Theo Valich from Bright Side of News has gathered lots of details about NVIDIA's GT300, which is called Fermi by NVIDIA insiders. Enrico Fermi was an Italian physicist credited with the invention of the nuclear reactor, and the GT300 board is reportedly nicknamed "reactor".
The GT300 is another major architectural change and if Bright Side of News is right the chip makes a big leap towards becoming some sort of GPU/CPU hybrid. NVIDIA's upcoming GT300 architecture will be made on a 40nm process by TSMC and reportedly has 3 billion transistors. BSN says you can expect 512 shader cores, configured into 16 Streaming Multiprocessors with 32 cores each. There's a 384-bit memory bus and support for up to 6GB GDDR5 memory with ECC. Consumer cards will likely see memory capacities of up to 1.5GB, while Quadro and Tesla parts can expect up to 6GB of GDDR5 memory.
Furthermore, the GT300 is expected to have 1MB L1 cache memory and 768KB L2 unified cache memory. The GT300 takes NVIDIA another step closer towards their GPGPU computing dream, BSN claims the chip support the latest IEEE 754-2008 floating-point standard and that there's native support for C [CUDA], C++, DirectCompute, DirectX 11, Fortran, OpenCL, OpenGL 3.1 and OpenGL 3.2.
The interesting bit is the type of IEEE formats. In the past, nVidia supported IEEE 754-1985 floating point arithmetic, but with GT300, nVidia now supports the latest IEEE 754-2008 floating-point standard. Just like expected, GT300 chips will do all industry standards - allegedly with no tricks.
Ferni architecture natively supports C [CUDA], C++, DirectCompute, DirectX 11, Fortran, OpenCL, OpenGL 3.1 and OpenGL 3.2. Now, you've read that correctly - Ferni comes with a support for native execution of C++. For the first time in history, a GPU can run C++ code with no major issues or performance penalties and when you add Fortran or C to that, it is easy to see that GPGPU-wise, nVidia did a huge job. To implement ISA inside the GPU took a lot of bravery, and with GT200 project over and done with, the time came right to launch a chip that would be as flexible as developers wanted, yet affordable.