The Future Technologies Group identifies and evaluates emerging architectures for potential use in high performance computing. This research thrust has recently focused on graphics processing units (GPUs) as accelerators for scientific computing, APUs (fused CPU+GPU processors), and FPGAs, with the goal of studying the impact of increasing core counts and heterogeneity.
Recent results from this effort include:
- The Scalable Heterogeneous Computing (SHOC) Benchmark Suite
- Our recent evaluation of the AMD Llano APU
- Our identification of NUMA effects in multi-GPU systems
In addition to new processors, this effort also addresses changes to the system interconnect and storage. In particular, recent efforts in the Blackcomb project have focused on the potential benefits of using non-volatile memory (NVRAM) as a complement or replacement or DRAM.