Already well known for their exceedingly eclectic and downright COOL videos, the SUSE team applied their well-honed skills to capturing all of the presentations delivered in their booth on the show floor at SC17 last month in Denver. In our wrap-up post for SC17 we promised to follow up regarding these presentations; this is the first of those three posts.
Other than an allusion in an HPCwire article this past February, you might regard Univa and SUSE as operating in ‘stealth mode’ when it comes to making public our mutual involvement in the TSUBAME3.0 supercomputer at the Tokyo Institute of Technology (TITECH). SC17 then, served as the ideal opportunity to start emerging from this mode, by sharing some of the recent developments from the Univa perspective.
In this first of our three talks, emphasis was placed on various projects Univa has completed to support the unique architecture of the TSUBAME3.0 supercomputer. If you have any interest in executing HPC and Deep Learning workloads via containers that are capable of leveraging GPUs, I can assure you that this presentation will whet your appetite. Because we anticipate you’ll also want to have some hands-on time, you’ll be delighted to know that we’ve rolled these TITECH-motivated enhancements into our mainline implementation of Univa Grid Engine – in other words, you don’t require access to a TSUBAME-style supercomputer to immediately reap the benefits of our efforts.
The video of the talk is available via Youtube on the SUSE channel (see below), whereas you’ll find the slides here. Key to our ability to deliver enhancements such as these is the unique way we abstract GPUs via Resource Maps (RSMAPS) in Univa Grid Engine. If you’d like to explore this enhanced abstraction for resources, our support for GPUs, and/or our integration with Docker containers, we encourage you to contact us and thereby join our ever-growing list of innovative customers like TITECH.
Please stay tuned for the details on the other two talks we delivered in the SUSE booth at SC17.