Computex Taipei: ARM Accelerating AI experiences from edge to cloud

First processors based on ARM DynamIQ
technology take a big step towards boosting AI performance by more than 50x
over the next 3-5 years

Nandan Nayampally, Vice President & General Manager, ARM Compute Product Group and (Right) Jem Davies, Vice president, general manager and fellow, Media Processing Group (MPG) announced the launch of ARM Cortex-A75, Cortex-A55 and ARM Mali-G72, accelerating AI experiences from edge to cloud

Artificial intelligence (AI) is already simplifying and transforming many of our lives and it seems that
every day I read about or see proofs of concept for potentially life-saving AI innovations. However, replicating the learning and
decision-making functions of the human brain starts with algorithms that often
require cloud-intensive compute power. Unfortunately, a cloud-centric approach
is not an optimal long-term solution if we want to make the life-changing
potential of AI ubiquitous and closer to the user for real-time inference and
greater privacy. In fact, survey data we will share in the coming weeks shows
85 percent of global consumers are concerned about securing AI technology, a
key indicator that more processing and storing of personal data on edge devices
is needed to instill a greater sense of confidence in AI privacy.

Enabling secure and ubiquitous AI is a
fundamental guiding design principle for ARM considering our technologies
currently reach 70 percent of the global population. As such, ARM has a
responsibility to rearchitect the compute experience for AI and other
human-like compute experiences. To do this, we need to enable faster, more
efficient, and secure distributed intelligence between computing at the edge of
the network and into the cloud.

ARM DynamIQ technology, which we initially previewed back in
March, was the first milestone on the path to distributing intelligence from
chip to cloud. Today we hit another key milestone, launching our first products
based on DynamIQ technology, the ARM Cortex-A75 and Cortex-A55 processors. Both
processors include:

To further optimize SoCs for distributed
intelligence and device-based Machine Learning (ML), we are also launching the
latest premium version of the world’s No. 1 shipping GPU, the Mali-G72. The new
Mali-G72 graphics processor, based
on the Bifrost architecture, is designed for the new and demanding use cases of
ML on device, as well as High Fidelity mobile gaming and mobile VR.

 

Cortex-A75: Breakthrough single-threaded
performance

I have been at ARM for more than a dozen
years and can’t remember being this excited about a product delivering such a
boost to single-threaded performance without compromising our efficiency
leadership. The Cortex-A75 delivers a massive 50 percent uplift in performance
and greater multicore capabilities, enabling our partners to address
high-performance use cases including laptops, networking, and servers, all within
a smartphone power profile. Additional performance data and a deep dive on
technical features can be found in this blog from Stefan Rosinger.

Cortex-A55: The new industry-leader in
high-efficiency processing

SoCs based on
Cortex-A53 came to market in 2013 and since then ARM partners have shipped a
staggering 1.5 billion units, and that volume is continuing to grow rapidly.
That’s an extremely high bar for any follow-on product to surpass. Yet, the
Cortex-A55 is not your typical follow-on product. With
dedicated AI instructions and up to 2.5x the performance-per-milliwatt
efficiency relative to today’s Cortex-A53 based
devices, the Cortex-A55 is the world’s most versatile high-efficiency
processor. For more performance data and technical details,
visit this blog from Govind Wathan.

Flexible big.LITTLE performance for more
everyday devices

When distributing intelligence from the
edge to the cloud, there are a diverse spectrum of compute needs to consider.
DynamIQ big.LITTLE provides more multi-core flexibility across more tiers of
performance and user experiences by enabling the configuration of big and
LITTLE processors on a single compute cluster for the first time.

The flexibility of DynamIQ big.LITTLE is
at the heart of the system-level approach distributed intelligence requires.
The combination of flexible CPU clusters, GPU compute technology, dedicated
accelerators, and the new ARM Compute Library work together to efficiently enhance and
scale AI performance. The free, open-source ARM Compute Library is a collection
of low-level software functions optimized for Cortex CPU and Mali GPU
architectures. This is just the latest example of ARM’s commitment to investing
more in software to get the most performance out of hardware without compromising
efficiency. On the CPU alone, ARM Compute Library can boost performance of AI
and ML workloads by 10x-15x on both new and existing ARM-based SoCs.

Mali-G72: Optimized for next-generation,
real-world content

Our system-level approach enables
innovation across multiple blocks of compute IP, including the GPU. The
Mali-G72 GPU builds on the success of its predecessor, the Mali-G71. The
Bifrost architecture enhancements boost the performance by up to 40 percent in
the Mali-G72, enabling our partners to advance the mobile VR experience and
push High Fidelity mobile gaming into the next realm. We have also designed the
Mali-G72 to provide for the most efficient and performant ML thanks to
arithmetic optimizations and increased caches, thus reducing bandwidth for a 17
percent ML efficiency gain. 

With 25 percent higher energy efficiency,
20 percent better performance density, and the new machine learning
optimizations, ARM can distribute intelligence more efficiently across the SoC.
To read additional technical details on the Mali-G72, visit this blog by Freddi Jeffries.

Distributed intelligence starts here

Today we’ve announced the next generation
of CPU and GPU IP engines designed to power the most advanced compute. The
image below represents the most optimized ARM-based SoC for your edge device. A
full suite of compute, media, display, security and system IP designed and validated together to deliver the highest-performing
and most efficient mobile compute experience. This suite of IP is supported by a range
of new System Guidance for Mobile (SGM-775) which includes everything from SoC
architecture to detailed pre-silicon analysis documentation, models, and
software – all available for free to ARM partners. For accelerated time-to-market
and optimized implementations to ensure highest performance and efficency, ARM
POP IP is available for Cortex-A75.

Leading software
ecosystem, from edge to cloud

Software
is central to future highly-efficient and secure distributed intelligence. The
ARM ecosystem is uniquely positioned to deliver the breadth of disruptive software
innovation required to kickstart the AI revolution. To further support our
latest CPU and GPU IP, we are also releasing a complete software development
environment. Our ecosystem now has the opportunity to develop software
optimized for DynamIQ ahead of hardware availability through a combination of
ARM virtual prototypes and DS-5 Development Studio.

As
ARM prepares to work with its partners to ship 100 billion chips over the next
five years, we are more agile than ever in enabling our ecosystem to guide the
transformation from a physical computing world into a more natural computing
world that’s always-on, intuitive and of course, intelligent. Today’s launch
puts us one step closer to our vision of Total Computing and transforming
intelligent solutions everywhere compute happens.