Screenshot of PSE visualization software.
David M. Koppelman
A collection of computers for research on microarchitecture.
Room 150F EE Building
- A 100-core GPU accelerated cluster.
- A novel cluster of three vastly different implementations of same instruction set (ISA).
- Funded by multiple BoR grants.
- Facilities for research and education.
The Computer Microarchitecture Research Lab consists of a collection of machines used for computer architecture research. The machines are housed in Room 150F EE Building, a raised-floor machine room, and are accessible from Internet. The lab consists of a novel specialized system for multithreading and computer architecture comparative studies, and a 10-node GPU-accelerated cluster for running simulations. The lab's most novel system is a single-ISA, three-implementation system, which consists of machines of very different design but sharing the same machine language and were designed at roughly the same time. The three machines are a Fujitsu M5000 (SPARC 64VI) named dynasparc.ece.lsu.edu, Sun T5120 (UltraSPARC T2) named bioctooris.ece.lsu.edu, and a Sun V490 (UltraSPARC IV) named static.ece.lsu.edu. All of these are multithreaded machines implementing SPARC V9 and as a set are an ideal platform to study the impact of vastly different design decisions, decisions made at about the same time and to implement the same language. The workhorse machine is petal.ece.lsu.edu, a 100-core, 10-node Dell cluster using Xeon E5560 and E5650 CPUs; five of the nodes are equipped with Fermi-class GPUs. The machine is used for computer architecture research and for some courses. The cluster runs simulations of new computer designs, generating detailed datasets that can drive visualizations needed to gain insights into performance bottlenecks and unexploited potential, or to estimate the overall performance of the designs.
- Fujitsu M5000 (SPARC 64VI) multiprocessor.
- Sun T5120 (UltraSPARC "Niagara" T2) many-core system.
- Sun V490 (UltraSPARC IV) multiprocessor.
- Dell cluster, including five 8-core CPU-only nodes and six GPU accelerated nodes.