Exophase said:
This should also gauge what a quick and dirty port of it would be like (since we already know this will be weaker than Gamecube, especially Wii, it'd be worse)
Additionally both N64 and GameCube are big endian, whereas presumably this new machine will be little endian and may have to do some more bit juggling to compensate.
I'm not sure the faster Wii speed is relevant at the moment, as a Wii in GameCube mode runs at the same speed as a GameCube (judging by Quake timedemos).
I've found the GameCube to run surprisingly fast, but I'm not sure whether that's down to GCC generating/optimising code well or the machine just being plain beefy. To be honest I've not looked at much of the code GCC has generated for Quake.
Oh, okay, so Gamecube speed all around. I'm not really sure how the Gamecube's CPU compares to an ARM Cortex-A8's, I think they both have about the same in terms of execution units, but Gekko has out of order execution. Cortex-A8 should be clocked higher, as well as having a denser instruction set. Both have comparable L1 cache. Don't know what the OMAP3430 has for L2 cache (Gamecube has 256KB but it's only 2-way set associative which is pretty bad for a unified cache). NEON might be more powerful at crunching floating point values, since its 128bit instead of 64bit.
So it's actually kinda hard to determine which CPU is more powerful. With some very well scheduled code they might be about even.
Of course Gekko would probably mop the floor with i.MX31's CPU.
Probably close to equal...
Gekko at 485 MHz: 1125 DMIPS
Cortex-A8: "over 2.0 DMIPS per MHz" which at 550MHz would mean ~1100+ DMIPS