Artificial intelligence (AI) and machine learning (ML) are two transformative technologies that are gaining traction in many areas of scientific research and business analytics. The cloud offers a platform from which organizations of all sizes can access the level of computing power necessary to conduct research and make effective use of these disciplines. Consequently, major cloud providers have bolstered their portfolios with cutting-edge products designed to attract companies interested in AI and ML.
Developing new solutions or delivering advanced functionality to popular platforms is one way a vendor can stand out against its rivals. A method of gaining a competitive advantage that is enjoying increased popularity among large providers is the use of custom-built hardware that is designed to optimize their software offerings. Let’s take a look at some of the hardware solutions and innovations that cloud vendors are employing to provide enhanced capabilities to their customers.
Google’s tensor processing units (TPUs) are a prime example of hardware dedicated to processing the workloads demanded by AI applications. These are custom-built chips designed specifically to provide the power required by AI systems. The chips are often favored over more traditional graphics cards for the higher speeds they offer.
Google has developed an infrastructure option called Cloud TPU Pods that use server racks packed with TPUs. They are configured with either 256 or 1,024 TPUs and the larger configurations offer speeds that approach those of supercomputers. The company powers its popular search engine and Google Translate with TPU pods.
Amazon Web Services’ Graviton processors are custom built by the company using 64-bit ARM Neoverse cores. The processors offer cost-efficient scalability for general-purpose, compute-optimized, and memory-optimized EC2 instances. Second-generation Graviton processors provide an improved price-performance ratio over x86 chips for diverse workloads such as video encoding and CPU-based machine learning. Security is enhanced with always-on 256-bit DRAM encryption, allowing developers to run native cloud applications securely.
Microsoft’s Project Olympus is an open-source solution that incorporates hardware and software modules to create a holistic rack architecture. Project Olympus is part of the Open Compute Project which is releasing technical information to the developer community that may help even the playing field in the server industry.
The schematics offer a well-defined starting point upon which manufacturers can build custom hardware solutions. Servers built under the auspices of Project Olympus will be infused with high debugging and testing capabilities that will help isolate intermittent and hard to duplicate problems.
As the computing world’s requirements for faster and more efficient processing continues to grow, the cloud-tech giants will continue to evolve their solutions to meet the demand. The immense resources they wield ensure that when the need calls for it, the financial cost of employing custom hardware will not be an obstacle.