Enhancing Data Center Energy Efficiency with DPUs

Energy efficiency is a top consideration for organizations as the world enters an era built on technology. Data and technologies, like AI, cloud, & 5G, are providing transformative benefits to organizations in the form of new business models, product and service enhancements, and new applications.?

Ronald van Loon is a NVIDIA partner and had an important conversation about reducing the environmental impacts of technology by enhancing the energy efficiency of servers and networks in the era of green computing.?

However, ushering in an era of advanced technologies that are resource and power intensive at a time when climate change is a top global agenda item adds additional complexities to the picture.?

If organizations are to continue to innovate with technology, they require new targeted energy-efficient solutions that can help them reduce their impact on the environment.?

Technology Innovation Meets Environmental Sustainability?

As organizations transform their IT infrastructure to support a data-driven enterprise, data center efficiency is a top target for improvement due to rising costs and further constraints on power capacity.?

Complex geopolitical landscapes in Europe are causing an energy crisis while businesses need to meet regulations, such as the Paris Agreement, to achieve net-zero carbon emissions. Amidst the disruption and uncertainty of current and future events, many organizations across industries are feeling the pressures of industrialization meeting environmental responsibility.?

It’s imperative that organizations acquire energy-efficient solutions so they can continue to innovate with new advanced technologies that are compute and network intensive, while also driving sustainability goals that meet regulatory conditions and fight climate change.??

One of the best ways to enhance energy efficiency is to transform infrastructure by using a Data Processing Unit (DPU) and SmartNICs to accelerate computing by accelerating and offloading infrastructure networking, storage, and security. This can reduce server power consumption by up to 30% .??

Reducing data center power costs by improving their efficiency is a perfect example of how organizations can develop new, agile, green strategies to keep pace with the evolution of technology so they can continuously push the boundaries of innovation.

Data Centers

Improving power consumption and reducing the costs of power for data centers are top priorities for organizations today as they weigh whether to expand existing data centers or target new ones. Data center managers are concerned with energy efficiency, seeking ways to improve performance while reducing costs.

Data centers consume a lot of power:

• In fact just over 1% of the world’s electricity production is consumed by data centers , that includes 1.8% of US electricity, and 2.7% of European electricity.?
• It’s estimated that this number could rise as high as 8% of all electricity production by 2030.?

Even cloud service providers are pressured to reduce the costs for SaaS, PaaS, and IaaS and keep pace with rising electricity costs per location.?

Geopolitical disruptions and supply chain constraints are causing electricity prices to skyrocket while many data centers are limited on how much power they can provide at any price. This creates an opportunity to target data centers for enhancing energy efficiency so businesses can innovate with new applications and receive higher productivity from existing data centers.?

Reducing Costs vs Improving Efficiency

Organizations, when they can, are building data centers in new locations that have less expensive electricity costs. Businesses are also trying to improve Power Usage Effectiveness (PUE) to maximize the amount of power that's delivered to the data center.?

Companies are also exploring cooling as another popular method to reduce data center power costs, such as through using data centers in colder climates to utilize free air cooling, or implementing liquid cooling.?

These strategies can reduce power consumption and costs, but they don’t improve the efficiency of servers. Achieving lower PUE ratios is an important goal for data centers, but PUE doesn’t account for server efficiency. A main challenge is that most data centers have already reached their PUE limit because they have already improved cooling and power distribution capabilities.?

However, it’s possible to maximize the utilization of each server with virtualization or containers, but this could result in increasing the power consumption of each server. A business can also outsource IT workloads to a public cloud with lower costs, but not all workloads can be outsourced due to enterprise data privacy and security concerns, for example.?

Businesses therefore need to improve server efficiency with domain specific processes, like DPUs. This will help them to target server efficiency while reducing the costs of energy consumption.??

Testing DPUs on the Data Center

Accelerating computing with DPUs is the next big development for improving data center efficiency. Businesses need sustainable computing to minimize their environmental impacts by changing the way systems and software are used.?

Energy-efficient computing is going to be a major solution to respond to climate change. AI and networking tasks are becoming more efficient as a result of GPUs and DPUs when run on supercomputers and data centers. Measuring AI efficiency is the next frontier for businesses as they balance efficiency and performance of AI workloads.?

DPUs run data center infrastructure tasks more efficiently than a CPU because they have dedicated hardware engines for accelerating networking, data encryption and decryption, key management, and storage virtualization.????

Organizations like telecom providers have very demanding network requirements. Researchers at Ericsson tested SmartNICs on workloads , which resulted in:

• Their CPUs significantly slowed down or slept when the servers were not as busy, saving power during periods of low networking loads
• Offloading networking from the CPU to SmartNICs also let the CPUs reduce power consumption.
• Using both strategies, energy consumption of server CPUs fell 23.7%.
• This single DPU application could cut power costs by nearly $2 million over three years for a large data center.

According to Ericsson’s CTO Erik Ekudden , “There’s a growing sense of urgency among communication service providers to find and implement innovative solutions that reduce network energy consumption.” DPU techniques “save energy across a wide range of traffic conditions.”

Performance

Similar DPU tests were conducted on Red Hat OpenShift where DPUs handled virtualization, encryption, and networking jobs. As a result, DPUs cut networking demands on CPUs by up to 70% and accelerated networking jobs by up to 74%.

Security

As companies embrace a zero-trust approach to network security, tests were made on one of the most widely used network encryption protocols. This resulted in improving performance and reducing power consumption by 21%. This could impact cost savings of up to $9 million over the course of the next 3 years for large data centers.

Green Energy IT Infrastructures

Green energy IT infrastructures are going to be more widely used as energy costs continue to rise and climate change is putting pressures on sustainability standards for businesses across the globe.

Enhancing network efficiency in the era of green computing will help organizations innovate with advanced technology like AI while reducing their impact on the environment.?

Be sure to check out NVIDIA for more information about enhancing data center efficiency with DPUs.