At the same time as launching the GeForce GTX 680, NVIDIA extended its GeForce 600M family with models derived from the Fermi generation (GeForce 400/500M) and, more excitingly, models based on a new Kepler generation GPU.
Let's remind ourselves first that the first GeForce 600Ms were unveiled in December and consisted mainly of a renaming of entry level GeForce 500Ms:
To add to these first GeForce 600Ms, NVIDIA has renamed some high end models and has also introduced what is probably the last Fermi generation GPU, the GF117. It has the peculiarity of being engraved at 28nm, which should make it a very low consumption model as it has a very small GPU that basically corresponds to a shrink of the GF108 (identical in terms of numbers of functioning units but engraved at 40nm).
The GeForce GTX 675M is simply a renamed GeForce GTX 580M, which corresponds to a highly underclocked GeForce GTX 560 Ti and probably gives performance more akin to a GeForce GTX 460. The GeForce GTX 670M is a slight development of the GeForce GTX 570M with the maximum GPU clock increasing from 575 to 598 MHz, a gain of 4%.
The GeForce GT 640M LE, in its Fermi version (there’s also another version), is a strange product in that it is similar to one of the variants of the GeForce GT 635M. It offers a slightly higher GPU clock (1%) and a lower memory clock (-13%).
A third [sic] variant of the GT 630M has also been released, this time based on the GF117. It offers more memory bandwidth and 20% more processing power than the second variant based on the GF108 to which the GeForce GT 620M, also based on the GF117, is very similar.
More worthy of a look are the GK107 models. Following the GK104 used in the GeForce GTX 680, the GK107 is the second Kepler GPU engraved at 28nm. They share the same architecture described in detail here
. Basically Kepler is a development of Fermi architecture designed to increase energy yield, notably by halving the clock of the processing units but doubling their number, which means it ends up consuming less energy.
The GeForce GTX 660M is the biggest model running on the GK107, an interesting development in comparison to the GeForce GTX 560M: + 8% processing power and + 60% memory bandwidth. All this comes within a thermal envelope that we imagine is similar or even lower than that on the GTX 560M.
There are other models on the programme, all based on this same GPU. The GT 650M exists in DDR3 and GDDR5 versions. The DDR3 version has a reduced memory bandwidth but a higher GPU clock than that used on the GTX 660M, while the GDDR5 version has the same memory bandwidth as the GTX 660M but a GPU with a lower clock.
Although the GT 640M also exists in DDR3 and GDDR5 versions, only the memory bandwidth is different from one to the other, with the GPU clock and processing power identical. The GT 640M is probably the best mobile compromise for the GK107 as its relatively low clock means it can give a very high performance/watt ratio. Indeed laptops equipped with this last GK107 card are the ones NVIDIA supplied for the first tests and which have been the first to come onto the market, in the Acer Timeline M3 (M3-581TG).
Finally, the GPU on the GT 640M LE is clocked even lower and only exists as a DDR3 version. Note that a second variant based on a Fermi GPU also exists and offers 24% lower processing power but 75% higher memory bandwidth.
Note that the GeForce 600Ms based on the GK107 support PCI Express 3.0 as well as DirectX 11.1, which isn’t the case with the other models. All however support Optimus technology which allows the user to benefit from the reduction in energy consumption afforded by the integrated GPU (part of the CPU) when it kicks in to display the Windows desktop. This technology does however come with a significant limitation: it is incompatible with 3D Vision. Given that NVIDIA uses a proprietary technique to transfer 3D images, an Intel GPU cannot take over for the display of these 3D images.
In the end this new mobile range from NVIDIA leaves us with rather mixed feelings. The new Fermi GPU graphics solutions give little extra in comparison to the GeForce 500M family, do not bring the energy efficiencies boasted of without distinction by NVIDIA for the 600M family and are above all there to create the illusion of innovation. Of course this never gives value for money to the consumer.
The few GeForce 600Ms based on Kepler architecture are the truly innovative products in the family and shouldn't be confused with the other models. They have a definite advantage as they give a significantly higher performance/watt ratio, which is the crucial thing for mobile graphics solutions. The sort of efficiency they afford finally allows us to glimpse the possibility of gaming on an Ultrabook type laptop.
While NVIDIA has got there first on mobile GPUs engraved at 28nm, AMD shouldn’t be far behind. Thus, its Cape Verde GPU, derived from the Radeon HD 7700s, should be capable of competing with the GK107 and Pitcairn, derived from the Radeon HD 7800s, should be able to offer real innovations when it comes to the high-end mobile segment.