Analog Devices Blackfin, Getting Started

Page: 42 / 154

Blackfin Processor Features
1-24 Getting Started With Blackfin Processors
Although these numbers provide a rough idea of a device’s performance,
they do not measure how an application runs on a device because they do
not take into account memory efficiency or instruction set efficiency.
Often, these peak specifications occur only momentarily (that is, they are
not sustained), and the sustained values are much lower. This is where
benchmark data can be useful.
“Benchmarks Against Other Processors” on
page 1-26 describes performance measurements reported by third parties.
Figure 1-3 shows a chart that demonstrates power consumption versus
speed for a 750 MHz Blackfin processor and various devices. What follows
is the Berkeley Design Technology, Inc. (BDTI) text description of the
figure:
BDTImark2000™ scores for fixed-point DSPs based on their fastest family
member. The BDTImark2000 is a summary measure of processors’ DSP speed.
The score is derived from a processor’s performance on BDTI’s DSP Kernel
Benchmark™ suite. A higher BDTImark2000 score indicates a faster processor.
System developers can leverage the wide range of performance options
available with Blackfin processors. Lower frequency, signal-core devices
scale up to high-frequency, high-bandwidth, dual-core devices.
ADSP-BF561 Blackfin processors provide additional options for power
management. Because this symmetric processor contains two identical
cores, traditional processing-intensive applications can be split equally to
run on each of the two cores. In this model, code running on each core is
identical; only the data being processed is different. In a channel stream-
ing application, the first core processes half of the channels and the other
core processes the other channels. In video and imaging applications, this
technique can be used to process alternate frames on each of the cores.
Dual-core processing melds with the Blackfin processor’s additional power
savings features. The energy consumed by a processor is based on both
static and dynamic components. Even when the application fits on a sin-
gle-core processor, you can employ a dual-core processor to reduce overall
energy consumption. Specifically, by running an application at half the
frequency of a single-core system, the processor core voltages can also be