A few features have finally be put to the blank face that has only been known as the GT300. A few key details of the upcoming NVIDIA generation of video cards have been leaked, the first of which is a name. Codenamed Fermi, after the developer of the first nuclear power reactor, Enrico Fermi, it appears to be quite suiting when looking at the rumoured specs.
Sources say that the card will be build on the 40nm manufacturing process and feature 3 billion transistors, a 384-bit memory interface running on approximately 1.5GB (3GB and 6GB versions also rumoured) of GDDR5 memory. Taking GPU computing in a new direction, NVIDIA is redeveloping ”shader cores” into CUDA cores, of which the card will feature 512. There are 32 of CUDA cores packed into each of the described “Shader Clusters” for a total of 16 clusters. The beauty of this new design is that these clusters are able to operate independently and balance load on calculations.
In another unique approach, they have added 1MB of L1 Cache onto the card, and divided it into 16KB chunks along with 786KB of unified L2 cache. NVIDIA is clearly pushing to embrace the video card as the new power house of the computer, and develop its parallel computing technology in mainstream video cards while still maintaining dominance in the visual graphics.
The icing on the cake is the native language and API support for CUDA (C), C++, Fortran, DirectCompute 11, DirectX 11, OpenGL 3.1, and OpenCL.
NVIDIA’s GTC conference is being held today, and we look forward to updating you later in the day as we hear more. It is rumoured that NVIDIA CEO Jensen Huang will be debuting a working model of the GT300 during his keynote.
For more technical details on the GT300 visit the Bright Side of News