The article author seems to think that the MT/s number is all that you need to tell the performance of a GPU. This hasn't been the case for 8-10 years now. You can't declare it to be faster without benching it, the theoretical numbers don't reveal much any more. (like comparing a 3GHz Intel to AMD CPU; Intel is much faster today at that clock; AMD was much faster in 2005).

Several PowerVR GPUs in modern phone SoCs (Snapdragon and so on) have higher claimed MT/s numbers but don't even approach the PS3.

The PS3 can also offload a lot of rendering and vector tasks onto the Cell SPUs, so the PS3 has higher effective performance. Games are also optimised for that GPU due to mature drivers and a single target for developers; a PC Gefore 7900 which would be similar hardware actually performs a lot worse on PC ports of those games and so would phone apps.

We'd need to see benches and screenshots before we can confirm this, basically. Ars usualy has higher journalistic standards than this; I'm surprised.

i barely understand what you just said :P.

Oohh it's comparable to the RSX... so the title is a little decieving since the best PS3 graphics take advantage of the CELL not the GPU.

PS3's gpu is pretty lackluster, so this isn't surprising. 

While I agree, I do have to say that if similar GPU performance can be found in a mobile device and is good on battery life...at a lower res screen, that GPU will kick major butt.

It's four years old. Merely on the 90nm -> 40nm transition since then, current GPUs should be ~5x faster for the same power use. Maybe even more given the unified shader transition since then and that the 79xx were not market leaders.

Not to mention PS3's version of 7900GT was gimped day 1 by having 128bit memory bus instead of 256 found in standard card.

Oh four sure. That gpu in a handheld would yield some amazing stuff.