Sorry, that is bogus. You cannot have AltiVec units independant of the PowerPC processor! The AltiVec unit is like the floating point unit, it can do nothing by itself. Think of it as a supercharger on a car's engine -- by itself it does nothing, it doesn't even spin and if it did spin its output couldn't be used for anything.
Even if they did add AltiVec units, it wouldn't do any good!!! I keep telling you, the current AltiVec units are heavily memory bound and adding more units won't make the system any faster.
As I said above, the DSP strategy has been tried before and the problem is that it requires arcane, specialized, and hard-to-write code. There were rumours of Apple adopting the TriMedia chips from Philips to fill this role before the G4 arrived, but then AltiVec made that a moot point. The advantage of AltiVec is that it is just like having a bunch of special functions on the PowerPC. With a DSP you have to write whole programs for the DSP, and they are weird machines that most programmers don't understand how to write code for.
An MPEG2 encoding chip (or add-on to the mobo chipset) is certainly possible, and since the MPEG2 encoder is a QuickTime component Apple could just update QuickTime to take advantage of it. How often do people encode MPEG2, and since the fast G4s can already do it in real-time (or near real-time) does it make sense to dedicate special purpose hardware to it? Consider too that this will be a new hardware feature in machines which will have faster-than-current processors.
Adding any piece of hardware has a cost, even if it is only an opportunity cost -- if Apple has room for another 10 million transistors on its chipset, and they don't use it for MPEG2 encoding, what else could they use it for? They can already encode MPEG2, shouldn't they spent their development effort and transistor budget on something new?
<strong>Sorry, that is bogus. You cannot have AltiVec units independant of the PowerPC processor! The AltiVec unit is like the floating point unit, it can do nothing by itself. Think of it as a supercharger on a car's engine -- by itself it does nothing, it doesn't even spin and if it did spin its output couldn't be used for anything.</strong><hr></blockquote>
Wasn't there a time when FPUs used to separate chips - math co-processors - and the Intel 80486 just a 80386 with one integrated?
It was also my concern that a DSP would be hard to code for. But isn't there a way that Apple can integrated some sort of middle ware into the OS that takes care of this issue automatically - like your MPEG2 encoding <-> Quicktime example indicates?
If the altivec can't be used without a processor, couldn't apple make a secondary processor, a GPU, which had altivec units built in, for purposes of rendering and encoding. On the other hand, all what we could see is a new multimedia chip, one that takes sound of the main processor and it turn processes some aspects of 3d like rendering etc. Is this possible, I doubt this very much, but hey, it would be good if apple could pull it off.
Wasn't there a time when FPUs used to separate chips - math co-processors - and the Intel 80486 just a 80386 with one integrated?
</strong><hr></blockquote>
Yes, but that was long ago when clock rates and floating point performance were much less. And the chip was tightly coupled to the CPU, and the CPU was designed to support coprocessors. None of this is true now.
[quote]<strong>
It was also my concern that a DSP would be hard to code for. But isn't there a way that Apple can integrated some sort of middle ware into the OS that takes care of this issue automatically - like your MPEG2 encoding <-> Quicktime example indicates?</strong><hr></blockquote>
They can't do that for AltiVec code because this code is composed of instructions in the PowerPC's instruction stream, and registers in the processor.
Middleware-into-hardware only works for APIs in the system, not at the much lower level of machine instructions. Accelerating QuickTime, Quartz, the audio system, Java, cryptography, etc would all be possibilities. Or something new. The trick is finding something that is of benefit to enough of the users and provides them with a capability they didn't already have.
"Yes, but that was long ago when clock rates and floating point performance were much less. And the chip was tightly coupled to the CPU, and the CPU was designed to support coprocessors. None of this is true now."
Wouldn't a 'multicore' G4 or G5 design effectively revive the 'co-processor' design?
I don't think you get his point. FCP doesn't contain code to encode an MPEG2 file - it asks QT to do it, and if QT is written to take advantage of a new chip, FCP doesn't have to be recoded.
</strong><hr></blockquote>
Actually, this isn't true - Final Cut Pro has tons of code to do all of it's different features.. it might use QT for some stuff, but the majority of it is custom code.
You'll notice that FCP came out with new versions, with new features, with no new version of QT.
As for all that Raycer stuff.. who knows? Maybe we'll get some stuff at WWDC this year (the one year I'm not going!)..
Wouldn't a 'multicore' G4 or G5 design effectively revive the 'co-processor' design?
I'd like to see a couple 'in there'...
