Thanks to BSN, the specifications for the NVIDIA GT300 or “Fermi” platform has been unveiled. It turns out NVIDIA has dropped the GPGPU and has gone for the kill. The GT300’s will be a cGPU, which is a type of hybrid GPU that combines the number-crunching abilities of the CPU, all on the GPU PCB.

Fermi was named after Italian physicist Enrico Fermi, who invented the first nuclear reactor. Reactor happens to be one of the codenames for the GT300 series.

Just as expected, the GT300 series will buff GDDR5. Unexpectedly, we will see this configurations in 1.5, 3, and 6GB of vRAM. Now here are the specifications, courtesy of BSN:

• 3.0 billion transistors
• 40nm TSMC
• 384-bit memory interface
• 512 shader cores [renamed into CUDA Cores]
• 32 CUDA cores per Shader Cluster
• 1MB L1 cache memory [divided into 16KB Cache - Shared Memory]
• 768KB L2 unified cache memory
• Up to 6GB GDDR5 memory
• Half Speed IEEE 754 Double Precision

Packing around 3 billion transistors. A little more than double that of the GT200 series (1.4 billion). In the die you will find 16 multi-core processors (new name for shader cores), with each packing 32 cores, that gives you a total of 512 cores. Once again, TSMC has manufactured the beast for NVIDIA, and just like the G80 platform, the GT300 will pack six 64-bit memory controllers. Giving you a 384-bit memory interface that is GDDR5 natively. A native GDDR5 memory controller means ECC (Error Correcting Code). Which may prove boastful in corporate uses.

Later, we can expect two additional memory interfaces. However, we will have to wait until the 28 or 22nm shrinks. The GT300 Fermi architecture will support CUDA, C++, DirectCompute, DirectX 11, Fortran, OpenCL, OpenGL 3.1, and OpenGL 3.2. The first of it’s kind to support C++ natively, without loss in performance. We can expect the C++ compatibility to be a major competitor to Intel’s Larrabee architecture. Also, just as we posted January this year in NVIDIA’s roadmap, the GT300 series will use CUDA 3.0 (Compute Unified Device Architecture). A type of parallel processing platform.

The line-up will consist of the usual consumer levels. High-end/Enthusiast GeForce, the commercial series Quadro, and the scientific Tesla. The GeForce series is expected to buff up to 1.5GB, while the Quadro is expected to use up to 6GB.