WHITE PAPER
Accelerating Big Data Processing and AI at the Edge
mrcy.com 5
MOVING AI TO THE EDGE
Unprecedented GPU Processing Power
NVIDIA revolutionized parallel computing when it invented
the GPU in 1999 and has continued to develop the multi-
core design concept. Since those early GPUs, with fewer
than 20 cores, technology has moved for ward on a
geometric growth cur ve. Individual GPUs from NVIDIA now
have thousands of cores, delivering teraflops of processing
per formance. They also implement extremely fast and
efficient data transfers to and from memor y, an essential
capability for processing high-bandwidth sensor input.
The first step in enabling AI at the edge was the explosion
in processing power delivered by GPUs, bringing
outstanding computing per formance to cer tain kinds of
applications by using many specialized cores to execute
math operations in parallel. This kind of computing is a
great match for rapidly rendering graphics, which is why
GPUs were first developed. However, the GPU architecture
is also a great match for AI algorithms, which rely on
linear algebra for a large por tion of their operations.
With this type of computing power in a single GPU chip, AI
at the edge became feasible – but more work still had to
be done. One step was ruggedizing NVIDIA's commercially
Environmental Resilience
Mercur y 's rugged edge servers are purpose-built to accelerate compute-intensive mission-critical applications in the most remote and harshest environments.
available GPUs to withstand harsh environments
often found at the edge. In parallel, the edge node
GPUs needed a software development environment
optimized for creating advanced AI applications.
Ruggedization
For almost 40 years, Mercur y has been ruggedizing
leading-edge commercial electronic components so they
can be used with confidence by aerospace and federal
customers. Bringing that experience and exper tise
to a technical par tnership with NVIDIA, Mercur y has
developed a range of deployable options for power ful
GPU computing. NVIDIA's A30/A40/A100 ser ver-class
products are now available from Mercur y in ruggedized
ser ver designs. Mercur y has also taken individual NVIDIA
GPUs and designed them into environmentally resilient
form factors, including 3U and 6U OpenVPX modules.
Able to withstand shock, vibration and extreme
temperatures, these solutions can operate on mobile
platforms and in the most difficult conditions, inter facing
directly with sensors and executing sophisticated
software analyzing an endless stream of big data.
