Last Tuesday, VR-Media and A*STAR A*CRC held their inaugural event, the first in a series focused on various aspects of high end computing. Here we bring the summary and key highlights.
After a long lull in the processor and system field, where the usual was just happening, i.e. yet another generation of Intel Core XX processors taking over the performance lead from one another, and AMD vs Nvidia on the GPU front, this past year brought along some refreshing news. AMD Fusion APU approach is one of them, whose tight integration of CPU and GPU will likely be followed by other processor vendors in the 2012 as well. And, since that year is to see stronger market entry by Chinese RISC CPU vendors as well, the multi-platform excitement of the years long gone by is likely to come back again.
Such innovations often find their first home in scientific and research High Performance Computing centre sites – in Singapore and this region, A*STAR's A*CRC is the leader in the field, both in the variety of large systems they have, as well as the expert staff strength. VR Media, the parent company of VR-Zone now presents the new development in system level technologies together with A*CRC through a series of events, the first of which was sponsored by AMD, HP and Kingston to coincide with the launch of 'Interlagos', the server version of the famed Bulldozer CPU platform.
The opening keynote speech by Dr Marek Michalewicz, Director of A*CRC, as well as technical brief by Stephen Wong, the Deputy Director, focused on supercomputing in Singapore as well as next generation large computer challenges respectively, in a broad sweep from visualising when Singapore should have its first petaflop system to the new platforms that we may see in such machines this coming year.
On the other hand, the two AMD evangelists, Sasa Marinkovic and Terry Makedon, both themselves GPU gurus from their ATI days, described the usage and programming benefits of the Fusion approach to a great detail. Their coverage also touched on how the end users and everyday consumers would benefit from the new approach. Other speakers from AMD and HP covered the new server CPUs as well as the systems using them.
The event saw over 80 HPC experts from Singapore, and a few more from the region, gather at A*Star's Biopolis in this first of our joint events. After the event, the feedback from the attendees, especially those involved in system design or programming, was that the new programming approaches like OpenCL for Fusion Systems Architecture open up a lot of latent unused performance in current systems, enabling some supercomputing code to even run on laptops, but the challenges are quite high still on how to extract that performance in the most efficient way until fully shared memory between CPU and GPU is enabled – something the next generation of GPUs and accelerators should enable. Also, the heterogeneous computing idea, where the application can efficiently use the different types of CPUs, GPUs, in clusters of hundreds of them both locally and in some kind of 'HPC cloud' where the different latencies, performance and architecture types have to be handled, can be a dream from the point of literally unlimited performance pool available, yet a very stressful exercise in actually getting to use all that capacity.
Overall, in the course of a few days after this inaugural event, the feedback was excellent and we are on course for more such events. Our readers are welcome to suggest, from your point of view, what could be interesting topics for the future forums – we are listening.