Arizona State University (ASU) has announced that its new supercomputer, “Sol,” will launch this summer. Sol, pictured in the header image courtesy of ASU’s Andy DeLisle, is set to supplement ASU’s existing supercomputer, Agave, which the university says is not yet slated for retirement.
Douglas Jennewein, senior director of ASU’s Research Computing division, told HPCwire that Sol—named for the Spanish word for “sun” — is a Dell-built system spanning 178 nodes. The system uses AMD Epyc 7713 CPUs (constituting some 18,000 cores, Jennewein said), with the bulk of the nodes carrying 512GB of memory and five large-memory nodes equipped with 2TB each. There are also 60 GPU nodes: 56 with quadruple Nvidia A100 (80GB) GPUs each and four with triple Nvidia A30 (24GB) GPUs each. The system is networked with Nvidia’s 200GB/s HDR InfiniBand and supported by 4PB of Dell BeeGFS scratch storage as well as 2PB of Dell PowerScale storage. (We expect this all might add up to a few petaflops of theoretical peak performance.)
ASU’s existing system, Agave, has been regularly upgraded since its launch in 2018. “Agave is now more than twice as big as it was when it launched,” Jennewein said. Currently, Agave (also a Dell system) consists of more than 16,000 Intel CPU cores (Broadwell or newer); 1,200+ Intel Phi CPU cores; more than 360 Nvidia GPUs (a mix of 1080 Ti, V100, A100 and more) and 448 AMD Epyc CPU cores. In April alone, Agave completed nine million CPU-hours of computing. But, ASU says, the resources are strained—and getting difficult to maintain.
“We’re sort of hitting our heads on the ceiling with Agave,” Jennewein said. “We’ve filled up all of the network ports on our core network switches and we’re running out of places to upgrade.”
“The Agave system is aging, and as parts age out, it becomes harder to keep them going,” he added. “Replacements are harder to come by and warranties expire.”
Sol, the university said, will offer Agave welcome reprieve, doubling the amount of CPU-hours available to ASU researchers. However, due to its increased size and cooling requirements, Sol will, unlike Agave, be hosted off-site at the nearby Iron Mountain Data Center. And, like its predecessor, Sol is expected to grow substantially: “I think we’ll see at least that same trajectory with Sol [as we did with Agave],” Jennewein said.
Sol is already in early access and is expected to reach general availability to ASU researchers this summer. “We have many bright, bright shining stars at ASU,” said Gil Speyer, director of the Computational Research Accelerator at ASU. “We have such spectacular faculty and I am always just blown away by the things that they do — and I get goosebumps thinking about the processing power that will soon be available to them through Sol.”