Decarbonizing Compute
Leveraging low-power compute and maximizing performance per watt, we can reduce technology’s carbon footprint for a carbon net zero future.
Decarbonizing Compute
How the Arm Ecosystem Is Laying the
Foundation for a Net-Zero Emissions Future
Our Vision
Decarbonize Compute, Decarbonize the World
The New Compute Imperative
The climate crisis is one of humanity's greatest challenges. As the IPCC signaled in its Sixth Assessment Report, we are at Code Red for humanity. It is indisputable that we are negatively impacting the planet – and urgent action is required. We must reach net-zero carbon by 2050 or face catastrophic climate change1.
We believe in the power of technology to build a better world for everyone, driving down emissions by unlocking efficiencies and reducing global energy consumption.
The World Economic Forum estimates that digital solutions could reduce emissions by 15% by 20302 by fine-tuning existing activities and replacing traditional solutions with intelligent ones. Reducing the environmental impact of the underlying technology, however, is critical. Connectivity cannot come at the expense of the planet.
Technology needs to be an integral part of national climate plans."
United Nations Climate Change Executive Secretary, Patricia Espinosa
Net Zero Goals
The tension between technology as a solution to our environmental problems and an exacerbating factor is not new. Technology roadmaps will have to place an even greater focus on efficiency to reduce overall power consumption as demand for cloud services and digital technologies grows.
Efficiency has long been a design imperative for battery-run devices. But in the race to reach carbon net zero, this efficiency is now being extended to the most performant devices, from electric vehicles to supercomputers.
That’s why we’re working hard to decarbonize compute, leveraging our expertise in low-power processing to maximize performance per watt. The next two generations of our highest-power CPUs, for example, will bring expected performance gains of 30 percent while keeping energy use flat.
What Is the Biggest Barrier to Climate Progress?
- Technology. We don't have adequate solutions yet to replace fossil fuels.
- Economics. We have solutions, but they are not nearly cheap enough to scale broadly.
- Policies. We need frameworks that will incentivize retrofits and adoption.
Inherently Efficient
Arm's energy-efficient processor designs and software platforms enable products from the sensor to the smartphone and the supercomputer.
Delivering more and more compute performance, while simultaneously improving energy efficiency, is what our partners ask from us every day. Decarbonizing compute is an ambition that makes both commercial and environmental sense.
By driving down carbon emissions wherever compute happens, we can help technology to achieve maximum benefit with minimal impact.
"As we become an increasingly digital world, the threat to the climate continues to grow but Arm gives us good reason to hope. Sustainable, innovative technologies that aim to help everyone everywhere can increase connectivity while limiting the impact on climate change, leaving the world a better place for every child and generations to come."
Fayaz King, Deputy Executive Director, Field Results and Innovation, UNICEF
Did you know…?
Arm's first CPU – the Arm1 – was designed to run at <1W. In testing, despite a problem with the power supply, the chip ran perfectly. It was found to be running on energy harvested from the surrounding chips, consuming just 0.1W. This energy efficiency laid the foundations for the low-power compute Arm is famous for today.
Powering Change
Connected Climate Solutions
The Connected Revolution
Technological innovation is deepening our understanding of the toughest environmental challenges – and giving us new ways to solve them.
A recent white paper from 451/S&P Global Intelligence, commissioned by Arm, explores how progress in four sectors—energy and utilities, buildings, datacenters, and transportation—will have a major impact on global emission reductions.
By taking a digital path to net zero carbon, we can drive efficiency and create opportunity, even as we reduce emissions.
Decreasing Energy Usage, Increasing Energy Efficiencies
Data Centers
Data Processing Units (DPUs) such as NVIDIA's Bluefield-3, which contains 16 Arm processors, increase throughput while lowering power consumption.
Transportation
EVerBendy, an experimental flexible chip based on Arm, will allow EV makers to wrap ML-capable plastic processors around every battery cell, so enhancing vehicle safety, range, and module life.
