The new GPUs supported by the 304.64 driver is the VGX K1 and VGX K2. The VGX boards are Kepler-based and designed for GPU virtualization, low-latency remote displays, multi-GPU designs, and other enterprise-grade requirements. The VGX K1 packs four low-end Kepler GPUs with 768 CUDA cores and 16GB of GDDR3 video memory while the VGX K2 packs two high-end Kepler GPUs with a total of 3072 CUDA cores and 8GB of GDDR5 video memory.
The 304.64 driver also adds in a missing 32-bit compatibility library for the libnvidia-opencl.so shared library for 64-bit Linux, fixes a backlight control regression for some notebooks, corrects a performance issue with recent Linux kernels about allocating and freeing system memory, fixes an issue with nvidia-settings, takes care of an X driver gamma manipulation bug after VT switching, adds an "--output-file" option to the nvidia-bug-report.sh script, fixes a hang for some OpenGL programs when using the SLI Mosaic mode, and updates the supported GPU documentation for NVS/Tesla/VGX products.
The NVIDIA 304.64 driver can be downloaded for x86/x86_64 Linux from NVIDIA.com. The 304.64 driver is a "certified" release, but for those wanting something a bit more bleeding edge, there is the NVIDIA 310 Linux beta driver.