zarx on 01 May 2012
| Model | Launch | Transistors (Million) | Memory (MiB) | Clock rate | Config core | Fillrate | Fillrate | Memory | Memory | Memory | GFLOPS |
| Texture (GT/s) | Pixel (GP/s) | type | Bus width (bit) | Bandwidth (GB/s) | |||||||
| RSX | 2006 | 300 | 256 | 550 | 24:8:24:8 | 13.2 | 4.4 | GDDR3 | 128 | 22.4 | 400 |
| GTX580 | 2010 | 3000 | 1536 / 3072 | 772 | 513:04:48 | 49.41 | 37.06 | GDDR5 | 384 | 192.384 | 1581.1 |
On paper the GTX580 is arround 10 RSX's welded together but with Unified shaders, Tesselation, Compute functionality etc which would crush the the RSX utterly in real world performance due to reduced bottlenecks and new hardware functions that would allow the 580 to do things that aren't even possible on the RSX (tesselation, compute functions like Physics accelleration, decompression of assets, advanced SSAO, AA and GI effects, and even supports working with better compression of assets without decompression so textures etc take up less working memory and less time to load). There is really no way to accurately compare the two.
@TheVoxelman on twitter
