Data-Level Parallelism in Vector, SIMD, and GPU Architectures - Computer Architecture: A Quantitative Approach - page 324

FIGURE 4.26 Key features of the GPUs for mobile clients and servers . The Tegra 2 is the
reference platform for Android OS and is found in the LG Optimus 2X cell phone.
The NVIDIA Tegra 2 for mobile devices provides both the system processor and the GPU
in a single chip using a single physical memory. The system processor is a dual-core ARM
Cortex-A9, with each core using out-of-order execution and dual instruction issue. Each core
includes the optional floating-point unit.
The GPU has hardware acceleration for programmable pixel shading, programmable vertex
and lighting, and 3D graphics, but it does not include the GPU computing features needed to
run CUDA or OpenCL programs.
The die size is 57 mm 2 (7.5 × 7.5 mm) in a 40 nm TSMC process, and it contains 242 million
transistors. It uses 1.5 wats.
The NVIDIA GTX 480 in Figure 4.26 is the first implementation of the Fermi architecture.
The clock rate is 1.4 GHz, and it includes 15 SIMD processors. The chip itself has 16, but to im-
prove yield only 15 of the 16 need work for this product. The path to GDDR5 memory is 384
(6 × 64) bits wide, and it interfaces that clock at 1.84 GHz, offering a peak memory bandwidth
of 177 GBytes/sec by transferring on both clock edges of double data rate memory. It connects
to the host system processor and memory via a PCI Express 2.0 × 16 link, which has a peak
bidirectional rate of 12 GBytes/sec.
All physical characteristics of the GTX 480 die are impressively large: It contains 3.0 billion
transistors, the die size is 520 mm 2 (22.8 × 22.8 mm) in a 40 nm TSMC process, and the typical
power is 167 wats. The whole module is 250 wats, which includes the GPU, GDRAMs, fans,
power regulators, and so on.