ICES - GPU computing for scientific simulations

General-purpose graphics processing units (GPU) can provide a significant part of the computational power used in scientific computing applications. Mastering parallel computing with these devices is therefore very interesting and becoming increasingly important. This talk gives a brief overview of GPU technology, available hardware and computational performance to expect from it. We will discuss a few use cases that show the spectrum of GPU computing in scientific simulations as well as KTH’s support for it.

Do not miss Session #5 with Andreas Nordin from Realtime Embedded.

Adapting software for hardware acceleration - bottlenecks, caveats and benefits.
The presentation will cover
• Available options - SIMD instructions, GPU, programmable logic, coprocessors
• Expected performance improvements
• Typical issues - bottlenecks, caveats, latency
• Planning ahead when designing algorithms, architectural issues