Trends Driving New Processing Architectures
With the rapid expansion of applications that can be characterised by dataflow processing, such as natural-language processing and recommendation engines, the performance and efficiency challenges of traditional, instruction set architectures have become apparent. To address this and enable the next generation of scientific and machine-learning
applications, SambaNova Systems has developed the Reconfigurable Dataflow ArchitectureTM, a unique vertically integrated platform that is optimised from algorithm to silicon. Three key long-term trends infused SambaNova’s effort to develop this new accelerated computing architecture.
First, the sizeable, generation-to-generation performance gains for multicore processors have tapered off. As a result, developers can no longer depend on traditional performance improvements to power more complex and sophisticated applications. This holds true for both CPU fat-core and GPU thin-core architectures. A new approach is required to extract more useful work from current semiconductor technologies. Amplifying the gap between required and available computing is the explosion in the use of deep learning. According to a study by OpenAI, during the period between
2012 and 2020, the compute power used for notable artificial intelligence achievements has doubled every 3.4 months.
Second, is the need for learning systems that unify machine-learning training and inference. Today, it is common for GPUs to be used for training and CPUs to be used for inference based on their different characteristics. Many real-life
systems demonstrate continual and sometimes unpredictable change, which means predictive accuracy of models declines without frequent updates. An architecture that efficiently supports both training and inference enables
continuous learning and accuracy improvements while also simplifying the develop-train-deploy, machine-learning life cycle.
Finally, while the performance challenges are acute for machine learning, other workloads such as analytics, scientific applications and even SQL data processing all exhibit dataflow characteristics and will require acceleration.
New approaches should be flexible enough to support broader workloads and facilitate the convergence of machine learning and HPC or machine learning and business applications.