Energy/Utilities
Korea Electric Power Company (KEPCO) deployed 30 million smart meters, based on Arm processor technology, to help manage energy consumption across its network.
Net Zero Buildings
Johnson Controls OpenBlue Enterprise Manager solution reduces energy spend by up to 30% and decreases carbon footprint by 20-40%. Tracking a building’s assets is a complex computing task, which is why Johnson Controls is partnering with Arm.
Energy and Utilities
A Smart Grid Is a Clean Grid
The energy sector accounts for more atmospheric CO2 than any other industry and will experience one of the greatest transformations over the next 30 years. The global trends are already evident: cars and heavy industries are switching from fossil fuels to electricity, and electric power generation is transitioning from natural gas and coal to intermittent renewables like solar and wind backed by microgrids and storage.
To achieve a net-zero future, worldwide clean energy investment will need to more than triple to over $4 trillion per year by 2030, with the percentage of electricity generated by solar and wind rising from 10% today to 70% by 20501.
Why Renewables Matter. The energy sector accounts for 76% of global CO2 emissions while electricity and heating alone account for 32%. In advanced economies like the U.S. and Europe, the power industry is also one of the largest consumers of water.2
A Digital Grid
Computing will play three key roles in the energy transition
1. Increased Efficiency. IoT, AI, and other technologies will boost the overall efficiency of equipment and appliances. While embedded processors have been employed for years, newer technologies for fine-tuning these systems are still relatively new: only 10% of fixed industrial machines installed in 2020 were IoT-enabled.3
2. Dynamic Control. Efficient appliances will be linked through building management systems and smart grids to service providers that unobtrusively balance power loads in real-time to lower costs, eliminate unnecessary consumption, and ensure comfort and security. Decarbonization can't happen without digitization of the grid.
3. Improved Power Supply. Hardware and software can also increase the power output of wind and solar while cutting downtime and cost. Renewable energy will also provide a rich source of data for AI: solar power projects are capable of generating up to 40x more data than fossil fuel plants.
Edge Impulse: A Safer Grid Through AI
Electrical grids have been called the largest and most sophisticated machines ever devised, daily delivering electricity to millions of customers in real-time across service territories that can span entire countries.
But grids also have their limitations. Utilities, for instance, typically learn about faults and outages the old-fashioned way: by a phone call from a worried customer.
The Symptom Before the Spark
The RAM-1—an intelligent grid sensor being developed in collaboration between Arm software partner Edge Impulse, research institute IRNAS, and equipment manufacturer Izoelektro—seeks to change that. The RAM-1 monitors and analyzes voltage surges and other parameters on live power lines in real-time to warn emergency employees about potential fires or emerging dangers.
Specifically, the device analyzes electrical waveforms on an ongoing basis. If the AI algorithms on the device detect an anomaly, data from the select event is sent to the cloud for further classification. If a problem is indicated, warnings are relayed to control rooms and field crews.
The system can also be used to optimize the distribution of power and reduce losses.
High Intelligence, Low Power
Although the inference engine continuously monitors activity, the RAM-1 uses little power by keeping analysis at the edge. Powered by a Nordic Semiconductor processor based on the Arm Cortex-M33, the system can last 20 years on a single battery. Low power consumption is possible due to the Arm-optimized Edge Impulse machine learning models, which enable energy usage orders of magnitude lower than in standard ML applications.
Over time, the AI analysis that is running on the device results in the longest and most efficient battery life.
Span: A Nervous System for the Sustainable Home
The electrical panel—the grey metal box likely in the corner of your garage—has been a fixture in homes for nearly 100 years. It’s getting a makeover.
Span, a startup from Silicon Valley, has developed a smart panel with integrated edge computing capabilities for greater visibility and control over power consumption.
