Jawed
Legend
It might be useful to turn off SSAA while you are trying to identify the apparent bottlenecks. I don't know what the 2x SSAA pattern is on G80, so I don't know the dimensions of the resulting underlying render targets it's working with.As i said, i'm working in the BR2 HD project, and i'm remaking the textures of the game in high resolution. Everything is going fine, but i'm concerned about the performance.
I use a 8800GTX (615/1500/2200) in my rig, and i use this config:
1280x1024 SSAA 2x AF 16x
I think the performance loss is simply down to bandwidth, as the larger textures will consume proportionally more of it, and cache usage, since the increased number of mipmap levels may be causing some (or increasing) cache thrashing.1) Texture Filtering: Quality -> No changes
2) Texture Filtering: Quality, Tri. Optimization: On -> 90 FPS
3) Texture Filtering: High Quality, Tri. Optimization: Off, AF: 2x -> 90 FPS
I suspect that i'm bounded by the TF units, cause, the trilinear optimization reduces the area in which the mipmaps are filtered trilinearly (reducing the number of texels to read & filter), and reducing the AF level also reduces the number of texels to read & filter.
The bandwidth cut, down to ~83%, is pretty significant and it seems to be dominant.I also did this experiment with the default textures:
Memory: 733 MHz (733MHz * 2 (DDR) * 384bit / 8 = 70 GB/s
Shader Clock: 1650 MHz (yes, it was stable).
Core Clock: 615 MHz (650 wasn't stable).
Basically i clocked my card like a 9800 GTX, to discover if the game is bounded by the memory bandwidth or by the shading power, and the frame rate went from 100 FPS to 80 FPS (20% slower).
Because G92 has extra TAs in its TMUs there's a chance that it will consume texels (and therefore bandwidth) in a different ordering, which might make it more (or less) cache friendly. It would make an interesting comparison. There hasn't been much analysis of the difference in performance of G92 and G80 with older games.
Jawed