No it isn't wrong, it's just not representative of what you would see in a game. It's a measurement of fillrate, NOT the ability of the platform to provide transformation + lighting. It's simple math: PS2's Graphics Synthesizer is clocked at 147MHz. It has 16 pixel pipelines. If you aren't using texturing then it can actually produce 16 pixels on the framebuffer, per cycle (the VRAM is fast enough to back this up too). Therefore you get 147 * 16 = 2352 million pixels per second. The polygons Sony is talking about are 32 pixel triangles (in other words, they're tiny). 2352 / 32 = 73.5 million of these triangles per second. Sony rounded the 2352 up to 2.4Gpixels/s fillrate. So are you asking me if a game has anything like 70 million triangles per second going on? Most games perform lighting and transformation on VU1 which is 150MHz. Sony says that it can do 3.08 single-precision gigaFLOPs per second. Does this add up? The VUs have 9 FPMACs and 1 FP divider unit. 3.08 gigaFLOPs at 150MHz means about 20.5 FLOPs per cycle. With all of the operational units that sounds feasible, I don't know what other pipeline restrictions it has that brings the number down to what it is, so we'll take it at face value. Now how many FLOPs are required per triangle? It's pretty complicated and has several variables. I'm not going to pretend to get something JUST right but let's try a very rough estimate. First the hard part: lighting. Look here: http://en.wikipedia.org/wiki/Phong_reflection_model This is the standard shading model used in Gouraud shading. You'll see that there are two dot products and 4 multiply/accumulates per light, and one multiply/accumulate additionally. A dot product for a color is 3 multiply/accumulates. So that's 10 multiply/accumulates per light, and 1 additionally. Let's say there's just one light (sucks but okay, whatever), then there are 11 FLOPs per vertex for lighting. Then transformation is a matrix + vector multiply and a vector division. Let's say we're dealing with 3x3 and 1x3 matrices and vertices respectively (homogeneous coordinate system). Then that's 3 dot products for the operation, and 3 fpdivs (they'd actually be reciprocal multiplications or something). So 12 FLOPs. Add them up, you get bare minimum 23 FLOPs per vertex. So you can do 134 million vertices per second on the VU1. But that's assuming all of those operations are worth the same common small amount. How many triangles does this equal? I really don't know what the vertex to triangle ratios for typical meshes are. Is it enough to feed 75 million? Since you have to be flat shading them that means one color per triangle, so probably. According to the Wiki node it gets this: "3D CG Geometric transformation: 66 million polygons/sec" Which is roughly in the ballpark (honestly, I don't know what goes into that computation) Now do any games actually reach this amount? I doubt it, because they do standard things like Gouraud shading and texturing which cuts your fill rate down a lot due to extra requirements on the pixel units and comparatively limited port space for texturing on VRAM. And this was only for tiny polygons to begin with - do you think most games have on average 32 pixel triangles? I doubt that too.