</strong><hr></blockquote>
A multi-core design is tightly coupled, most likely on the same chip, at the very least in the same chip package (a la IBM's POWER3/4 monsters). The Apple hardware in question would be in the motherboard chipset, not integrated into the processor. Add-ons to the processor design are part of the direction that IBM & Motorola are both taking the PowerPC family (and it holds considerable promise), but its not what was stated at the top of this thread.
Comments
Even if they did add AltiVec units, it wouldn't do any good!!! I keep telling you, the current AltiVec units are heavily memory bound and adding more units won't make the system any faster.
As I said above, the DSP strategy has been tried before and the problem is that it requires arcane, specialized, and hard-to-write code. There were rumours of Apple adopting the TriMedia chips from Philips to fill this role before the G4 arrived, but then AltiVec made that a moot point. The advantage of AltiVec is that it is just like having a bunch of special functions on the PowerPC. With a DSP you have to write whole programs for the DSP, and they are weird machines that most programmers don't understand how to write code for.
An MPEG2 encoding chip (or add-on to the mobo chipset) is certainly possible, and since the MPEG2 encoder is a QuickTime component Apple could just update QuickTime to take advantage of it. How often do people encode MPEG2, and since the fast G4s can already do it in real-time (or near real-time) does it make sense to dedicate special purpose hardware to it? Consider too that this will be a new hardware feature in machines which will have faster-than-current processors.
Adding any piece of hardware has a cost, even if it is only an opportunity cost -- if Apple has room for another 10 million transistors on its chipset, and they don't use it for MPEG2 encoding, what else could they use it for? They can already encode MPEG2, shouldn't they spent their development effort and transistor budget on something new?
<strong>Sorry, that is bogus. You cannot have AltiVec units independant of the PowerPC processor! The AltiVec unit is like the floating point unit, it can do nothing by itself. Think of it as a supercharger on a car's engine -- by itself it does nothing, it doesn't even spin and if it did spin its output couldn't be used for anything.</strong><hr></blockquote>
Wasn't there a time when FPUs used to separate chips - math co-processors - and the Intel 80486 just a 80386 with one integrated?
It was also my concern that a DSP would be hard to code for. But isn't there a way that Apple can integrated some sort of middle ware into the OS that takes care of this issue automatically - like your MPEG2 encoding <-> Quicktime example indicates?
<img src="graemlins/hmmm.gif" border="0" alt="[Hmmm]" />
<strong>
Wasn't there a time when FPUs used to separate chips - math co-processors - and the Intel 80486 just a 80386 with one integrated?
</strong><hr></blockquote>
Yes, but that was long ago when clock rates and floating point performance were much less. And the chip was tightly coupled to the CPU, and the CPU was designed to support coprocessors. None of this is true now.
[quote]<strong>
It was also my concern that a DSP would be hard to code for. But isn't there a way that Apple can integrated some sort of middle ware into the OS that takes care of this issue automatically - like your MPEG2 encoding <-> Quicktime example indicates?</strong><hr></blockquote>
They can't do that for AltiVec code because this code is composed of instructions in the PowerPC's instruction stream, and registers in the processor.
Middleware-into-hardware only works for APIs in the system, not at the much lower level of machine instructions. Accelerating QuickTime, Quartz, the audio system, Java, cryptography, etc would all be possibilities. Or something new. The trick is finding something that is of benefit to enough of the users and provides them with a capability they didn't already have.
Wouldn't a 'multicore' G4 or G5 design effectively revive the 'co-processor' design?
I'd like to see a couple 'in there'...
Lemon Bon Bon
<strong>
I don't think you get his point. FCP doesn't contain code to encode an MPEG2 file - it asks QT to do it, and if QT is written to take advantage of a new chip, FCP doesn't have to be recoded.
</strong><hr></blockquote>
Actually, this isn't true - Final Cut Pro has tons of code to do all of it's different features.. it might use QT for some stuff, but the majority of it is custom code.
You'll notice that FCP came out with new versions, with new features, with no new version of QT.
As for all that Raycer stuff.. who knows? Maybe we'll get some stuff at WWDC this year (the one year I'm not going!)..
<strong>
Wouldn't a 'multicore' G4 or G5 design effectively revive the 'co-processor' design?
I'd like to see a couple 'in there'...
</strong><hr></blockquote>
A multi-core design is tightly coupled, most likely on the same chip, at the very least in the same chip package (a la IBM's POWER3/4 monsters). The Apple hardware in question would be in the motherboard chipset, not integrated into the processor. Add-ons to the processor design are part of the direction that IBM & Motorola are both taking the PowerPC family (and it holds considerable promise), but its not what was stated at the top of this thread.