When introducing every new GPU generation traditionally we go through a repetition of the grade, when talking about the basics of the business philosophy of the whole industry, which means that with the process of “harvesting” you need to use as much as you can from the manufactured chips. This process is crucial for keeping a commercial success when dealing with GPU’s, which have amazing number of transistors implemented inside- between 3.5 and 4.5 billion. Depending on the complexity of the design and improvement of the manufacturing process, the number of perfectly made processors can vary more or less, and it is left to the manufacturer to find a golden middle on which to base the final chip, which will come out of the original chip.
The graphical chip that we are reviewing came out of the chip that goes into the GTX 680 named GK104 and through a process of harvesting this imperfect chip became GeForce GTX 670.
Let’s start with the details. Instead of 1536 CUDA processors which are implemented in the GTX 680, the GTX 670 has 1344. This is not a big difference. The difference is small also in the number of textural units- 128 in the GTX 680 and 112 in the GTX 670. The number of ROP units is the same-32. It works on the same RAM frequency of 6 GHZ and also it is a 256 bit, as well as having a 2GB of memory. Obviously the difference between the two is minute.
When doing the synthetic tests both give out almost the same performance statistics and still the GTX 670 is much cheaper. This makes the GTX 670 the preferable choice and this is the step that NVidia was expecting thus expanding the market and getting new revenue by doing recycling.