Videos
ETI at Research@Intel Day 2011
Rob Knauerhase of Intel Labs recently demonstrated the company’s Runnemede exascale system research using ETI’s SWARM technology at June’s Research@Intel Day 2011. Intel is currently investigating advanced software techniques using fine-grained execution models to exploit the full power of high core count systems, and Rob uses implementations of the Graph500 with MPI and SWARM to demonstrate the efficiency and speed of problem solving delivered by the ETI solution for many-core systems.
Check out the video on the Intel Software Network.
Light Cube Videos
The following videos demonstrate the difference between current technology (bulk synchronous parallelism or BSP) and ETI's technology (asynchronous dataflow task parallelism or ADTP). As shown on the left, BSP suffers large delays when synchronizing all nodes to each other to commence the next phase of an applications. Shown on the right, ADTP allows the runtime to synchronously schedule tasks when their data dependencies are met. Therefore, the runtime is able to find 'slack' in the schedule and assign new work to cores that finish earlier than others. ETI's technologies employ dynamic ADTP to maximize application performance in the face of unknown conditions or difficult-to-schedule complex fine grained tasks.