Iron Tiger said:
Actually I have.
PSP
PVR MBX
That's not to say I believe Sony's far-fetched performance numbers. But going on what I've seen of the PSP (around DC quality from the best looking titles) and what I've seen from the MBX (the Lite version in my Dell Axim X50v), I'd say the PSP is defintitely packing more gaming power.
Leaving aside issues of console v PDA system optimisations:
The PVR site implies that the Lite version has less than half the polygon throughput of regular MBX, and Lite has half the fill rate of regular MBX for the same clock speed. Fill rate scales linearly with clock speed, polygon rate doesn't.
So, comparing the Texas Instruments
OMAP2, using regular MBX at 200MHz, to the Intel
2700G, using MBX Lite at 50-75MHz and part of the Axim X50v ...
The 2700G datasheet states 831K triangles per second (tps) processing capability, 84M pixels per second (pps) fill rate at somewhere between 50 and 75Mhz. OMAP2 gives polygon throughput at "up to 2 million". I can't see fill rates for OMAP2, but 500-600 Mp/s seems plausible. I can't see power consumption either.
The polygon throughputs for PSP in the GameSpot article seem a little fanciful. Maybe if the hardware did nothing but generate unfilled triangles in one big strip. Fill rate numbers are believable - and perhaps comparable to raw OMAP2 fill rate. How would they compare when we factor in PVR's pixel-perfect HSR? Does PSP use any overdraw reduction technique?
Comparing power consumption of OMAP2 to PSP's array of processors could be interesting too.
This Khronos presentation from Feb 2004 gives an example of MBX+VGP, with
-Sustained 2.5M triangles/second
-Sustained 240M peak 480M pixels/sec, Single-texture, bilinear MIP-map, 2x1 FSAA
It doesn't give clock speeds - from the date, I'm guessing 120MHz on 0.13um.
Anyway, I seem to have convinced myself that MBX "effective" fill rate absolutely murders PSP fill rate, when on the same process. The Register (stop sniggering at the back)
reports that N-Gage 2 is expected to be OMAP2 powered. Hmm...