The latter manages 797-915 MHz under load, which is still decent. 1,029-1,097 MHz during 3D operation is significantly higher than the Kepler version. The Maxwell chip is particularly superior in terms of clocks. This is even superior to the GTX 770M (960 CUDA cores). Compared to the GTX 765M, the Kepler GTX 860M gets 50% more CUDA cores, more precisely 1,152 instead of 768.
Thanks to the huge efficiency improvement, 640 CUDA cores are sufficient for the Maxwell GTX 860M to compete with the in-house rivals. Much more important is the number of shader units. It does not really matter if Nvidia equips the graphics card with 2 or 4 GB memory based on our experiences. Other similarities of the two cards include the memory interface: The GTX 860M uses modern GDDR5 video memory, but a 128-bit interface is not up to date anymore in 2014. Smaller notebooks use the soldered GPUs with the Maxwell chip GM107, whereas bulkier gaming cases with MXM slots use the significantly bigger Kepler sibling (GK104 chip), which is also manufactured in a 28 nm process. It is easy to understand why Nvidia produces two versions of the GeForce GTX 860M. We compared the performance of the two DirectX 11 chips and found a winner. Besides a version based on the Maxwell architecture, there is also a GPU from the old Kepler generation.
Nvidia offers two different versions of its latest high-end GPU GeForce GTX 860M. Review Nvidia GeForce GTX 860M Maxwell vs.