Powered by a quad-core Arm Cortex-A chip, Span’s smart panel lets residents remotely control all lights, outlets, and appliances or develop a long-term, automated savings strategy. It can also inform homeowners through its smartphone app when devices like water heaters or air conditioners show symptoms of pending failure.
More importantly, Span provides a glide path to zero emissions living. The system can manage EV charging to match the power output of a home solar system, sell power from a battery storage system to a local utility so it can avoid turning on a dirty “peaker” plant, or improve resilience by islanding a home during blackouts or natural disasters.
Replacing gas-powered water heaters or stoves with cleaner electrical fixtures also becomes more feasible and economical. Span helps avoid unnecessary and costly home wiring upgrades by intelligently managing the available power from sources like the grid, solar, and batteries.
Green Mountain Power, a Vermont electric utility, is currently collaborating with Span to study how making data more readily available to consumers changes power consumption.
25 million people endured power outages in the U.S. lasting 15 hours on average in 2020. Span's smart panel automatically manages power to keep homes online.
Awesense: An Ecosystem Approach to Energy
Founded in 2009, Awesense is decarbonizing the grid with data.
It developed a solution—adopted by utilities such as BC Hydro—that combined an Arm-based line sensor and software for detecting faults and electricity theft.
Today, its Digital Energy Platform serves as the data curator for software, hardware, and services targeted at the energy transition. The focus is on the distribution grid: the complex web that links cities and customers to power sources.
The company’s goal is to reduce emissions by 10 million tonnes by 2022 and 100 million by 2025.
“The grid needs data and there’s not a lot of visibility into distribution,” says founder and CEO Mischa Steiner.
The company is collaborating with Doosan GridTech and Washington’s Snohomish Public Utility District to maximize renewables in microgrids and leveraging EVs as a peak power supply.
It is also working with cities and service providers to develop long-range capacity plans for EVs by identifying areas needing upgrades or that have excess capacity.
Other ecosystem partners include Peak Power, which provides energy management services for commercial and industrial buildings, and Rainforest Automation, which has created an edge gateway for controlling pool pumps, water heaters, and other appliances.
Awesense's Digital Energy Platform (left) and the Raptor 3 sensor
Buildings
Pervasive Computing = Pervasive Efficiency
Homes and commercial buildings, and their construction, account for nearly one third of global emissions and, despite substantial progress, average efficiency remains relatively low. The U.S. EPA estimates that 30% of energy delivered to commercial buildings gets wasted1 because of insulation gaps, unmonitored lighting, and appliances drawing power while in standby mode.
By 2050, the global stock of floor space is expected to double.2 Air conditioning is also on the rise: left unchecked, cooling will consume as much power in 2050 as China and India do today.3
New building codes and technologies will be needed. Success will also require minimizing disruption and cost while implementing safeguards for privacy and security.
Digital decarbonization for buildings is largely focused on three areas:
-
Replacing existing gas appliances and fixtures with electric. Natural gas, which heats 85% of homes in the U.K., will be phased out in favor of electric heat pumps starting in 2025.
-
Deploying AI, IoT, and cloud and edge services to precisely monitor and manage lights, HVAC, and other energy-consuming fixtures.
-
Installing microgrids, energy storage, and renewables. Universities have become early incubators for microgrid deployments.
5G Buldings. Lenovo is experimenting with 5G microcells built around Arm-based silicon for functions such as security, energy management, occupant health, and safety. Lenovo is also testing 5G for optimizing factory production.
Smart Appliances: Arçelik
Refrigerators are the second largest consumers of electricity in homes, accounting for 7-13% of the total. While manufacturers have achieved steady efficiency gains since 1990, current technologies are reaching a plateau. Power-saving alternatives such as better vacuum insulation panels could add 30% to retail prices.
Arm partner Arçelik, a global appliance manufacturer, is adding AI to standard refrigerators – using their existing computing resources – to conserve energy. The company developed a lightweight Reinforcement Learning (RL) algorithm for Arm Cortex-M processors that analyzed in-home behavior – not reams of training data – to balance compressor speed thus reducing power consumption and continually adjusting for the residents' daily patterns.
