New GPU architectures don't come quite as
frequently as processors, which is why the NVIDIA GTX 680 is such a monumental
arrival.
The kepler architecture
NVIDIA's Kepler GPU architecture is
designed from the ground-up with the purpose of maximising performance while
maintaining optimal performance per watt. In other words: Power Efficiency.
Kepler succeeds the previous Fermi
architecture, which was the first to bring two key advancements: full geometry
processing to the GPU (enabling DirectX 11 tessellation), and a great
improvement to the GPU's performance in general computation. While undoubtedly
powerful, it also consumes a lot of power -- something the Kepler aims to
rectify.
The Kepler's architecture features two key
changes. Firstly, they redesigned the streaming multiprocessor (the most
important building block for GPUs) for optimal performance per watt. Secondly,
they added a feature called GPU Boost, which increases clock speed to improve
performance within the card's budget.
This gives birth to Kepler's new SM, called
SMX. SMX eliminates the Fermi's "2x" processor clock and uses the
same base clock across the GPU. SMX balances out this change by using an
ultra-wide design with 192 CUDA cores. The total cores on the GeForce GTX 680
is at 1,536 cores, easily outperforming their older cards.
GPU boost
GPU's today operate on a Thermal Design
Point (TDP) that leaves a lot of headroom when running most games. GPU Boost
automatically adjusts the clock speed based on the power consumed by the
currently running app (instead of the most power hungry one), and as such will
dynamically increase the clock speed to take advantage of the extra power
headroom. It uses realtime hardware monitoring as opposed to application-based
profiles, reading a huge amount of data such as GPU temperature, hardware
utilisation and power consumption. Depending on these conditions, the GPU will
raise the clock and voltage accordingly.
More monitors
Previous GTX cards require two cards to
power three monitors. The GTX 680 only requires one card for three, with an
additional monitor for non-gaming display. The display ports support here
include HDMI, DVI-D and mini DisplayPorts.
FXAA and TXAA
FXAA is a new anti-aliasing method that
works differently from the typical MSAA method (which renders everything at
four times the resolution); it picks out the edges in a frame based on contrast
detection, subsequently smoothing out the aliased edges based on gradient. The
result is equally smoothened edges without the consumption of additional
memory.
TXAA is a higher step in AA mode, designed
for direct integration into game engines. It combines the raw power of MSAA
with sophisticated resolve filters similar to those employed in CG films. It
comes in two modes: TXAA 1 (8 x MSAA with performance similar to 2 x MSAA) and
TXAA 2 (better image than 8 x MSAA with performance comparable to 4 x MSAA)
Adaptive V-SYNC
Vertical sync (V-Sync) is a remedy to
screen tearing during games, but V-Sync often causes frame rate drops and
notable stuttering. The new NVIDIA 300 drivers introduces the Adaptive V-Sync
feature, which automatically enables or disables itself based on whether the
frame rate exceeds the refresh rate of the monitor. That way, V-Sync becomes a
much more attractive option for gamers.
Vapour- chamber cooling
While other manufacturers will inevitably
go for their own proprietary cooling methods, the GTX 680 has a vapour-chamber
cooling system by default.
SLI, PHYSX and 3D
Naturally, the GTX 680 supports NVIDIA SLI
technology (combine three cards for world-changing performance), NVIDIA PhysX
(accurate physics emulation) and NVIDIA 3D Vision (for, well, 3D support).
