The wiki is deprecated and due to be decommissioned by the end of September 2022.
The content is being migrated to other supports, new updates will be ignored and lost.
If needed you can get in touch with EGI SDIS team using operations @ egi.eu.
Difference between revisions of "GPGPU-WG KnowledgeBase Batch Schedulers LSF"
m (moved GPGPU-WG:GPGPU Working Group KnowledgeBase:Batch Schedulers:LSF to GPGPU-WG KnowledgeBase Batch Schedulers LSF) |
|||
| (2 intermediate revisions by 2 users not shown) | |||
| Line 1: | Line 1: | ||
{{Template:Op menubar}} {{TOC_right}} | |||
[[Category:Task_forces]] | |||
'''[[GPGPU_Working_Group| << GPGPU Working Group main page]]''' | |||
LSF version 8.0.1 has some limited support for GPUS, via the esub plugin mechanism. The GPU support is supposed to be improved in LSF 9, but this has not been tested yet. | LSF version 8.0.1 has some limited support for GPUS, via the esub plugin mechanism. The GPU support is supposed to be improved in LSF 9, but this has not been tested yet. | ||
The plugin is capable of setting CUDA_VISIBLE_DEVICES for GPUs allocated to a job by using LSF application profiles, which have to be specified on the command line. However testing discovered some limitations - the environment variable is identical across all ranks of the MPI jobs, even when running on different hosts with different GPUs in use. And this has no effect on OpenCL binaries | The plugin is capable of setting CUDA_VISIBLE_DEVICES for GPUs allocated to a job by using LSF application profiles, which have to be specified on the command line, with a profile each for how many GPUs were required. However testing discovered some limitations - the environment variable is identical across all ranks of the MPI jobs, even when running on different hosts with different GPUs in use. And this has no effect on OpenCL binaries. | ||
Latest revision as of 16:06, 22 January 2015
| Main | EGI.eu operations services | Support | Documentation | Tools | Activities | Performance | Technology | Catch-all Services | Resource Allocation | Security |
<< GPGPU Working Group main page
LSF version 8.0.1 has some limited support for GPUS, via the esub plugin mechanism. The GPU support is supposed to be improved in LSF 9, but this has not been tested yet.
The plugin is capable of setting CUDA_VISIBLE_DEVICES for GPUs allocated to a job by using LSF application profiles, which have to be specified on the command line, with a profile each for how many GPUs were required. However testing discovered some limitations - the environment variable is identical across all ranks of the MPI jobs, even when running on different hosts with different GPUs in use. And this has no effect on OpenCL binaries.