Can PS2 be considered a pure 128 bit machine??

[Fafalada]

Actually there is a common convention.

Depends who you ask.
From what I remember, the original convention(well at least so were were told in school) goes by size of 'memory word', which happens to usually coincide with size of general purpose register(but EE happilly breaks that), as well as width of the bus in many cases.

Granted, if we followed that particular definition, R5900 is 32bit, but both VU0/1 are 128bit.

address space and data-word size were all 32-bit in all but the more esoteric variants IIRC

68x had some really esoteric variants too - 68008 in QL had 8bit external bus. Actually funny thing is back then people weren't afraid to call something 8/16/32bit hybrid, but with consoles they always got stuck on one number.

How is a 64 bit OS and applications going to speed things up?

Some arithmetic&algorithms can run faster with wider registers.
For instance - 32bit integer multiplies have 64bit results, so code compiled for 64bit would be both shorter and faster then 32bit for such arithmetic.

 

[swaaye]

Suffering thru my infinite ingorance as usual, I was reading an article that said that AMD64 only doubled the # of registers available (to some still paltry #) and that other RISC CPUs have had bunches more registers for a long time. Why didn't AMD scale it up more? Is there some extra cost that these overblown x86 CPUs bring with them there too?

 

[Jacob]

If you don't know what a 128 bit CPU is, how can you know whether EE is 128 bit or not?

With that comment, I was just emphasizing how much people have overused the term "bit" as a way of comparing the overall capabilities of two machines or promoting their system of choice as the better hardware.

PS: My definition doesn't vary much from what some have posted above, but I have yet to see a definitive convention of a X bit CPU that is accepted by everyone (he, you just have to read the replies). I don't know if you disagree with me, but I don't care as I find this particular discussion pointless. What really annoys me is that it seems like every console specific fanatic now comes to this place looking for ammunition. I wish we could go back to more civilized times, but I guess things will calm down once these next generation machines become old news. Hopefully.

 

[thejeek]

Suffering thru my infinite ingorance as usual, I was reading an article that said that AMD64 only doubled the # of registers available (to some still paltry #) and that other RISC CPUs have had bunches more registers for a long time. Why didn't AMD scale it up more? Is there some extra cost that these overblown x86 CPUs bring with them there too?

There are diminishing returns as you add more registers to a machine and an associated cost in terms of the instruction size - AMD did extensive simulations and determined that adding 16 registers would improve the situation considerably compared to ia32 for only a small code size increase.

Ffor position independent ia32 code you sometimes have only 5 general purpose registers available, amd64's 16 registers plus IP relative addressing nearly triples this. AMD were able to add the 8 additional registers whilst adding only a single byte to those instructions that reference them - had they gone with 32 GP registers they would have required two additional instruction bytes. This would have affected code density and consumed additional decode bandwidth.