Official Competition Rules, Regulations, and Details
A team consists of five or six students with up to two official coaches/mentors. Additional mentors can work with the teams at any time except during the actual competition period from Marchth to March 19th, 2021. During that period, no coaches or outside mentors can help students.
Teams are typically composed of students from the same university or associated with a research institution. However, teams can be from different institutions as long as they compete under a common team name.
Any student who has not received an initial degree by July 30th, 2020 is eligible to participate in this competition.
Students who have received their two-year degree but have not received their four-year degree by July 30th, 2020 are eligible for this competition as are masters students who have not received their second degree by July 30th, 2020.
High school students of any age are also eligible to compete either as an entire team or participating as part of a university team.
Masters or PhD students are eligible, but for each Masters/PhD student added to a team, the team will be reduced by two (2) undergraduate students. Thus a team with one PhD student can only have an additional four undergraduate students.
Each team can have as many informal coaches or mentors as they want, however, they can only have a maximum of two official coaches.
Coaches/mentors do not have to be faculty members of the team institution.
Teams are encouraged to ask their sponsor for assistance during the months leading up to the competition.
Teams cannot receive any assistance from coaches/mentors during the competition period (March 15-19).
Each team will have at least one sponsor which will provide technical assistance, remote hardware access and general support for the team.
Teams are not allowed to acquire their own sponsor. All sponsors will be acquired by the Winter Classic Invitational Student Cluster Competition sponsorship committee.
Sponsors and teams are encouraged to closely interact with the goal of forming a tight and successful partnership during the competition.
Sponsors should provide practice clusters well in advance of the competition and ensure that students receive training on how to access and use the cluster properly.
Applications: Four HPC applications that are being used daily to help battle the COVID-19 pandemic.
OpenFOAM: (Open-source Field Operation And Manipulation) is a toolbox typically used for computational fluid dynamics. OpenFOAM is currently being used to analyze and model aerosol transportation and virus exposure for COVID-19. Open source, scales well, and has versions for CPU and GPU processors. Read more about it here.
LAMMPS: (Large-scale Atomic/Molecular Massively Parallel Simulator) software is used for molecular dynamics simulation. Open source, highly scalable and widely used. There are versions that use GPU and other accelerators as well. Read more about it here.
NAMD: (Nanoscale Molecular Dynamics) is used for molecular dynamics simulation. NAMD is being used today to run groundbreaking COVID-19 simulations on the Frontera supercomputer at the Texas Advanced Computing Center. Opensource, highly scalable with CPU and GPU versions. Read more about it here.
Gromacs: (GROningen MAchine for Chemical Simulations) is a molecular dynamics application typically used to simulate proteins, lipids and nucleic acids. GROMACS is being used to simulate COVID-19 on supercomputers around the world, including the #2 Summit system at Oak Ridge National Laboratory in the US. Here’s a tutorial and a user manual.
These are very well documented applications. There is a wealth of information on the internet about how to install, run, and optimize performance. Student teams will be provided with a sample data set to use for practice sessions.
Hardware is not allowed to be rebooted during the entire period of the competition, March 15-19. If a system encounters a failure, inform competition committee immediately for reboot permission and observation.
Students will be provided with access to clusters that have 80 TFLOP/s theoretical Rpeak performance.
Work with your vendor sponsor to come up with the best configuration to tackle the applications in the competition.
Students will be able to configure CPUs, memory, GPU accelerators and other components, depending on your sponsor.
Winter Classic Competition staff will be available to handle questions during the competition.
In order to get credit for completing an application, students must write an Application Brief for each benchmark and application they run.
An “Application Brief” is a short presentation that must include the following components for each benchmark/application:
The ‘out of the box’ initial performance of the application on their cluster. This means the results from the first successful application/benchmark run on your cluster.
Steps your team took to understand the application and optimize it.
Final performance results on the application – we’re interested in your speedup and how you achieved it.
If you had problems with the application, please let us know what they were and if/how you overcame them.
Application Briefs will be THE key factor for judging how well your team did in the competition and your final position in the field.
Hardware Rules & Regulations
All hardware used must be announced and available for sale at the time of the competition. In other words, no NDA systems allowed, and all technical specifications of the cluster (including software stack) must be provided to the competition committee one month before the start of the event.
No overclocking of hardware is allowed, this includes CPUs and GPUs. The 80 TFLOP/s system cap is in place to provide a level playing field for all. Here’s a handy calculator for figuring out your CPU peak TFLOP/s (remember convert the GFLOP/s result to TFLOP/s by dividing by 1,000). CPU cores x CPU Frequency x Instructions per cycle.
For the GPU portion of your configuration, use these numbers:
For NVIDIA V100 processors, regardless of memory, use 7.8 TFLOP/s per GPU x # of GPUs in your configuration.
For NVIDIA A100 processors, regardless of memory, use 9.7 TFLOP/s per GPU x# of GPUs in your configuration.
Add your CPU TFLOP/s and GPU TFLOP/s to get your total system TFLOP/s number
We will be providing you with a piece of code that will allow you to see your current cluster competition in real time. We will randomly require you to provide us with a screen shot of the output of that application which will include a time stamp. This is to ensure that all configurations remain consistent during the competition.
Software Rules & Regulations
Teams may use any o/s and software stack desired, provide it is approved by their hardware sponsor.
We recommend that teams preload their entire software stacks on their cluster well before the competition begins.
Teams should also be working to optimize their applications before the competition begins.
Each team will receive a webcam and will be required to use the provided webcam for team interviews, judging sessions, etc, unless we use another tool like Zoom.
Teams can expect to be interviewed three or four times during the months before the competition and during the actual competition as well.
Publicity interviews will be hosted on the competition website as well as on major HPC publications like HPCwire, InsideHPC, and others.
The Brueckner Award
This is a scholarship named after Rich Brueckner, long time tech journalist who passed away this year. He was a big fan of the cluster competitions and would love to see that his name is now associated with this competition.
The award is two $2,500 scholarships for the most outstanding male and female competitor.
In order for students to be eligible for the award, they need to submit:
A copy of their undergraduate transcripts.
A one page essay covering why they think they are the best choice for this award, their background, academic performance, etc.
These materials need to be sent to the competition committee by March 1st, 2021.
All submissions will be kept strictly confidential.
Tentative Competition Timeline
November: sponsors matched with teams, sponsors announced
November/December: teams work with sponsors to configure systems, work on systems, and optimize competition applications
February: Final technical specifications for each cluster must be provided to the competition committee
Monday, March 15th: Competition begins with “Benchmarks Rise & Shine” where students optimize and turn in results for HPL and HPCG. Once they turn in results for the benchmarks, they get the data set(s) for their applications.
Tuesday – Friday, March 19th: Students work on applications, turning in all results and application briefs by end of day Friday.
Monday – Thursday, March 22nd -25th: judge interviews will all teams