Technical data / specifications
GeForce GTX 680 is NVIDIAs new flagship model regarding their GPU portfolio.
This card is based on the GK104 chip which is being manufactured using TSMCs new
28 nanometer process technology. At a first glance you can see that NVIDIA has
changed the logo as well as the design of the cooler. Generally we like it.
Looking at cock speeds we see that the GPU runs at 1006 MHz and the GDDR5
memory clocks at 6000 MHz. Furhtermore there is 256 bit memory interface we
results in a XXXX Megabyte per second memory bandwidth. In total there is 2'048
Megabyte of GDDR5 memory.
The GPU itself consists of 3.54 billion transistors. The architectural
hierarchy starts with a so called GigaThread Engine, which marshals processed
and unprocessd information between the GPU and the PCI-Express 3.0 system
interface. Further down in the hierarchy there are four Graphics Processing
Clusters (GPCs), which can be seen as a common resource. At this point you'll
also find a the raster eingine and two streaming multiprocessors (SMs). Altough
NVIDIA decided to change their name which means that they're now called SMX. A
closer look at the SMX shows that they support a new PolyMorph 2.0 engine, an
instruction cache, 192 CUDA cores and furhter first level caches. If you do the
maths based on four GPCs with two SMXs each, 16 SMXs, 192 CUDA Cores, we get a
total of 1536 CUDA cores. This is three times as much as you can find on a
GeForce GTX 580, which supported "only" 512 CUDA Cores. There are also four
raster units, which means that there are 32 ROPs, eight geometry units, whereby
each features with a tessellation unit as well as third level cache.
Discuss this article in the forums
||GeFroce GTX 680
||GeForce GTX 580
||Radeon HD 7970
||Radeon HD 7950
||2'048 MB GDDR5
||1'536 MB GDDR5
||3'072 MB GDDR5
||3'072 MB GDDR5
|Maximum board power