Massively Parallel Computing

WorldScape has demonstrated the world’s highest performance radar processing benchmark

WorldScape’s Scalable Processing Platform (SPP) is based on advanced massively parallel processing technology in scalable, high-bandwidth architectures capable of supporting a variety of needs across many application areas.

Massively parallel processing technology has long been known to offer up to orders of magnitude more computational performance than traditional single-CPU machines. There are multiple ways to achieve massively parallel implementations. A common method is to create computer clusters, such as Beowulf clusters. While these implementations can provide greatly increased throughput, they come with several downsides that, for one or more reasons, can limit or restrict use. Commonly referenced downsides of clusters are: the physical size of the cluster, the aggregate power requirement for the cluster, complexity of networking and programming, and cost of ownership.

Single Instruction Multiple Data (SIMD) is an alternative architecture for implementing massively parallel systems. SIMD architectures have been around since the 1960's and are made up of many processing elements (PEs) operating in lockstep. All processors perform the same operation at the same time on different parts of the data. This data parallel approach applies to many common types of processing being handled today by traditional computing hardware. Distributing the data and performing computations in parallel on numerous processors can dramatically increase performance.

It has been well known that for applications that map well onto SIMD architectures [point to page with powerpoint slides], up to two or more orders of magnitude increased performance can be achieved with well implemented coding. In general, applications that are known to map well onto SIMD architectures include:

•  Image Processing
•  Signal Processing
•  Compresssion/De-compression
•  Encryption/De-cryption
•  Image/Text Search
•  Network Processing
•  Certain other supercomputing applications

In addition to the raw performance benefits of SIMD architectures, current fabrication technologies enable large numbers of processors to be integrated onto a single chip. This dramatically reduces the power dissipation (Watts) as well as reduces the size. The graphs below illustrate both raw power improvements of the SIMTAP technology as well as the performance per watt superiority.

WorldScape Sensor Processing Platform (SPP)
WorldScape has developed and is implementing a scalable processing architecture specifically tailored to handle simultaneous inputs from multiple sensors. In-line processing is available per channel (e.g., compression) and cross-channel processing is enabled for more complex applications (e.g., multi-spectral imaging, target recognition, etc.).