SuperMicro SuperServer MLA V100 (Vulcanite)
Vulcanite is a Linux cluster located at the ERDC DSRC to support HPCMP research efforts associated with developing and deploying GPU-enabled capabilities. Vulcanite is a SuperMicro SuperServer MLA V100 that consists of two login nodes and 39 compute nodes with a mixture of GPGPU configurations.
- 26 compute nodes, each with two NVIDIA V100 GPGPUs
- 8 compute nodes, each with four NVIDIA V100 GPGPUs connected on node by NVLink
- 5 compute nodes, each with eight NVIDIA V100 GPGPUs connected on node by NVLink
See the Vulcanite User Guide for more information.
Vulcanite is a small exploratory system with very limited storage space and therefore not capable of production jobs. But it has Machine-Learning/Data-Analytics software that can use the GPGPUs.
Its purpose is to provide users experience with:
- Using MLA (machine learning accelerator) applications and developing a mythology that enhances their capabilities to process big data sets on HPC systems.
- Developing workflows that foster productivity and produce results that meet the requirements.
It is expected that users will provide feedback to the ERDC DSRC with lessons learned and suggestions on the configuration of future HPC acquisitions that will provide the best value to HPCMP users needing Data Analytic and Machine Learning capabilities.
You must limit your home directory disk usage to 30 GB. Each compute node has a private mounted SSD drive for scratch space during your batch jobs. It is mounted as the /tmp directory.
2. Project Proposal Criteria
To obtain an account on Vulcanite, your proposal must include the following information:
2.1. Research Description
Limit each proposal to two pages. Include the header on each page as shown here:
Vulcanite Research Proposal
<Principal Investigator Name>
Briefly discuss the following topics:
- What is your application (will it use multiple nodes, multiple GPUs, hybrid
- CPU/GPU computation, etc)?
- How does your application use the GPUs (what computations run on GPU, what
- frameworks is used to access GPUs, etc)?
- How will the varying densities of GPUs affect your application?
- How will the on-node scratch SSDs affect your application?
2.2. Usage Estimate
Estimate the overall project duration.
Do you plan or need to use the entire system as part of your investigation?
Duration in weeks.
Number of nodes by GPU configuration:
Number of nodes with 2 GPUs:
Number of nodes with 4 GPUs:
Number of nodes with 8 GPUs:
2.3. Additional Users
List the name and organization for each researcher participating in the investigation. If the researcher(s) does not already have an HPCMP user account, they will need to request an account to gain access to Vulcanite. Information regarding requesting a user account may be found HPC Centers: Obtaining an Account.
2.4. Results and Constraints
Are you willing and/or able to share your results with the HPCMP?
Would you be willing to share performance assessment information in conference papers and publications?
2.5. Submitting Your Proposal
Notice of acceptance should come within two weeks of receipt of access proposal.