ARC hosted a user's meeting on April 6, 2021 to share updates and ongoing developments at ARC. The slides from the presentation are available here ARC_user_meeting_6Apr2021.pptx and contain much of the information which was covered and links to some resources that may be useful to the research computing community at Virginia Tech.
Twice each year, the Division of IT at Virginia Tech organizes a Departmental Computing Support Symposium (DCSS) to help comminicate updates or important information which may be of value or interest to anyone in a role of computing support at Virginia Tech.
VT OpenPOWER Hackathon –
April 19th through May 3rd
The VT OpenPower Hackathon is a first-of-its-kind AI (artificial intelligence) Hackathon at Virginia Tech that will focus on accelerating AI model training, with the ambition to reap and share the power of AI technologies with Virginia Tech’s student community. Prizes will be awarded to winning teams: 1st place = $2,000, 2nd place = $1,000, and 3rd place = $500!!!
The Hackathon is being organized by the OpenPOWER Foundation Academic Workgroup, in collaboration with Virginia Tech faculty from Advanced Research Computing, and the IBM Systems Client Experience Center. The event takes place between April 19 and May 3, 2019. The initiation and problem statement will be shared on Friday, April 19 (Hackathon Launch Day). Hackers will have two weeks to work independently. The Hackathon will end with power hackers submitting projects to be presented and judged on the Hackathon Awards Day, Friday, May 3, 2019.
Participants will help to create a new wave of high performing AI models that have the potential to impact all our lives.
The theme of the Hackathon is “Accelerating AI at Scale.” The goal is to demonstrate that optimizing the time needed to train a model can significantly impact the time required to obtain results. Students will be encouraged to use an existing model they have trained on Power Systems and demonstrate how their training time can be accelerated using the latest state-of-the-art techniques on the same Power Systems to the same accuracy.
Details of the event will be provided two weeks prior to the May 3rd judging day, allowing participants to prepare and access system resources to develop their submission. The aim of the competition is to unite students from different creative fields to accelerate the training of AI model prototypes using AI tools in a limited amount of time. The competition will test the students’ fortitude and ability to work under pressure.
Location/Dates and Times
Hackathon Launch Day - Friday, April 19th, 11 AM – 2 PM, 1100 Torgersen Hall
Hackathon Awards Day - Friday, May 3rd , 10:30 AM – 2 PM, 1100 Torgersen Hall
How to prepare for the Hackathon
Students are recommended to select a starting Machine Learning/Deep Learning model they are working on or to select a model from other sources (github, kaggle.com, modelzoo, etc.) from which they can collect a baseline training time and test accuracy.
Students will evaluate ways to accelerate the training time for their model by exploring advanced techniques and the use of:
- Multiple GPUs with a server
- Multiple GPUs across multiple servers
- Accelerated machine learning libraries
- Other innovative approaches
Come to the Hackathon Launch Day to learn about the technologies you can apply. Be ready to put the final touches on your accelerated model to demonstrate your test accuracy and speedups at Hackathon Awards Day. We encourage the use of open datasets and the sharing of code, jupyter notebooks, etc., as part of the Hackathon’s skill- and team-building goals.
Technologies and Available Resources
Some of the more popular training acceleration tools and techniques will be covered on Launch Day and can be found in IBM PowerAI on the AC922 Cluster at VT (Huckleberry – arc.vt.edu) including:
- Tensorflow benchmark samples
- IBM Distributed Deep Learning
- IBM Snap ML library
- Nvidia RAPIDS.AI library
- IBM Large Model Support (note the need for higher resolution datasets)
- Frameworks optimized for POWER8
Review the PowerAI education and technical resources to get started.
The Hackathon is open to all Virginia Tech students. Register soon, as space is limited! To register, create a team and follow this link - https://forms.gle/oDaqjUNYG6Zhz5gC6
Please note that teams should be 2-6 members (in the spirit of fairness).
