Imperial technology workshop on software technologies for accelerated computing
Important: this event has passed!
Many software products that use computationally-intensive algorithms may benefit from exploiting accelerators, such as Graphics Processing Units (GPUs). For certain workloads, accelerators potentially offer much higher performance and power efficiency than symmetric multicore processors. Unfortunately, to realise this potential software developers have to resort to tedious and error-prone low-level programming, which affects code portability and maintainability.
The Accelerator Revolution: Accelerator architectures are developing extremely quickly — with existing products such as IBM’s Cell BE blades, ClearSpeed’s Advance cards, NVIDIA’s Tesla and AMD’s FireStream cards, as well as more exotic options such as FPGAs, and new product announcements coming out every few months.
The Software Challenge: Software developers are faced with a spectrum of software technologies to gain access to the potential power of accelerators. Some progress can be made with automatic parallelisation, and OpenMP. Hand-optimised libraries are an essential consideration. NVIDIA's CUDA has proved effective, and the Khronos Group's OpenCL standard may extend this success. Meanwhile, promising higher-level software technologies are emerging from R&D, including Intel’s Ct and TBB, and Codeplay’s Sieve C++.
The technical programme will include the following topics:
· Performance, power and reliability of accelerators
· Strategies of porting applications to accelerators
· Expressing parallelism and orchestrating data movement
Calculating in single, double and mixed precision
· Generating efficient low-level code from productive high-level specifications
Professor Paul Kelly: Professor Kelly leads Imperial's Software Performance Optimisation research group (http://spo.doc.ic.ac.uk). His research spans compilers, architectures, operating systems, and programming languages – with a particular focus on software tools for exploiting multicore hardware. Recent work has focused on “active”, self-optimising libraries, programming models for managing data movement, and whole-program pointer analysis. Professor Kelly has published more than 60 research articles and is a member of the programme committees of several international research conferences:
Program committees:
§ Vice-chair, Software Track, IPDPS 2008
§ Programme committee member, Compiler Construction 2009
§ Programme committee member, GPCE 2007
§ Programme committee member, Algorithms Track, IPDPS 2007
§ Programme committee member, International Conference on Supercomputing (ICS 2006)
§ Programme Committee member, EuroPar 2006, Topic 4: Compilers for High Performance
Dr Anton Lokhmotov: Dr Lokhmotov is a Research Associate at Imperial’s Software Performance Optimisation research group. His research interests include software development tools, programming languages, computer architecture and high-performance computing. Dr Lokhmotov received a PhD in Computer Science from the University of Cambridge and an MSc in Applied Mathematics from the Moscow Institute for Physics and Technology.
A workshop fee of £190+VAT is required.
Please contact, Anne Mathiot, Assistant Project Manager, Tel: 0207 594 1558, Email:
Ticket: Tickets to be purchased in advance
Monday 8th December 08, Ballroom, 58 Prince's Gate, South Kensington Campus