It found such a system could reduce power by up to 10% without adding to the bill of materials. Deployed across refrigerators in Europe, it could save enough power to shut nine small coal-powered power plants.
Arçelik is also looking into ways to leverage AI to improve industrial and commercial refrigerators. The RL model could also have application in air conditioning – with big potential for emissions reduction.
Waste Not: Arçelik also found RL could keep food fresh around 10% longer by moderating temperature fluctuations. Approximately one third of food gets wasted annually. If food waste were a nation, it would rank third in GHGs (8% of the total) and consume water equivalent to the annual discharge of the Volga.4
The Future: Smart Cameras on a Sustainable Edge
Computer vision and AI are increasingly being added to cameras for city planning, supplement elder care and reduce traffic. An estimated 770 million public-facing cameras already operate and the market is growing by 13% per year. 5,6
But smart cameras also have a significant carbon footprint. A 1080p smart camera operating at 30 frames per second can generate 2GB of data per hour7 or 17.5TB per year. Worldwide, that translates to 13.5 zettabytes per year, not including the data generated by deeper AI analysis.
Deploying AI at the edge (see chart) can cut emissions by 42%.8
Edge AI Can Cut Emissions by 42%
|
| 12 Months |
Cloud AI |
Edge AI
|
|
20 cameras
|
@ 1080p |
@ 1080p |
| Camera (kgCO2 e) |
522 |
1,304 |
| Wi-Fi (kgCO2 e) |
938 |
0.052 |
| Cloud (kgCO2 e) |
1,410 |
353 |
| Total CO2 |
2,870 |
1,657 |
| Percent change |
|
42% Less |
|
CO2 avoided (770M cameras)
|
19.2Mt |
Datacenters
More Performance, Less Power
There are nearly 230,000 datacenters worldwide containing 14 million server racks. Since 2013, operational floor space has more than doubled.1 Datacenter workloads and internet traffic, meanwhile, grew by 8x and 12x respectively in the past decade.
What hasn't grown is energy use. Datacenter power usage rose by only 6 percent from 2010 to 2018 and currently represents 1-2% of global consumption.2
Datacenters are also helping other industries to decarbonize: the 7% decline in 2020 emissions partly came through replacing travel with videoconferencing.3
But without continued improvement in performance per watt, power consumption will likely rise, increasing cost, emissions, and regulations.
A Sustainable Shift: Hyperscale cloud providers can shift workloads to regions with cleaner, cheaper power while minimizing performance impacts. Above, moving a VM from Sydney to Taiwan cuts carbon from 6 to 4 grams and cost from 95 to 80 cents while adding only 13 ms to latency. (Source: 451 Research.)
How did power consumption stay flat for a decade? Moore's Law, workload consolidation, and better cooling strategies increased the amount of work performed by datacenters per unit of energy at a regular, rapid cadence.
Unfortunately, many of these techniques are approaching limits. Leading hyperscale datacenters now post Power Use Effectiveness (PUE) ratings below 1.1, meaning little power goes to cooling. Meanwhile, capacity in multi-tenant datacenters is expected to grow by 8% through 20261.
Specialized processing – purpose-built CPUs, GPUs, DPUs, and other specialized processors that optimally allocate workloads – provides a way to continue to increase performance while reducing power usage, emissions, and cost.
A Cleaner Cloud. Servers based on Arm Neoverse can perform 33% to 64% more work (SPECint® 2017) while avoiding 42% to 74% percent of the CO2 emissions of servers using equivalent traditional processors.4 Deployed broadly, Neoverse could increase cloud capacity by 25% or more by 2030 without increasing energy consumption.5
Cloudflare: Reducing Carbon Emissions in the Cloud
Cloudflare’s mission is to build a better internet—one that's safe, performant, reliable, and that also consumes less energy.
