Industry executives and experts share their predictions for 2023. Read them in this 15th annual VMblog.com series exclusive.
Acceleration of ARM Adoption for Server and Edge Applications
By Bill Conrades
For years, ARM processors have powered the vast
majority of the world's smartphones, largely because their low power consumption
helps deliver all-day battery life even for gamers and users with other
compute-intensive workloads.
Today, performance improvements and other
advances are expanding the use of ARM technology to newer uses like laptops, servers and edge
computing applications, offering benefits ranging from energy efficiency and
low price points to optimized support for cloud computing, artificial
intelligence and machine learning.
The shift to the use of ARM architecture in
enterprise computing vastly accelerated in 2020, when Apple announced that it was transitioning
a portion of its Mac computers from the Intel chips it had used since 2006 to
its own ARM-based M1 system-on-a-chip (SOC). The trend has been steadily
accelerating ever since. Here are several examples.
1
- ARM arrives in the data center: After a series of false starts that led to the shutdown of
companies like Calxeda a decade ago, Ampere
Computing is achieving substantial
success in providing an ARM-based, cloud-focused server architecture with its
Altra® SOC solutions featuring the world's only 128-core cloud native
processor. The low TCO makes Ampere's Altra reference platforms well-suited for
edge computing with heavy workloads, AI/ML and GPU workloads, and applications
demanding high core counts, flexible storage and peak memory.
Major ODMs like Supermicro, Gigabyte, Foxconn,
ADLINK and Wiwynn have added Ampere Altra-based server offerings to their
portfolios in the last year. So have Google Cloud, HPE and Microsoft, which has
introduced Azure VMs with Ampere Altra
ARM-based processors.
This clearly signals the dawn of a new
infrastructure model, with a 2022
study by Trendforce predicting that ARM architecture will have a 22% share of
data-center servers by 2025 led by cloud data centers.
2
- Netflix tests ARM platform to double server bandwidth: In 2021, in an effort to boost the
TLS-encrypted bandwidth of its video servers from 200 Gbps to 400 Gbps, Netflix compared servers equipped with AMD EPYC 7002 (Rome), Intel Xeon
Platinum 8352V (Ice Lake) and Ampere Altra Q80-30 processors. The AMD
configuration topped the field at 380 Gbps when TLS encryption was offloaded to optimize the
performance and avoid hardware bottlenecks, but Altra's ARM-based solution
finished a close second at 320 Gbps.
Netflix testers theorized that the slower
performance was caused by a PCIe-specific problem because they noticed low
processor utilization and saturated
NICs with lots of output drops. Considering benefits like energy and cost
savings, ARM will likely be a strong contender for content delivery networks
going forward.
3
- ARM CPUs drive AI at the edge: NVIDIA Jetson models combine an ARM architecture central
processing unit and an NVIDIA GPU within a SOC
designed by NVIDIA. The Jetson platforms take advantage
of ARM's high-performance and low power to run artificial intelligence and
machine learning workloads at the edge. This is fueling the development of new
edge-based inferencing solutions in areas ranging from medical, safety and
security applications to robotics, industrial automation and more.
One example is MBX Systems' Kori, the first mobile cart that can be outfitted with different
cameras, workstations and peripherals to accommodate the needs of different
computer vision applications. Preferred
compute options currently utilize the NVIDIA Jetson Xavier NX with plans to
transition to the next-generation Jetson Orin NX for single camera inferencing
and the Jetson Orin AGX for inferencing multiple camera streams. MBX will
have a dual camera AI demo running on the AGX Orin development kit at the
MD&M West trade show in Anaheim, CA, in February.
The cart also uses an ARM-based Raspberry Pi
module to power the 157 LEDs that can be individually programmed with different
color and sequencing patterns to indicate usage status, alerts and other
functions.
In hospital environments, for instance, Kori can
be used to deploy solutions that can detect patient falls, screen body temperatures,
facilitate patient-healthcare provider communication and enable remote live
surgical theater collaborations.
The low power consumption made possible by the embedded ARM CPU is particularly
useful in hospital settings because it allows inferencing to be performed
against the camera stream even if the cart is unplugged and running on
battery.
ARM
adoption beyond smartphones is just beginning. As this is written, for example, NVIDIA is
launching an industrial-grade edge AI computing platform called NVIDIA IGX that
will achieve a new level of performance per dollar and thereby make large-scale
edge inferencing a reality. Potential applications range from bringing AI to
entire traffic systems, smart cities and autonomous factories to medical uses
such as robotic surgery controls, endoscopy, diagnostic
imaging, radiation therapy, and microscopy.
The trend is clear. ARM is entering the
mainstream, opening new opportunities for
energy-efficient, performant, lower cost of ownership product development that
will continue to transform the technology landscape.
We'll all be watching.
##
ABOUT THE AUTHOR
Bill Conrades
is Director of Sales Engineering at MBX Systems
(www.mbx.com), a provider of custom computing hardware engineering,
manufacturing and support services for ISVs, OEMs and other technology
companies that deliver complex products on turnkey hardware.