Director – Power Systems
I attended SuperComputing 2016 in Salt Lake City, and I was amazed!
The attendance was over 16,000, with people from all over globe, there were in-depth lectures and sessions over 4 days, along with over 3 football fields of exhibits. The exhibits ranged from Universities, hardware component providers (eg. GPUs, FPGAs, switches, storage, etc), software providers (eg. Redhat, SUSE, Nice, etc.), and major players in supercomputing like Nividia, Mellanox, Intel, Lenovo, Dell, and IBM.
IBM was very impressive, with their presence of systems, storage, technology, and innovation, offering two days of sessions on various aspects of supercomputing (i.e., HPC).
IBM Systems: IBM showcased the 822LC systems with integrated NVLink to 4 Nividia P100 GPUs, 20 cores of POWER8, and 1 TB of RAM. Sandia Labs gave outstanding presentations on the performance results of these systems that they are using for various HPC workloads.
IBM Storage: IBM showcased the Spectrum Scale storage to provide the massive capacity and performance support that is needed for supercomputing.
IBM Technology: IBM showcased Power CAPI, CAPI-2 in POWER9, and the coming OpenCAPI. CAPI has been around on POWER8 since the announcement of POWER8, and OpenCAPI takes the concept further in POWER9. In OpenCAPI, the interface in POWER9 will enable coherence to a range of attached accelerators like GPUs, memory, NVMe, adapters, ASICs, etc., via an open standard. With POWER9, which was also showcased by IBM, coming very soon (2017) OpenCAPI will enable another boost in performance by enhanced coherency between accelerators and cache in the processors. IBM has also announced the OpenCAPI Consortium where members can cooperate in creating OpenCAPI devices, software, and connections.
IBM Innovation: I think the main “innovation” that IBM was presenting was POWER9 and the “Open” consortiums of Power and CAPI, as well as what they are producing, and what they will produce. IBM has created a programming model and framework for FPGAs and CAPI that will enable C and C++ programming to utilize the capabilities of ASICs. Previously, the programming by vendors (eg., Redis Labs) could have been considered very low level and complex. The new CAPI SNAP (CAPI Storage, Networking, and Acceleration Programming) framework will be available to everyone on Github soon. For a view of CAPI accelerators available today, see ibm.biz/powercapi_examples. Also you can Google CAPI for many other references and technology views.
IBM Sessions on SuperComputing: IBM offered two days of no-charge sessions, on various aspects of supercomputing and HPC. These introduced the topics of Machine Learning (ML), Deep Learning (DL), cognitive, analytics, Watson, and the future needs based around Big Data. As this topic is very broad and insightful, I will comment on that in another blog soon.
In summary, SuperComputing 2016 in Salt Lake City as awesome!
The next Supercomputing will be in Denver, in 2017. I highly recommend your attendance, as supercomputing is not just for shooting rockets to Jupiter and Pluto (yes… it is a planet again!!), but rather, high-speed computing is becoming a reality in all forms of business and business transformation.
Please contact your Mainline Account Executive directly, or click here to contact us with any questions.