Imperial College London > Talks@ee.imperial > CAS Talks > Making Heterogeneous Computing Accessible

Making Heterogeneous Computing Accessible

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Grigorios Mingas.

Heterogeneous Computing is a computing paradigm that attempts to make use of the growing viability of alternatives to CPUs such as GPUs and FPG As, as well as advances in Multicore CPUs. The key challenge is how to use these radically different platforms cooperatively, so as to not only achieve the aggregate of the constituent platforms’ performance, but exceed it.

I will argue that an application domain specific approach is an effective strategy for realising the opportunity of heterogeneous computing. The domain specific methodology focuses on supporting a thematic grouping or domain of high level computational tasks as opposed to generic logical or arithmetic operations.

For heterogeneous computing, I will demonstrate three benefits for this domain specific approach:
  1. End users can express their problems for an array of platforms with concepts that they are familiar and comfortable.
  2. Knowledge from the domain can be used in the optimisation of the platform-specific implementations of tasks.
  3. Partitioning of tasks across the available platforms is enhanced using relationships that are made explicit in the application domain.

To illustrate this demonstration, I will use the example of the domain of forward looking derivatives from computational finance. I have built a domain specific application framework, the Forward Financial Framework (F3) which supports valuation of forward looking financial problems upon Multicore CPUs, GPUs and FPG As. For a benchmark of Option Pricing Tasks (link), F3 performs better than manual, platform-specific implementations of individual tasks. Furthermore for the full set of benchmark tasks, F^3 is able to partition tasks across the available platforms more efficiently than a random or heuristic-based partitioning schemes.

This talk is part of the CAS Talks series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

Changes to Talks@imperial | Privacy and Publicity