|
Brief acquaintance with fermi architecture , used in GF100 video chip, assert that the new graphic processor NVIDIA in some degree is oriented to the market for calculation acceleration , whose capacity reaches $1,1 billion per year. If we speak about the market for discrete graphical solutions, then its capacity is measured by one-and-one-half two billion US dollars per year.
From the forecasts of independent analysts, it becomes known that 512 stream processors of the new NVIDIA card can operate at the frequency of approximately 1.5 GHz. Such characteristics ensure speed at the level of 1,5 trl operations with floating point per second. The cost of a system with the speed level 1 teraflops several years ago could reach $10 000. In this case, the fermi architecture allows to create supers-computer, which are ten times faster than the existing now. Fermi supports up to 6 Gb GDDR- 5 memories, which use 384- bit bus. Analysts assume that the frequency of utilized memory chips will reach 2 GHz (4 GHz QDR). This will allow to exceed GT200 on the memory speed , even with the slower memory bus from 512 to 384 bits.
NVIDIA reported about the prospects of using fermi architecture in super-computer, which will build the national laboratory Of oak ridge. This computing system will be ten times faster than the existing supers-computer. Moreover, graphic processors NVIDIA in it will be combined with central processors. The super-computer oak ridge will be occupied by calculations in the energy sphere, and it will help to study the problem of climate variation. The architecture in the next ten years proposed will allow to reach the speed of more than thousand petaflops, which corresponds to millions of trillions of operations with floating point per second. Related Products :
|