In the GPU Technology Conference, Nvidia you reveal certain elements from the new architecture of processors graphic that it will present in the next months. The architectural &? quot? Fermi&? quot? , as announced the CEO of company, Jen-Hsun Huang, it is that better it has presents the industry of processors graphic (GPU) and brings, this complicated units of treatment in a other level.
Substantially, Nvidia made known, that at an early date, it will not present in the market one simple GPU, but cGPU (Computational Graphics Processing Unit) after it supports, and in particular gigenos (natively) beyond CUDA, DirectX 11 (Shader Model 5. 0), OpenGL 3. 1 and 3. 2, OpenCL and Fortran or DirectCompute and - you are kept - C++!
With few, reasons. the new architecture of Nvidia is more near that of processors, from never previously. The possibility of implementation of code C++ (and C, CUDA and Fortran) they undeniably constitute a sonorous blow in the &? quot? Larrabee&? quot? Intel, and they show, that the processors graphic can at last make a lot of, many more things than shoot. frames per second in a screen, even in high definition analysis.
Before few days, the AMD, presented the really exceptional line of cards of graphic ATI Radeon HD 5800, which offers also support in the model ATI Stream, OpenCL, DirectX 11 (Shader Model 5. 0) and DirectCompute. As long as it concerns in the use of processor graphic, as general use of processor as we say, Nvidia, advanced appears a step parapera with the support - in hardware - the language C++.
the AMD with the new line of ATI Radeon HD 5800, presented the new line with basic arm the support in DirectX 11 API of Microsoft - even if it gave big importance and in the support of OpenCL and DirectCompute - but appears that Nvidia, even if naturally it will support the new publication of model, rather wanted it downgrades lightly the particular characteristic. And how it didn’t make him, can somebody say, when it has a GPU in her hands, that can run code C++ in hardware, prosferontas innumerable possibilities in the programmers and in the companies of growth of applications?
new chip that follows the architectural &? quot? Fermi&? quot? (Enrico Fermi, Italian natural that discovered the nuclear reactor, for this and in certain cases, certain &? quot? axiomatoychoi&? quot? Nvidia they were reported in the GT300 chip with code name &? quot? reactor&? quot) it incorporates more from three billions transistors (does the ATI Radeon HD 5870 incorporate roughly 2,15 billions transistors) and will occupy surface lightly smaller - in all probability - from 576mm&? sup2? the GT200, even if it will be manufactured by the TSMC in scale of completion 40nm. Enormous chip for such scale of completion.
In the parapano photograph you can distinguish die the GT300 (Fermi GPU), that allocates in total 512 CUDA processing cores, and that with their line are organised in 16 SM blocks (Streaming Multiprocessors) with each one from them it allocates 32 cores (cores).
Each core from the 32, constitutes a completely pipelined unit of management of action entire and action of mobile comma that supports the most recent model IEEE 754-2008 (which is supported also by the line Radeon HD of 5800 ATI). As long as it concerns in the memory bus, it is constituted from six channels of 64-bit that globally …