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.