February 2016
47
B
oards
& M
odules
SoC processors enable COM
Express modules from 12W up
By Christian Eder,
congatec
With the launch of the first
SoC variant of the AMD Embedded
R-Series processors this high-end class
of desktop processors has also
become available for embedded
applications. For the first time and
thanks to their scalable TDP from
12W upwards, they can now even be
used in fully enclosed, fanless systems.
congatec supports the new processor
platform on COM Express
computer modules.
AMD Embedded R-Series SoC processors
predecessors combined the CPU and GPU on
a single die to an APU (Accelerated Process-
ing Unit). The new System-on-Chips (SoCs)
now also integrate the platform controller
hub (formerly Southbridge) which makes
board designs significantly more compact and
energy efficient. The new CPU cores provide
around 5%more performance while requiring
approximately 40% less energy. At the same
time, the scalability of the TDP is extremely
wide, ranging from 12W to 35W. When con-
figured for 12W to 15W, they can now be used
for the first time in fully fanless designs.
Besides high-performance embedded com-
puting, this opens new applications for
the AMD Embedded R-Series in areas that
demand extreme robustness or a high degree
of system safety. For instance, fully enclosed
systems are a must in clinical applications that
require the highest standards of hygiene. In
the industrial environment, protection from
dust and dirt is important. Low maintenance
is another important aspect that speaks for
fanless systems. So there is a multitude of
applications where AMD Embedded R-Se-
ries processors have become a viable option
for the first time. In many cases, the argu-
ments for the AMD Embedded R-Series are
the same as in the higher performance range
where fans are always required. It is primar-
ily the high-end embedded applications that
benefit from the embedded high performance
AMD Radeon graphics, or applications that
use AMD support of OpenCL and the Hetero-
geneous System Architecture (HAS) to their
competitive advantage.
The processors are therefore used in appli-
cations that require a particularly powerful,
highly integrated graphics and/or parallel pro-
cessing power. Examples can be found among
high-end gaming, e.g. digital pinball and
arcade machines, in demanding digital sig-
nage installations with large 4k displays, as
well as image and video analysis in industrial
vision systems and medical imaging technol-
ogy. Other applications that benefit from the
high GPGPU performance of the new pro-
cessors include security applications such as
video surveillance with face recognition, net-
work firewalls with deep packet inspection, or
IoT systems with integrated big data analytics.
Let’s look at the specific improvements that
have been achieved over the previous APU.
While the competition continues to imple-
ment its processors as multi-chip modules,
the AMD Embedded R-Series is the first SoC
design in the x86 high performance segment
to integrate GPU, CPU and I/O controllers on
a single die. AMD has used this higher level
of integration to optimize energy efficiency
and reduce power draw. The CPU is based on
the new Excavator architecture where two x86
cores share one L2 cache, just like R-Series
APU designs. In addition, some functional
units have been optimized to save space. For
example, the floating-point scheduler, the
fused multiply accumulate (FMAC) units as
well as the instruction cache has all been com-
pacted to lower energy consumption.
AMD has also refined the power management
featured in earlier CPUs and GPUs. Each
Excavator core uses 10 so-called adaptive volt-
age scaling frequency (AVFS) modules. They
optimize the voltage and clock rate of the indi-
vidual functional units, so that each core can
utilize the existing power and thermal budget
to the maximum. Yet the top TDP of the SoCs
remains the same as that of previous APUs -
despite the integration of the Southbridge that
consumes a significant 7.8W (i.e. 22% of the
maximum TDP of 3W) in the shape of the
fusion controller hub A75.
At the same time, it has increased the perfor-
mance of the Excavator architecture. Overall,
it provides 5% more instructions per clock
cycle at 40% less power and 23% smaller
die area compared to its predecessor Steam-
roller and therefore offers significantly more
performance per watt. When calculating the
performance increase per given watt, a the-
Figure 1. The conga-TR3
computer module from congatec
comes equipped with a 12-35W
AMD Embedded R-Series
processor. Providing
high-performance SoC graphics
and HSA 1.0 support, it can
theoretically yield up to 75%
more performance per watt.
