mix bus for example), so it needs more refined algorithms and psycho-acoustic research.

[Johnny B]

Jon,

I agree with much of what you said.

As far as the Power PC vs. Intel/AMD...emmm...I don't really wanna go there...

I like it when you can avoid Disk i/o...more memory helps with that.

I like big fast bus structures...there's a lot to like about 64-bit.

I think you will see some advances in the DSP as well.

In fact, I sense we will see some incredible innovations when the 64-bit thing becomes more standard...

One thing we may also see, is "peer review" of some of the code and more Open Source.  

I could be wrong about any or all of this.  I often am wrong, just ask my wife.

[Johnny B]

Jon,

Here's a chip that appears at first blush to be working at 32-bits...I dunno all the details tho' cuz I haven't spent time reading all there is to know about it.

Maybe if you look at it more closely, you can say what is good about it, if anything, and what is bad about it, if anything.

Since this was a year ago, I'd imagine the quantity pricing may now be less than $4.85 per chip in 10K quantities.

http://www.cirrus.com/en/press/releases/P434.html

[Jon Hodgson]

Johnny B wrote on Wed, 05 October 2005 20:09

Jon, Here's a chip that appears at first blush to be working at 32-bits...I dunno all the details tho' cuz I haven't spent time reading all there is to know about it. Maybe if you look at it more closely, you can say what is good about it, if anything, and what is bad about it, if anything. Since this was a year ago, I'd imagine the quantity pricing may now be less than $4.85 per chip in 10K quantities. http://www.cirrus.com/en/press/releases/P434.html

There's not much to say really, it's pretty much what it says on the box.

It's effectively a single use DSP, it does sample rate conversion and that's it. Looking at the numbers it seems it keeps any noise and distortion down in the 24th bit and below, which should make it inaudible, unless you're doing something very silly in your implementation.

If I needed an SRC for a system I'd certainly consider it, and from the numbers given it would appear that the effective gain in quality or performance in this particular application from going to 64 bits would be zero.

[Johnny B]

Where are we, are we talking about the rest of the "food chain?"

I think you agree that 64-bit has a place in the food chain, IIRC?

[Jon Hodgson]

Johnny B wrote on Thu, 06 October 2005 17:40

Where are we, are we talking about the rest of the "food chain?" I think you agree that 64-bit has a place in the food chain, IIRC?

I'm saying that for what this chip does, extending the precision of its processing would not gain you anything tangible, even if there were other places in your system that did benefit from 64 bit processing. Considering doubling the length of the multiplier quadruples its size, I can think of better things to do with those gates and power.

[Johnny B]

Such as?

[Jon Hodgson]

Johnny B wrote on Fri, 07 October 2005 02:25

Such as?

Process four channels at once for the same power consumption?

Leave them out and keep my chip yield up and the price down?

Process the signal more accurately in other places where it is beneficial?

Of course I could just put them in there, extend it to 64 bits, put some marketing pseudo science on the box and charge 10 times the price to suckers who think they can hear the difference, but I've got standards.

[Johnny B]

What kind of tests and experiments would you run?

What kind of proto's would you build?

[Jon Hodgson]

If I wanted to make sure of the numbers then I'd just build a board that let me feed in a digital signal and read another one out and change the settings.

Write a sample rate converter for comparison purposes, hell I might aswell use 80bit floats since I'd probably be writing it on a PC. Feed in a whole load of test signals to the hardware converter and the software one. If the output of the chip and the output of my program differ as little as the datasheet says (140db down on the signal), you're never going to hear it.

If it differs by more than that, then I'll take a closer look.

[Johnny B]

IIRC, someone was just talking about using 80-bit or greater float in an EQ product.

IIRC, the poster thought that partcular EQ was among the best he'd ever used.