OT: RE: [rescue] XBOX vs PS2

[Joshua D Boyd]

On Tue, Jan 22, 2002 at 07:28:44PM -0800, Chris Kennedy wrote:
> > > I'll stick with my PS2.  A nice R5600/200 or so Mhz MIPS processor,
> > > lots of memory IO...and TWO vector processing engines.
> 
> Not R5600.  The CPU2 core is the TX5900, which executes an extended 
> R5000 instruction set.  It's called a 128 bit processor, but 128
> bit instructions lock step the two 64-bit integer pipes, which 
> puts a damper on multiple issue; other weirdness includes on-board
> scratch pad RAM (a page worth) that is a DMA _target_, introducing
> the somewhat weird concept of doing DMA _to_ the processor.
> 
> It's true that there are two vector engines, although they differ in 
> the number of floating point units and hardware visibility (one
> can see the GPU2 chip, the other can't).  Both share a slightly
> brain-damaged implementation of single-precision IEEE floating
> point (as does the FPU on the 5900 core).

IMO, for games, and most things in general, 32bit floats are good enough.
For scientific work, obviously not as good as 64bit floats, but still, 
nobody is buying a PS2 for scientific work, and for the most part, people
aren't buying R5ks for scientific work either.

> Well, actually, the poor bastards who have to cobble together bits
> and pieces of code by hand care.  Game processors are weird
> beasts; you can't do a process change, die shrink or anything else
> that might affect timing once the thing ships because these guys
> still do some hand-scheduling of instructions. I personally think
> the x86 ISA is horrid and I'm certainly no fan of the TX5900 ISA
> either, but at least with the TX5900 I _can_ schedule things so
> I have no (or minimal) slips and stalls.  While the same can be
> said for an x86 implementation it's harder to predict because the
> ISA is more opaque with respect to underlying hardware state.

Those people seem to complain about the PS2 an awefull lot.  I would
welcome the challenge of working on such a beast, but I'm pretty
unqualified.

> The fact that the memory cards for the PS2 go for $40 is definitive
> proof that game consol hardware is given away.  Why SCE insisted
> that we use RAMBUS for the TX5900 -- and why we had to support both
> the sync and async specs -- remains a great puzzlement.
> 
> > The graphics look to be about the same to me on 
> > both platforms but PS2 has had much more time to mature.
> 
> The graphics appear to be a wash, but given that people 
> were still finding new ways to use PS1 hardware as recently
> as three years ago I remain unconvinced that the PS2 set has
> even scratched the surface on what to do with all of those
> vector units.  The same may or may not be said for the "modified"
> PIII that lurks in the Xbox.  Time will tell :-)

I doubt people will find many better ways to tweak out the P3, but
the shader system that the graphics system has will probably be 
producing tweaks for some time.  Still, this isn't as flexible as 
the PS2s to my understanding.  But, I don't have thing official on 
the machine, only what I've managed to scavenge.

Which reminds me, I still need to be seeing about getting an AR Pro for
my PS1 so that I can try my hand hacking code for it.

-- 
Joshua D. Boyd