Nvidia's Fermi GPU architecture

[devpat]

hi
I have saw the news about Nvidia's Fermi GPU architecture but its make so special for graphics or its just another new product which have nvdia like parents ? just enthusiastic to know ....

[Marthaz]

new architecture of NVIDIA Fermi has the number of stream processors, increased from 240 the previous achitetture GT200 to the current 512, NVIDIA has chosen to call these components as CUDA Cores and not as stream processors, but in fact are the same in terms of architecture of the GPU.NVIDIA has incorporated the caching architecture Fermi particularly complex.For each stream there is a microprocessor dedicated cache of 64 Kbytes of capacity, partition as shared memory and L1 cache as: the relationship is 1:3 or 3:1

[ApPy10]

NVIDIA has also improved the management of the GPU switching from graphics mode to that CUDA, reducing by 10 times the time to switch over GT200. The direct benefit is better performance with applications that require many times the transition from one mode to another and vice versa. Fermi GPUs also allow you to manage parallel transfer to and from the CPU, no longer in a serial fashion as with the previous architectures.

[David23]

Related to graphics processing in these GPUs certainly introduce performance improvements, while leaving unchanged the overall scenario for applications that NVIDIA GPU Computing lays its highest expectations. Fermi may be the turning point for NVIDIA in a market that is extremely interesting and potentially profitable, but which until now has seen the U.S. company dominate even with a volume of sales is not sufficient to justify the investment architectural technology required for a project like this.