We also have an open invitation for industry experts to serve as mentors during the Hackathon, helping students refine their ideas. If you are interested in supporting this event as a mentor, please contact email@example.com. For other questions/concerns, please email firstname.lastname@example.org.
- We encourage participation in teams of 2-6 participants to foster collaborative sharing of skills and knowledge.
- We want to see innovative code, not just PowerPoints! However, creative presentation of the results and re-use of code is also important and encouraged.
- Access to the VT Power Systems cluster resources will be provided.
- Participants are not restricted to any AI framework, but models should be machine learning or deep learning models.
- On various occasions throughout the event, photographs and video will be taken. By applying, participants are consenting to being photographed and filmed.
Be Excellent To Each Other: This is, by all means, the most important rule. This is a very diverse environment, made up of people from many backgrounds. Any behavior or presentation that objectifies or belittles others will be halted, with the presenters immediately disqualified and removed from the event.
- Utilization of AI Technology (30 points): How well does the concept follow our theme? How well is the technology implemented? Does the project scale?
- Impact (30 points): Is there a clear benefit for the solution? How much acceleration was achieved? Did the accuracy of the model suffer?
- Creativity (20 points): How creative was the team in developing an innovative solution for the challenge?
- Technical Effort (20 points): Is the project technically impressive? Complex? Does it seem remarkable that someone could achieve this acceleration as part of the Hackathon?
Prizes (All prize are in USD)
- 1st Place: $2,000
- 2nd Place: $1,000
- 3rd Place: $500
Launch Day Agenda
11:30 AM- Noon: Hackathon Launch!
Noon: Break for Pizza
12:30 – 2:00 PM: Hackathon Tools you can use
Awards Day Agenda (subject to change based on # of teams)
10:00 AM: Coffee, Tea, and Check-In
10:30 – 11:00 AM: Welcome and Overview
11:00 AM - Noon: Judge /Mentor Walkthroughs and Questions
Noon: Sponsored Lunch
12:40 – 1:40 PM: Final presentations (10 min max per team)
1:40 – 2:00 PM: Hacking Ends and winners announced!
Even though the hacking is over doesn’t mean that you shouldn’t carry on with the great work you started! We encourage you to keep going, consider becoming a mentor in future hackathons, and share your results with the community.
Inside HPC News Story:
The Institute for Creativity, Arts, and Technology (ICAT) had already forged a partnership with the university’s Division of Information Technology, which created a high-speed connection to its high-performance computing networks for the institute’s data-intensive projects in the Cube.
Virginia Tech’s Advanced Research Computing (ARC) group in the Division of Information Technology offers shared high-performance computing resources to support the Virginia Tech research community, as well as non-research users who need more high-end computing power than departmental systems can provide. Offering high-performance computing services on ARC’s clusters means the computational processing, management, and data storage needs are all serviced remotely.
From any ARC login node, you can explore how to use ARC software by checking the directory /opt/examples
Details on the examples directory are in
The gfortran, ifort, and pgf90 compilers can be instructed that variables declared as "real" should actually be stored and handled as double precision quantities. Details are at https://188.8.131.52/userguide/fortran_realsize/
Click to learn more about the events on April 9-10th, 2018 at Torg. 1100 and the Inn at Virginia Tech.
The Visionarium Lab (Torgersen Hall 3050) is open for drop-in hours M W F 12:00-1:00 pm.
A COMSOL batch job can run in parallel, using multiple nodes or processors. Information on how to automatically set the appropriate COMSOL command line parameters for parallel execution is available, including an example script, athttps://secure.hosting.vt.edu/www.arc.vt.edu/userguide/comsol/.
ARC will offer many free classes on scientific computing, sponsored by Network Learning Initiatives (NLI) and held in room 1100 Torgersen. NLI encourages you to register for these classes through their web site https://nli.tlos.vt.edu, but unregistered auditors are welcome to sit in.
For a list of dates, times, and titles for the classes, refer to