PSC’s Bridges supercomputer
The Bridges supercomputer was installed at PSC in 2016 and ran until 2021. It was based on ~800 HPE Apollo 2000 nodes, along with a number of K80 GPU nodes. I contributed to the design of the Omnipath interconnect. Conventionally, a chassis switch would be used to connect all components. Here, we used exclusively edge switches, which saved on cost and reduced the compute->storage node latency. This was enabled by the limited presence of jobs running on >42 nodes. This project was funded by a $17.2M grant from NSF.
