NVIDIA GeForce ULP K 1 ( Tegra K1 Kepler GPU )

The NVIDIA Tegra K1 Kepler GPU (successor of the GeForce
ULP series) is the integrated graphics card in the Tegra K1
(formerly known as Tegra 5). Compared to its predecessor,
the GeForce ULP, it made the move to a unified shader
architecture . It integrates a full-featured Kepler-based
graphics card that includes a single SMX with 192 CUDA
cores clocked at up to 950 MHz, 8 TMUs and 4 ROPs.
Therefore, Tegra K1 offers about half the power of a GK208
chip (384 shaders, 32 TMUs, 8 ROPs) such as a GeForce GT
730M or 740M . This is a big step forward compared to the
GeForce ULP in the Tegra 4 SoC , which was based on an
old architecture with dedicated pixel- and vertex shaders
that is similar to the GeForce 6 series of 2004 and supported
only OpenGL ES 2.0. The new graphics card supports
OpenGL 4.4 and DirectX 11 in hardware, or well above the
current OpenGL ES 3.0 standard that is most common
amongst the majority of mobile operating systems.
Furthermore, the Kepler GPU can be used for general
calculations using CUDA and OpenCL 1.1 (1.2 in theory). As
an example, Nvidia has shown how to perform real time
video effects using the GPU and CUDA (Chimera 2).
Performance
In a 7-inch Tegra K1 reference tablet, Nvidia claims the
ability to render 60 fps in the GFXBench 2.7.5 T-Rex
offscreen test. In our benchmarks, the PowerVR G6430 in
the iPad Air was the fastest ARM-based SoC GPU, which
achieved only 27 fps. The similarly sized iPad Mini
Retina managed 25 fps in the same benchmark. 60 fps
would be similar to a Haswell-based Intel HD Graphics
4400 (e.g. in the 4200U ). The Adreno 330 achieved 20 - 24
fps in our benchmarks (e.g. in the Galaxy Note 3 or 10.1 ).
Even the announced Adreno 420, which should be up to 40%
faster than its predecessor, is not expected to outperform
the Tegra 5.
More information on the SoC and performance data on the
CPU can be found on the SoC page of the Tegra K1 .