14 DAY TRIAL // The entry-level GeForce 600 series cards are just rebranded 500 series models, but this may change when NVIDIA finally brings out the GeForce GT 640.
According to CPU World, the Santa Clara, California-based company will bring out the GT 640 during Computex 2012 (June 06-09).
Of course, it would be more accurate to use the plural, since there will be six versions of the GT 640 for some reason.
But here is where things get a bit confusing, as each card has its own clock speeds, CUDA core number, memory capacity and VRAM interface.
If we were to guess, some GeForce GT 640 cards will indeed rely on the GK107 GPU, but some will be powered by rebranded Fermi.
The GT 640 “series” has two entry-level versions, two mid-range models and two top-shelf iterations. Each pair is made up of a retail product and an OEM-exclusive card.
The top shelf ones have the Kepler GK107 GPU with 384 CUDA cores and a clock speed of 950 MHz. 1 GB or 2 GB of memory will back it up (128-bit interface).
The mid-range boards are a curious bunch. The report says they have the same 192-bit interface, 1 / 3 GB VRAM and 720 MHz GPU clock speed, but it says that the OEM card has Fermi (144 cores) and the other Kepler (unknown CUDA core number). We think this might be a mistake, and that both have Fermi chips, but we might as well be wrong.
Finally, the entry-level GT 640 cards are Kepler-powered. The OEM version features 384 CUDA cores, 1 / 2 GB VRAM, 128-bit interface and 797 MHz GPU clock. The retail one has an unspecified CUDA count but otherwise identical specs.
NVIDIA's GT 640 cards will be most popular among HTPC builders, due to their low profile (not all of them are half-sized though) and length of 5.7 inches (144 mm).