A new parallel world


From Moore's Law to AI

The Problem

We are in an era of exploding volumes of data. Organizations are inundated with massive amounts of information, that they must process quickly to identify emerging market trends and respond to shifts in consumer demand. For years, computers have ‘automatically’ gotten faster thanks to Moore’s Law. However, Moore’s Law is rapidly coming to an end. In response, hardware companies have pitched devices such as ‘multi­core’ CPUs and GPUs to address the demand for more processing power. However, such chips are much more difficult to program and require fundamentally new tools to use effectively compared to traditional ‘single core’ CPUs. Furthermore, recent years have seen the deployment of several ‘application specific’ processors in chips ranging from smartphones to servers. Chips these days often have over 10 different types of processors! Making the best use of such a heterogenous range of processors is a daunting software programming task, requiring expert knowledge of each piece of hardware that must be programmed. As new versions of chips are released frequently, it is an unsustainable strategy to expect programmers to learn about each new piece of hardware to make the best use of it. This situation will lead to companies spending more money to buy even more underutilized hardware, as programmers scramble to keep up with every new hardware iteration.

High­level overview of the YETI platform.

Our Approach

YetiWare's solution to this problem is based on a new abstract computer model that unlocks a novel programming paradigm; one that shifts the complexity of heterogeneous software performance and portability to the proprietary YETI platform. This platform completely isolates software developers from the complexities of the underlying processor architectures, thus simplifying development and maximizing both performance and portability across a wide range of computing platforms. The YETI platform enables dynamic optimization of algorithms for a wide range of hardware at runtime using cutting­edge machine learning techniques. Based on user settings, YETI will continuously search for optimal latency, energy, and throughput optimizations as the application is running. When an optimization is found, it will be reported back to YetiWare where it will be distributed to YETI platforms that subscribe for optimization updates so that user applications transparently increase in performance – this creates an optimization marketplace with data crowdsourced by running YETI platforms. Thus, all of YetiWare’s customers will benefit when more people switch to the YETI platform!