Over 25 million internet properties run on Cloudflare's global network, which spans more than 250 cities in over 100 countries.
Cloudflare recently started to deploy its 11th generation servers across its network and will use Arm-based CPUs designed by Ampere Computing to power them.
Compared to its 10th generation servers based on traditional CPU architectures, the Arm-based servers process an incredible 57% more internet requests per watt.
Cloudflare customers benefit as well: any internet property on Cloudflare’s network will automatically reduce emissions and seamlessly contribute to a greener, more sustainable world.
Cloudflare Nitin Rao discusses how Cloudflare is deploying Arm Neoverse in it's 11th generation of servers.
Specialized Processing in Action
Innovative chips and systems from Arm partners are reducing carbon emissions while achieving unprecedented levels of performance.
Created by RIKEN and Fujitsu and containing nearly 8 million Arm-based processors, Fugaku is the world’s most powerful supercomputer. It is being used to simulate extreme weather, develop COVID-19 vaccines, and other tasks.
Single-threaded, multicore Central Processing Units (CPUs) optimized for the cloud, such as AWS's Graviton2, have helped customers increase work per watt by 30% or more while simultaneously reducing costs by 20% or more.
Delivering 3x better performance and 4x better performance per watt than its predecessor, Marvell's Octeon 10 Digital Processing Unit, based on Arm Neoverse N2, will bring leading performance to 5G and edge applications.
Transportation
A Bumper-to-Bumper Transformation
Transportation accounts for approximately 24% of worldwide and 29% of U.S. emissions and stands as one of the most difficult challenges to moving to a net-zero carbon world.
A primary tool for cutting emissions, electric vehicles (EVs) have grown rapidly, thanks in part to a rapid decline in battery prices, pro-EV policies, and improvements in driving range, performance, and customer choice. By 2025, 40 million EVs should be on the road, according to 451/S&P Global Intelligence.1
Still, that represents only a fraction of the world's 1.2 billion vehicles. To achieve net-zero goals, zero-emissions vehicles will need to account for over 80% of car, bus, and light-duty truck sales by 2040.2 Step changes in performance, cost, and infrastructure will be needed.
Driving a Grid Upgrade. EVs consumed 6 terawatt-hours (TWh) of electricity in 2016. By 2040, they will consume 2,090 TWh, or roughly half of the capacity of the U.S. today3. Cloud and edge applications for dynamically balancing charging with renewable capacity will be vital.
The new and improved e-powertrain provides the latest Nissan LEAF model with 110 kW of power output and 320 Nm of torque. The newly developed inverter has enabled an improvement in vehicle acceleration and has had a positive impact on driver experience.
These improvements have been made possible with the new power module which has increased the motor drive current. It features an Arm core processor, the Cortex-R4, and the latest control technology to increase efficiency and output – enabling approximately 1.3 times more torque and power than the previous Nissan LEAF.
In controlling the inverter, the Arm-based microcomputer core accurately repeats a series of processes such as sensing, calculation, and control output for the events that occur in 1/10,000 second cycles. Nissan found that the Arm Cortex-R family of cores was the right choice due to its efficient performance and responsiveness enabling it to dependably deliver the precise control needed within such a tight computation window.
Smart City Meets Smart Vehicle: Better Traffic Control with NXP and Arm
EVs, hybrids, and other vehicles will become vital nodes on smart city networks. With driver consent, utilities will be able to orchestrate EV charging over large geographic areas, for example, to avoid firing up dirty peak power plants.
Data sharing will also help drivers and transportation officials reduce the emissions, energy consumption, and lost time caused by traffic congestion. Before COVID-19, U.S. drivers on average lost 99 hours a year in traffic, generating millions of tons of excess emissions. And for residents of growing megacities like Bogota and Istanbul, it is even worse with drivers stuck up to 190 hours a year moving at 20 km/h or slower.4
NXP's Green4TransPORT (see video), based on Arm technology, is an innovative solution aimed at improving traffic flow and reducing emissions. Intelligent traffic infrastructure allows truck drivers to request priority at traffic crossings. Smart controllers reply, sending speed advice and a “countdown” to the next green light. Heavy trucks get priority, clearing crossings sooner and reducing congestion. The 'straight through' drive increases vehicle efficiency and reduces emissions in the city.
