Videocardz reports that Nvidia is working on a new high-end graphics card for laptops, based on its GK104 'Kepler' GPU. This chip has 1536 shader units and is quite a bit faster than the GF114 chip that is used in the current high-end notebook cards. It seems that in practice, the performance gap won't be as large. According to Videocardz, half of all SMX clusters will be disabled, resulting in only 768 active shader units. While it is unclear whether the amount of ROPs is the same, rumours indicate that the memory interface remains unaltered.
The chip will be combined with 4 GB of GDDR5 memory, twice the amount of high-end cards like the GeForce GTX 670 and 680. In 3DMark 11, the card is identified as a GeForce GTX 680M and scores 4905 points using the 'Performance' preset. According to Videocardz, this is a 37% improvement over the score of a GeForce GTX 670M. Despite that, the score is still somewhat disappointing as rival AMD's Radeon HD 7970 Mobility allegedly reaches scores of P5800 in 3DMark 11.
A photo of the circuit board has also been released, showing the GPU and eight memory chips. The code on the GK104 chip reveals that it's an N13E-GTX 680M, based on 'A2 silicon'. The image, hard to trace due to its numerous watermarks, does not show that the card will support SLI configurations. It's quite peculiar that Nvidia has disabled so many cores, which makes the GTX 680M even slower than a fully-enabled mid-range AMD card. The Radeon HD 7970M uses a 'Pitcairn' core, derived from the Radeon HD 7800 desktop series.
Although the power consumption of the GTX 680M is not explicitly mentioned, rumour has it that a TDP of about 100W is to be specified. The power requirements of the card completely overshadows the processor's, as a modern Ivy Bridge chip only uses about 45 watts. The card will be introduced during the Computex, early next month.
The GTX 680M
Nvidia's new GTX 680M scores just under 5000 points in 3DMark 11, quite a bit less than the Radeon HD 7970M