bryanb: It isn't a Celeron, though... it's about halfway between a Celeron and a Coppermine. I forgot the numbers, but the cache isn't literally split in half - the associativity of the cache is still P3-level. Celerons have it cut in half.
SL5S puts it in the mobile Celeron family... Doesn't really matter what you call it (Celeron and Pentium are just marketing names for specific product segments).
would x86-64 (Athlon 64) be a good choice for Xbox2? The extra registers would allow for increased performance due to compiler efficiency, while maintaining backwards compatibility with x86 for original Xbox games.
The extra registers are only usable in 64-bit mode and I doubt you're going to double your code's memory footprint just to use 8 extra registers...
The one downfall I'd see would be the integrated memory controller. I don't think MS would be able to take an off-the-shelf PC Athlon 64 and create a compelling gaming solution based around its, say, dual channel PC2700 memory system (atleast for 2005).
Actually that could be a good thing. Better CPU/memory performance would be a boon. A lot would depend on the system architecture as a whole though. If you're going to build the same type of system architecture as the current Xbox, then it might not be so hot. If you segment it in a similar fashion like the Dreamcast then it might be advantageous.
AMD could have been in XBox 1. Intel came into the project late and musculed AMD out with some ridiculously low pricing on those Celeron's that you see in the current XBox. Intel is selling those Celerons at or below cost. The same will happen with XBox 2.
Actually it wasn't Intel coming in late, they were a candidate from the get go, even though AMD participated in the early system development. The Athlon's boom in success caused AMD to renegotiate their offer. They still offered a cheaper price on Athlons, but also wanted additional money up front. Intel made a better offer ( a 733MHz part vs. a 600MHz part for the same price, using depreciated factories, devleping the motherboard, providing debugging tools) in the end. AMD needed higher margin sales at that time, that 'Death Star' fab (Fab 30) in Dresden was still spinning up, and Intel (even though they hate selling cheap, low-margin parts) had the capacity to spare...
I would expect that GS3 will be roughly equally outclassed by the NV5X as the GS2 is outclassed by the NV2A.
I'd like to see how the NV2A outclasses the GS2 since there isn't a GS2 out...
IIRC a few developers said that the Gekko can pull just as much work as the EE's r5900i (model number right?) and VU0...
In terms of scalar performance it beats the EEcore quite handily. However it largely depends on the work you're doing. The one gotcha with Gekko is you can't run vector ops in parallel with the scalar core.
If Sony is beating its chest about the CPU power in the PS3, one can only assume that again, the T&L logic will be decoupled from the graphics chip.
Is there some law of physics that I don't know about that's going to make this a forgon conclusion (and is such necessarily bad)?
I don't think its logical to assume that if SCE goes with this CELL wackiness, that it'll be as nearly as happy to program for as an NVIDIA platform.
Hmmm... The NV2A is grossly parrallel, must to a total bitch to program for...