VT GPGPU
General Project Information
- Leader: John Walsh (TCD, Ireland)
- Mailing List: to be setup
- Status: Proposed
- Start Date: -
- End Date: start date + 6 months
- Meetings: -
Motivation
GPU computing or GPGPU is the use of a GPU (graphics processing unit) as a co-processor to accelerate CPUs for general purpose scientific and engineering computing. The GPU accelerates applications running on the CPU by offloading some of the compute-intensive and time consuming portions of the code. The rest of the application still runs on the CPU. From a user’s perspective, the application runs faster because it is using the massively parallel processing power of the GPU to boost performance. Experiments with the use of GPGPUs for scientific computing are already ongoing within various EGI institutions. This project aims to integrate these efforts from EGI partners, define and implement a roadmap to integrate GPGPU services into the European Grid Infrastructure for existing and potential EGI user communities.
Output
add output or expected output
Tasks
- Collect requirements for GPGPU usage from generic and community-specific user support teams (VRCs, NGIs, VOs, projects)
- Investigate how the collected requirements could be addressed on the production infrastructure (technologies, service configuration, policies)
- Make a proposal for the introduction of new technologies, service configurations and policies to the production infrastructure
- Implement the proposed technologies, configurations, policies on selected sites/VOs of EGI
Members
- Name 1, Affiliation
- Name 2, Affiliation
Resources
The project is seeking for members to carry out the following activities, answer the following questions:
- Requirements, existing practices:
- Collect requirements and best practices from existing and potential new user communities of EGI, from NGIs
- Information system:
- Glue Schema requirements/definition and Information System plugins (Glue expert)
- Benchmarking? What benchmark to use?
- Capabilities - speed, #streaming cores, memory etc? (For example, to consider some of the potential capabilities, see: http://en.wikipedia.org/wiki/Comparison_of_Nvidia_graphics_processing_units)
- Batch integration support:
- Feedback on experience for i) AMD GPGPUs, and ii) LSF, (S)GE and SLURM
- How do we ensure secure exclusive access to the GPGPUs?
- How do we deal with multiple GPUs on a single physical card (e.g. Nvidia S2050 has 4 GPUs on single PCI-e card)?
- Other features such as GPU peer-to-peer support?
- Accounting:
- Do we need to extend the accounting system. Torque only reports the CPU time?
- Runtime support:
- API frameworks, versions. What other applications are compiled against CUDA - e.g openmpi, torque?
Progress
- Task 1
- Task 2
- ...
- Task N