The apocryphal tale of Benjamin Franklin venturing out into a storm with kite and key in hand set the stage for the 21st century’s voracious appetite for electricity. Artificial Intelligence has proved to be a glutton for energy consumption. The amount of computing power necessary for AI machine learning is doubling every six months. Data center behemoths like Amazon Web Services, Microsoft, and Google are looking to reduce their carbon footprint, increase efficiency, and reduce energy costs. To comply with this trend, the IT infrastructure is evolving towards a more distributed compute architecture and a broader variety of processor cores and types (such as Arm-based CPUs and system-on-chip processors). JB Baker, VP of Products for ScaleFlux says, “Large-scale AI workloads rely heavily on memory and networking, putting much of the current focus on alleviating bottlenecks in these areas. However, with the petabyte and exabyte scale of the datasets used in AI, power and efficiencies of the underlying storage infrastructure are quietly a major challenge. Improving the pipeline between storage, memory, and GPUs with solutions like solid-state drives (SSDs) enhanced with computational storage capabilities is critical in enabling companies to achieve their efficiency and energy sustainability goals.”
A data center is a tangible space dedicated to housing IT infrastructure. Its purpose is to construct, operate, and provide applications and services while also serving as the storage and management hub for the associated data. Once privately owned for the exclusive use of one company, data centers have evolved into remote facilities managed by cloud service providers that house a virtualized IT infrastructure shared among multiple companies and customers. Of the 10,978 data center locations worldwide, 5,388 are in the United States. Newmark’s new U.S. Data Center Market report says that the U.S. data center footprint will more than double from 17 gigawatts in 2022 to 35 gigawatts in 2030.
Troubling statistical data strongly suggests that data centers and data transmission networks are infringing on global efforts to improve the sustainability of energy resources and minimize harm to the environment. The Department of Energy reports that data centers consume up to 50x as much energy per floor space as conventional commercial office buildings.
In January, the International Energy Agency (IEA) released its forecast for global energy consumption for the next two years. Notably, the report included projections for electricity usage linked to data centers, cryptocurrency, and artificial intelligence for the first time. According to the IEA, these sectors combined accounted for nearly 2% of global energy demand in 2022. The agency predicts that by 2026, demand for these services could double, reaching a level approximately equivalent to the whole of Japan’s electricity consumption.
AI models are trained using copious data, and the more complex and popular an AI model is, the more requests it receives and the more energy it uses. Training can take from just a few minutes to months, based on the amount of data and the AI’s level of sophistication. GPUs, the electronic circuits or computer chips that perform the computational tasks used in machine learning, run 24 hours a day, devouring vast amounts of energy while the AI model deduces a user query, processes it, thinks about it, and provides the answer.
This perfect storm of data-hungry tech and its unquenchable need for more energy to power increasingly sophisticated data processing needs cannot be calmed without changes to the IT architecture.
A solid-state drive (SSD) is a mass storage device with no moving parts, unlike a spinning hard disk drive (HDD). It stores data on integrated circuits, utilizing flash memory for faster data access compared to traditional HDDs. SSDs boast considerably faster accessing speeds than HDDs, resulting in accelerated file transfers, reduced power consumption, increased durability against physical damage, and quieter operation. SSDs offer instant system boot-up, seamless multitasking, better system cooling, and flexible storage options.
Also Read: Rambus Expands Chipset for Advanced Data Center Memory Modules with DDR5 Server PMICs
Data center operators prioritize power efficiency, component dependability, RAS (reliability, availability, and serviceability), and storage and performance density. SSDs have transformed enterprise data centers with high-speed performance and agility, critical in split-second decision-making. SSDs offer rapid data access, energy savings, and compact size, optimizing space and scalability.
In computational storage solutions like SSDs, aspects of data processing become localized on the drive where the data is stored, providing additional processing capability to the server. Rather than transmitting data over long distances, localized data processing minimizes latency, energy consumption, and bandwidth utilization. And since data stays on the drive, computational storage solutions enhance privacy and security.
Arm-based computing uses an advanced RISC (Reduced Instruction Set Computer) microprocessor architecture design that features lower power consumption, better efficiency, lower cost, and higher performance than traditional x86 CISC (Complex Instruction Set Computing) architectures. Arm server processors offer extensive scalability and can manage workloads from lightweight web servers to high-performance computing tasks. Their lower power consumption and higher performance make them a popular choice for AI and ML algorithms in data centers.
ScaleFlux is the computational storage leader, helping businesses get more from their data infrastructure and budget at scale. Their transformational SSDs embed intelligent computational storage technology to perform storage computing directly in the data path, optimizing the data pipeline for enhanced scalability. ScaleFlux‘s smarter SSD transparently compresses and accelerates data to double performance and efficiency. It supports its customers in reaching their sustainability goals, enabling more data storage per gigabyte of flash, reducing the requirement for extra components, and decreasing environmental waste.
