StartupTunes

GeForce GTX 680

Except for good GPU’s, NVidia is master of creating a big hype, especially before their launch. AMD should learn something from them. With every new generation they raised a lot of dust, about how great their new product will be, upon which they would present a lot of test, reviews, and thoughts, about their new technology. Now they presented the newest Kepler architecture and with it again the same story.

Kepler or GK104 is NVidia’s GPU, made in 28nm technology. They didn’t want to experiment that much, so they kept the basic principles that they picked up from Fermi. Even though the architecture is the same in the basic level, GK104 is basically a new GPU, a product of fine tuning the Fermi architecture. The architecture is more efficient than the previous ones.

The lower production process has enabled a higher working frequency to be achieved. So the new GTX 680 works on 1 GHz, NVidia didn’t change the number of ROP units, because it seems that 32 is the optimal number of units for a card that uses a 256bit memory bus. The number of texture units is raised to 128, but the most impressive part is the CUDA units that now number 1536. That is exactly four times more than the previous GF114 and three times more than the processor that Kepler is replacing- GF110. All of that is packed in a chip with 3.5 billion transistors, which take a surface of 294mm².

Also they implemented some other tricks. For example their memory works on over 6 GHz and also it can be overclocked on demand. NVidia claim that every company that uses the given references and puts better coolant system can overclock the card up to 1110 MHz. NVidia calls this technology GPU boost and it is totally transparent up to the end user and the operating system.