High-Performance Computing Research Update
By Dr. Scott Yockel, HPC Services Manager
In the past 12 months, since I have been in the position of HPC Services Manager, I've witnessed first hand the tremendous growth of research computing here at UNT. There has been a growth of over 50% in new research groups whose numerically intensive computations are now being performed on Talon, as opposed to desktop computing. With this growth CITC has begun expanding Talon to service the needs of the HPC research community. In the past year, we saw a huge demand for largest memory compute nodes (64GB) and have now added 24 of Dell's R410 servers (which include the Intel® Xeon® Processor X5650 with 12 cores) to bring the count to 56 nodes with 64GB of memory. Additionally we added a “fat node” with 256GB of memory (a Dell R910 with four Intel® Xeon® Processor X5650 for a total of 32 cores); this is the largest single computing node on campus. Many of the computational chemistry are excited about this addition as it will allow for accurate thermochemistry of much larger chemical systems to be computed, where many of the matrix elements will be able to stored in memory instead of swapped out onto disk. Calculations which store arrays in memory instead of disk are at least 10x faster.
Additionally, we have added the first compute nodes on campus with an Nvidia's Tesla M2050 GPU computing module; these include 448 GPUs per node! Software codes with CUDA instruction sets have the ability for a 30x speed up in computation time. Professor Tom Cundari hopes that this speed up will allow his groups enzyme kinetics simulations to run long enough (hundreds of nanoseconds) to be able study in detail the temporal effects on catalytic activity as experimentalist have hinted at. Professor Krishna Kavi's group is currently working along side of Oak Ridge National Laboratory to develop tools to help streamline the process of writing software codes for this type of hybrid CPU/GPU nodes and will finally have a resource on campus for a test bed.
The dawn of a new age ...
We are also at the dawn of a new age of computing on campus which is not numerically intensive, but data intensive. A pilot program for a centralized large-scale research storage solution has also begun that will include hundreds of TB of storage. This storage solution is intended for researchers with single data sets larger than 1 TB, and will continue to need easy access to this data throughout their funded project. This storage solution will be interfaced with Talon so that compute intensive research can be performed without the timely process of transferring TBs of data over traditional Ethernet. Lastly, this storage solution will be expandable, and researchers will have the option to use grant money to purchase allotments of dedicated storage beyond the shared storage space. The HPC Services Team is excited about the opportunity this opens up for a whole new direction in research computing at UNT.