Fermi is Nvidia's first foray into specifically building supercomputers. It's thus more than just a competitor to the 5000 series, which had predictable graphics performance.
Nvidia widened its market focus, so you should widen your perspective.
Like back in the day when nVidia bad mouthed tessellation because other brand could do it & they could not. Or when nVidia essentially talked MS from leaving tessellation out of DX10.
Or the in ability of one nVidia card to power three monitors. The power usage of just powering two monitors is astounding.
GPGPUs are nice and all and nVidia is ahead of the curve with Cuda. But OpenCL is coming and AMD & probably Intel will support that open standard
PhysX isn't used by many games let alone good ones. AMD is actively working with Havoc & Bullet so the other more open & more used physics engines will have very good ATI support. Meanwhile nVidia blocks PhysX hardware accelration if you have an ATI card present.
But all of this means nothing to normal people who just want to play a game on an average level PC.
They won't even be able to run a GTX 400 series without a new PSU & stronger cooling options.