With Maxwell, NVIDIA seems to be taking a bottom to top approach by releasing the entry level models first and high-end graphics chips later. We still don’t have any concrete details regarding
the Maxwell architecture to the process these chips would be based on
but NVIDIA seems to be focusing alot towards power efficiency with these
chips on the mobile front and NVIDIA would expect some design wins with
such a chip considering its low power consumption and low temperatures.
The most interesting thing regarding Maxwell would be the Denver CPU which is possibly going to be introduced with the high-end cards. Denver CPU has already been introduced on NVIDIA’s Tegra K1
mobile chip and has been confirmed as a dual-core, 64-bit ARM v9 CPU
which offloads some of the tasks from the GPU over to itself. We can
expect NVIDIA to offload some of their PhysX processing which is handled
by the CUDA Cores to be offloaded to the Denver CPU on the
high-performance Maxwell GPUs to gain more performance and faster GPU
physics processing. One thing is confirmed that Denver is ready (atleast
of paper) and would find its way on Maxwell cards eventually.
It is possible that NVIDIA may unveil more information on their
Maxwell core during the GTC 2014 (GPU Technology Conference) on 24th
March 2014 so its going to be a pretty interesting event similar to
AMD’s GPU14 where the announced their Volcanic Islands R200 series
lineup. A few NVIDIA Maxwell GeForce chips have also been spotted which
include engineering samples of GM108 and GM107 for
notebooks which could possibly hint NVIDIA introducing their mobility
lineup based on the 28nm architecture first and then introducing their
high end desktop chips for both GeForce and Tesla markets.