High Performance Computing
Unparalleled performance. Unprecedented results.
While addressing "ease of use" of an FPGA platform, HANSA’s software stack enables
Macro level Parallelism - Where the problem may not be computationally intensive, but requires immense parallelism for serving multiple requests - for example running 1000 simple derivative calculations concurrently in a SaAS framework. The advantage of utilizing this method is speed to market. Utilizing HANSA’s Macro Level Parallelism requires no re-writing of your C/C++/Java codes. Since the parallelism is contained within a single box, and the processor scheduling aspects are enabled through simple to use API’s, this eliminates the need for complex infrastructure enablers to support the underlying tasks.
Micro Level Parallelism/Acceleration - Where the problem is computationally intensive, and one expects to achieve a significant increase in speed. By utilizing the HANSA API’s to distribute the problem components across multiple cores, one can expect gains of more than 200x(Matrix operations) -1000x (BLASTp/n) versus a dual core CPU running at 2.8GHz. The API’s eliminate the need for knowing/understanding hardware design aspects and the familiarity with VHDL.
Expanding the problem domain - Inherent in HANSA’s unique architecture is the ability to handle large data sets through its massive memory of 256GB that are closely tied to the processors; HANSA could operate on matrices up to 64K x 64K without hitting the network.
The HANSA platform is currently available with "native" routines for Java, C++, Matlab and R. Enabling your application with Micro Level Parallelism (and therefore significantly increased speed) is as simple as incorporating these routines into your application. If your library compiles with the standard gcc compiler, a simple recompile on our platform without any code modifications will allow you to leverage the Macro level parallelism. The following are sample implementations for HP-Filter and Trinomial tree option valuation in Java, C++, Matlab and R. The examples reflect micro level parallelism, and include the ability to work on matrices of sizes up to 64K x 64K.
HANSA Implementations
HP Filter
Original Source as implemented in MATLAB on a desktop
Trinomial tree option valuation
Original Source as implemented in MATLAB on a desktop
Take the Next Step
Kuberre's HANSA platform is part of our complete framework of custom solutions for the Investment Management enterprise. To learn more about how we can tailor our solutions specifically to your business, please contact us.