October 1, 2009 at 02:26nVidia GT300 Fermi Specifications
NVidia has published detailed specifications for the new GT300 chip, code-named Fermi . The GPU will be executed, as expected, using the 40-nm process technology and will contain 3.2 billion transistors, which, by the way, is much more than 2.15 billion transistors in the AMD RV870. Accordingly, Fermi is larger in size than its competitors and, more importantly, more expensive to manufacture.
The total number of GT300 shader cores is 512, 32 CUDA cores for each of the 16 stream multiprocessors. Each individual multiprocessor, by the way, in each thread is capable of performing an operation with an integer or a floating point number in one clock cycle. Other features: 384-bit memory bus, 1 MB cache in the first level, 768 KB of unified cache in the second level and the ability to use up to 6 GB of GDDR5 memory.
In terms of technology, Fermi hardware supports CUDA, C ++, DirectCompute, DirectX 11, Fortran, OpenCL, OpenGL 3.1 and OpenGL 3.2. The manufacturer notes that the GT300 is the first-ever graphics processor that can execute C ++ code without sacrificing performance.
The GT300-based product line will include GeForce, Quadro, and Tesla (“scientific”) accelerators. The cost, unfortunately, is not yet known. One can only hope that it will be reasonable. But in general, judging by the dry numbers of technical specifications, nVidia's answer to the successful AMD RV870 and its implementation in Radeon 5870 was a success.
PS In a year, when the price of cards with this chip drops to an acceptable level, they will become very good candidates in choosing an upgrade for mere mortals.
via hothardware
The total number of GT300 shader cores is 512, 32 CUDA cores for each of the 16 stream multiprocessors. Each individual multiprocessor, by the way, in each thread is capable of performing an operation with an integer or a floating point number in one clock cycle. Other features: 384-bit memory bus, 1 MB cache in the first level, 768 KB of unified cache in the second level and the ability to use up to 6 GB of GDDR5 memory.
In terms of technology, Fermi hardware supports CUDA, C ++, DirectCompute, DirectX 11, Fortran, OpenCL, OpenGL 3.1 and OpenGL 3.2. The manufacturer notes that the GT300 is the first-ever graphics processor that can execute C ++ code without sacrificing performance.
The GT300-based product line will include GeForce, Quadro, and Tesla (“scientific”) accelerators. The cost, unfortunately, is not yet known. One can only hope that it will be reasonable. But in general, judging by the dry numbers of technical specifications, nVidia's answer to the successful AMD RV870 and its implementation in Radeon 5870 was a success.
PS In a year, when the price of cards with this chip drops to an acceptable level, they will become very good candidates in choosing an upgrade for mere mortals.
via hothardware