Speaking of performance, we’ll get to that in a bit, but for now we’ll finish our discussion of GF100’s graphics units. Each SM has 4 texture units, which have changes somewhat from the GT200. In GT200, the texture units were at the TPC level; here they are per SM. The texture units are also affected by the cache reorganization of GF100; for L1 cache they still have the same 12KB of texture cache per set of 4 texture units, while they all share GF100’s 768KB L2 cache, which is now used by all execution units and not just textures. Each unit can compute 1 texture address and fetch 4 texture samples per clock, a significantly higher texture fetch rate than on the GT200. Furthermore in conjunction with DX11, NVIDIA’s texture units now support DX11’s new compressed texture formats. Finally, texture units are now tied to the shader clock and not the core clock. They run at 1/2 the shader clock speed of GF100.