Difference between revisions of "VT GPGPU Use Case Computer Science"
Jump to navigation
Jump to search
Line 1: | Line 1: | ||
{{Template:Op menubar}} {{TOC_right}} | {{Template:Op menubar}} {{TOC_right}} | ||
[[Category:Task_forces]] | [[Category:Task_forces]] | ||
<b>[[GPGPU-WG:GPGPU_Working_Group| GO TO THE GPGPU WORKING GROUP]]</b> | |||
= Scientific Discipline = | = Scientific Discipline = | ||
Computer Science - Hybrid CPU/GPGPU workload management | Computer Science - Hybrid CPU/GPGPU workload management |
Revision as of 07:56, 18 February 2014
Main | EGI.eu operations services | Support | Documentation | Tools | Activities | Performance | Technology | Catch-all Services | Resource Allocation | Security |
Scientific Discipline
Computer Science - Hybrid CPU/GPGPU workload management
Institute or Community
- Trinity College Dublin, Ireland
Keywords
- MPI
- OpenCL
- Hybrid CPU/GPGPU
Description
We have investigated a new dynamic workload sharing algorithm using the CPU and GPGPUs. A hueristic is used to determine how much work should be assigned to a given GPGPU or CPU. We use OpenCL with multiple ICDs (Nvidia/intel/AMD) to ensure that we have access to the CPU as an OpenCL device. OpenMPI is used to help distribute work across multiple machine.