The ECC memory protection, which is important for professional applications, was completely omitted and the FP64 hardware was shortened (only 1/3 of the shader are FP64-capable and thereby only 1/12 of the FP32’s performance). In worst case scenarios, the performance of the Fermi architecture can also be worse than expected. As there are still only 2 warp schedulers (versus 3 shader groups), Nvidia now uses superscalar execution in order to utilize the higher amount of shaders per SM more efficiently. In theory, the performance per core should be greatly improved over previous generations. They feature more shaders (3x16 instead of 2x16), more texture units (8 instead of 4) and more SFUs per streaming multi-processor (SM). In contrast to the GF100, which was designed for professional applications, these latter chips target the consumer market. Unlike the GF100, the smaller GF104, GF106, and GF108 cores were not only shortened, but also considerably modified. Therefore, the architecture is not directly comparable to the old GT215 (e.g., GeForce GTS 350M) or GT216 (e.g., GeForce GT 330M) cores. Except for the memory controllers, the GF108 can essentially be considered a halved GF106. The GF108 core of the GT 525M is related to the GF100 core found in the GeFore GTX 480M and offers 96 shaders and a 128 Bit memory bus for DDR3 VRAM.

Download NVEnc 7.25