In 2012, we completed the multi-year HPC resource expansion plan that started in 2007. Our major milestones include exponential growth in research computing capacity, significant speedup improvement in research computation and the launch of HPC Cloud services.

Exponential Growth in Research Computing Capacity

In terms of CPU hours generated, the amount of research computation done on the central HPC systems () had grown more than 16 times since 2006. For the first time, we breached the 10-million mark when 11.9 million CPU hours usage was clocked at the end of 2012. The capacity expansion was made possible through the multi-year resource expansion plan as shown below:

Our multi-year approach, as opposed to a big-bang approach, allowed the latest technologies to be made available to the researchers within a short period of time. For example, the Intel Nehalem and Sandy Bridge processors were made available within a year after they were launched. A GPGPU cluster with the latest GPU and a total of 16,000 GPU cores was also introduced in 2012 (capacity not included in the above chart). Check our website on how to gain access to these resources.

Significant Speedup Improvement in Research Computation

As you can see from the table, we achieved an overall productivity improvement of 571% in the average speedup of all computational jobs. To the researchers, it may mean a simulation that took a month to complete previously can now be completed within days. Some researchers had demonstrated that their applications could achieve speedup ranging from 10 to 80 times through OpenMP, MPI and Grid parallel computing. Find out how you can also achieve similar speedup here.

Launch of HPC Cloud Services

The launch of HPC Cloud Services in March 2012 allows researchers to purchase dedicated HPC computing clusters within hours, as compared to months if they were to purchase and install it themselves. These services also enable researchers to focus on their research rather than the operation and maintenance of the HPC system. The Pay-Per-Use service is recommended for ad-hoc usage, whereas the Condominium service is more suitable for researchers who need long-term dedicated use of the HPC cluster. Details of the services are available here.

Future Direction for Research Computing Support

We foresee an exponential demand for data storage capacity and management solution due to the following factors:

• More research simulations being done with greater speed

• More data being processed and generated

• Compliance requirements for data retention

In the next few years, we will be focusing on developing a scalable and cost effective data storage and management service. At the same time, we will continue to refresh our HPC clusters, bringing in new many-core technologies that are expected to improve the research productivity further. Other existing services such as the HPC Cloud, scientific visualization, research application software and HPC consulting support will also be reviewed and improved over time.