Sparsix Technology
Sparsix Corporation’s mission is to be the market leader in providing solutions that radically accelerate critical functions based on large sparse linear systems. One of the ways that Sparsix is pursuing this mission is the development and commercialization of ground-breaking new technology in the area of hardware acceleration for sparse linear systems analysis.
Problem
Sparse linear systems analysis is a mathematical tool universally used by scientists and engineers to model and analyze complex systems. Solving sparse linear systems quickly and efficiently is critically important to many different industries because the speed at which sparse linear systems analysis can be performed directly affects computer simulation run-times, design cycles, and production costs.
These industries face a common problem: sparse linear systems and their associated sparse data structures wreak havoc on computer memory controllers. The result is serious performance degradation and very long computation times, even on powerful computers.
Faster processors, faster memory chips, faster bus speeds or parallel processing techniques might help improve performance, but these solutions fail to address the fundamental problem: general-purpose computers are not designed to handle sparse data.
Solution
To offer a solution to this problem, Sparsix has developed a custom application accelerator – the Sparse System Accelerator (SSA).
The SSA accelerates sparse linear systems analysis in two steps. The first step uses Sparsix’s unique Intelligent Memory Controller (IMC) technology and its innovative method of bypassing a computer’s standard memory controller to reduce cache misses and dramatically increase the efficiency of accessing and analyzing sparse data.
The second step further accelerates this analysis by implementing in a high-performance field programmable gate array (FPGA) those algorithms and iterative solvers that are fundamental to sparse linear systems analysis.
The
SSA combines this FPGA – the heart of the product where all control
and computational logic is implemented – with a specialized memory
system on a custom-designed printed-circuit board that is hosted in
an industry-standard server, workstation or PC. By implementing the
algorithms and solvers in an FPGA, compute-intensive processes are
off-loaded from the central CPU to provide significant application
acceleration.
Preliminary tests show that by combining these two steps, the SSA is capable of performing calculations on sparse data up to 20 times faster than a general-purpose computer working without an SSA.
Sparsix has completed design of the SSA hardware platform and is testing prototype units.
Who will benefit?
Sparse data is ubiquitous – encountered by engineers simulating mechanical structures and electronic circuits, geologists analyzing oil and gas reservoirs, and financial engineers modeling futures markets, to name a few. Analyzing this data using sparse linear systems analysis creates one of the most troublesome bottlenecks in high-performance computing and affects industries as diverse as automotive, aerospace and defense, oil and gas exploration and production (E&P), telecommunications, bio-sciences, and investment banking.