It began with the recognition of the parallel software implementation gap

There are three insights of parallel software application development one should know:

The first is that application domain expertise is key to enable efficient application. Parallelization by hacking sequential code often does not yield much speedup. The process is error-prone, and generates new performance bottlenecks as old ones are removed. That means wasted money, performance and technological investment.

The second is that deep parallel programming expertise is required to architect parallel software. Application domain experts may be able to adopt a new API, but without a deep understanding of parallel processor technology roadmap, the software often cannot scale to faster performance on the future generations parallel processor. That means high non-recurring engineering (NRE) cost as the hardware scales.

The third is that both application domain expertise and parallel programming expertise are required to develop efficient parallel applications. Application domain expertise alone is not enough for evaluating the performance implications of design alternatives. Parallel programming expertise alone is not enough for exploring the full application-level design space. This is the Implementation Gap, which imposes a significant barrier for organizations to conquer computational bottlenecks and develop efficient parallel software.

Passion to conquer computational bottlenecks fuels 3-year-long research

What does it take to conquer computational bottlenecks in various application domains? Jike decided to find out.

Looking at the application acceleration process of application domains (including automatic speech recognition, computational finance, computer vision, and medical imaging), they takes months, and the outcomes (50x or 500x speedup) are difficult to estimate.

With the required cost, and risk in technology and process, unless there’s significant gain for an organization, there’s no reason to invest the time, energy and resources. Even with years of experience working on intricacies of parallel programming, a process has to defined and tested to put in place.

The goal: help organizations accelerate their businesses and save costs

With our Ready-Set-Go model executed in real environments, our clients measured application performance increases of 14x to 750x. They not only could postpone building of new data centers and computational power, but can also use their computing power to offer additional services.

There is potential to achieve orders of magnitude speedups at a fraction of a cost with highly parallel platforms. Parasians are here to empower, catalyze and execute!