Зарегистрируйтесь сейчас для лучшей персонализированной цитаты!

Новости по теме

Cisco UCS powers HPC clusters for student competition

Nov, 14, 2019 Hi-network.com

In the rolling hills of North Carolina, two renowned universities, Wake Forest University and North Carolina State University (NC State), engage in a friendly academic rivalry. This relationship is about to get tested as both universities have been selected to participate in the Student Cluster Competition (SCC) at SC19, the premier HPC conference, which will take place in Denver on Nov 18-22.

The competition

After an intense application process in the spring, 16 teams were invited to participate in the 2019 Student Cluster Competition. The competition exposes undergraduate students to HPC by having them complete real-world scientific workloads in a non-stop 48-hour challenge. In order to compete, teams must get sponsorship from hardware vendors. The teams then design and build a small cluster on which they will run the HPC benchmarks and workloads.

Cisco is proud to sponsor two teams this year. Wake Forest University, who we also sponsored in 2018 and who finished second out of the 6 first-time competitors, and NC State who is participating for the first time.

Cisco has been working with the students for 6 months, providing them access to the hardware and helping them build HPC clusters in preparation for the competition.

The students come from a variety of academic background with majors in computer science, mathematics, computer and electrical engineering, finance and even one history major. The key challenge for these students is that in order to compete successfully they need to learn a lot of HPC skills that are not part of the standard undergrad curriculum. Many learned the skills over the spring and some even had to stay on campus this summer to work on acquiring these skills and learn how to build an HPC cluster.

The cluster

Both universities will compete using a cluster built on Cisco UCS but using different architectures.

-Wake Forest will use one Cisco UCS C480 ML M5 server, which comes with 8 NVIDIA Tesla V100 GPUs for parallel processing and leverages NVIDIA NVlink for GPU communications and three Cisco UCS C240 M5 servers. The UCS C480 ML is purpose-built for high-parallel processing and used for deep learning algorithm training. The UCS C240 is the workhorse of the data center with a flexible configuration and easy expandability.

-NC State will use two Cisco UCS C480 M5 with 4 PCIe-connected NVIDIA Tesla V100 GPUs each. The UCS C480 M5 offers flexibility in the number and model of GPUs in a 4-socket server.

One of the key requirements of the SCC is the power limitation meant to create an even playing field for the competitors: they must use less than 3,000 watts of power.

Denver-bound

The Student Cluster Competition is a great example of academic and industry collaboration in helping the best and brightest students get real world experience and learn skills that are in high demand today.

Even though the two North Carolina universities will compete fiercely in Denver, Wake Forest has shared their experience from last year with NC State and the two teams have had several conversations about what to expect and how to prepare.

In addition to sponsoring the SCC, Cisco will also have a booth at SC19 where we will showcase our Unified Computing System as well as networking solutions for HPC and AI. So if you happen to be in Denver as well, stop by booth 801 and talk to our technical staff about your needs.

For more information about our UCS portfolio visit: cisco.com/go/ucs

For more information about our infrastructure portfolio for AI visit: cisco.com/go/ai-compute

May the best team win, preferably a team from NC!

 


Join the Conversation

@FrancoiseBRees -@NCState -@WakeForest -#SC19 -@CiscoDC


tag-icon Горячие метки: Cisco UCS HPC (HPC) cluster

Copyright © 2014-2024 Hi-Network.com | HAILIAN TECHNOLOGY CO., LIMITED | All Rights Reserved.
Our company's operations and information are independent of the manufacturers' positions, nor a part of any listed trademarks company.