We've seen more of the GeForce GTX 680 graphics card than most people probably need, and now we get to learn something about the second Kepler-based adapter(s).
The GeForce GTX 680 is the high-end video card powered by the GK104 GPU (graphics processing unit).
As it happens, the same GPU will be used in the making of the GTX 670 and GTX 670 Ti, a pair of upper mainstream video controllers.
For those wondering why NVIDIA would waste fine GPUs instead of pushing them to the max, the answer is simple: the company actually doesn't.
What is really going on is that not all chips come out of the assembly lines fully functional.
As such, by fusing off the parts that don't work, the GPU can be given a new brand and role.
Really, it isn't all that different from how Intel (and Advanced Micro Devices for that matter) salvage flawed CPUs.
What we have here is a report stating that the GTX 670 Ti will use recycled GK104 unit clocked at around 900 MHz.
Seven SMX clusters will be available, instead of the maximum 8 (192 units per cluster), as well as 112 texture units (TUs), 32 raster units and four graphics processing clusters (GPC).
As for the rest of the card assets, GTX 670 (Ti) will possess 2 GB of GDDR5 VRAM at a frequency of 1 GHz QDR (5 GHz effective) and a memory interface of 256 bits, leading to a video memory bandwidth of about 160 GB/s.
Keep in mind that the clocks are speculation only. NVIDIA may very well settle on different final numbers.
The GeForce GTX 670 and GTX 670 Ti will battle AMD's Radeon HD 7950, and will have prices of about $50-100 lower (38-76 Euro). That means that buyers will have to part with $349-399 (266-304 Euro according to exchange rates). Announcements should happen this May (2012).