About Arm and References
Bringing brilliant people together to spark the world's potential
Wherever Computing Happens
Arm designs technology building blocks such as Central Processing Units (CPUs), Graphics Processing Units (GPUs), and Neural Processing Units (NPUs) that semiconductor designers, equipment manufacturers, and others use to create silicon chips and specialized compute systems to power a host of devices, applications, and services across global markets.
Our partners have used our technology to enable the mobile revolution, make vehicles safer, drive down the cost and emissions of data centers, and connect individuals and businesses in new, exciting ways.
With billions of transistors and specialized subsystems operating at extreme speeds, processors remain some of the most complex devices ever produced. By leveraging Arm's proven intellectual property, our partners can get to market quickly and reliably, delivering secure, unique system-on-chip (SoC) products that meet any performance need.
From One, Many: The Arm Cortex-A53 is one of our most popular CPUs. Semiconductor companies such as Qualcomm, Samsung, MediaTek, and NXP have adopted it to create a broad spectrum of entry-level, mid-range, and high-end SoCs.
These SoCs in turn have served as the foundation for smart speakers, computer vision systems for buildings, gaming controllers, a handheld for detecting bacteria in water for emerging nations, a device that lets cars park themselves, and billions of IoT devices and smartphones.
An Inventive Ecosystem
An ecosystem of hundreds of innovative partners has made Arm technology pervasive worldwide. Because of their efforts and ideas, Arm technology can be found in everything from TV remote controls to satellites.
The extended ecosystem includes Fortune 500 companies, cloud native developers, freshly minted startups, national laboratories, and technology distributors across the globe.
Arm also fosters close relationships with semiconductor foundries and EDA developers to help ensure that our designs can be smoothly transitioned to products.
References
Our Vision: 1. IPCC. 2. World Economic Forum and Exponential Group.
Electricity: 1. IEA: Net Zero by 2050; 2. European Environmental Agency and USGS. 3. Omdia. 4. Renewable Energy World.
Buildings. 1. U.S. EPA and Dept. of Energy. 2. Programme for Energy Efficiency. 3. IEA: The Future of Cooling. 4. Food and Agriculture Organization of the UN, UNEP. 5. Comparitech, IHS (770 million). 6. IDC ($50 billion, 13% CAGR) 7.. Makeuseof.com (2GB). 8. Arm calculations. We assumed the 20 cameras would run 18 hours for an entire year. Dynamic transmission draw for the cloud-centric setup is at 0.018 kw while the draw for the smart edge camera is .000001 kw with identical Wi-Fi hardware: the difference comes in the larger amount of data sent via the cloud case. Device power draw for the cloud-dominant camera is 0.01 kw while the smart camera draw is 0.25 kw. Carbon calculations are based on the U.S. average power mix.
Datacenters: 1. 451/S&P Arm report. 2. Science. 3. Univ. of East Anglia, Univ. of Exeter, Global Carbon Project. 4. Arm calculations. The power draw of the Arm processor is 284.4 watts. The power draw of the competition processors in the calculation is 336 and 388 watts, respectively, based on publicly available documents. The SpecInt ratings are based on Arm benchmark testing. We assumed one instance and 2 CPUs per instance running for a month in similar datacenters. Carbon calculations are based on the U.S. average power mix. 5. Cloudflare. 6. Cloudflare.
Transportation. 1. 451/S&P Arm report. 2. Bloomberg New Energy Finance. 3. BNEF (6TWh) 451/S&P Global Market Intelligence (2090 TWh), Arm estimates. Inrix Global Traffic Scorecard.
