OSG has limited support for GPU jobs. To request a GPU slot, simply use the HTCondor request_gpus attribute in your submit file, as well as request_cpus and request_memory. Example:
request_gpus = 1 request_cpus = 1 request_memory = 4 GB
Currently, a job can only use 1 GPU at the time.
Just like CPUs, GPUs are shared with the OSG community only when the resource is idle. Therefore, we do not know exactly what resources are available at what time. When requesting a GPU job, you might land on one of the following types of GPUs:
- Tesla M2070
- Tesla K20
- Tesla K40
- Tesla P100
- GeForce GTX 1080
It is currently not possible to specify exactly what type of GPU you want, but you can match on for example CUDA compute capability. For example, use the following requirements expression in your job:
requirements = CUDACapability >= 3
A good example on how create a software stack for GPU use is our TensorFlow example:
This page was updated on Sep 22, 2021 at 17:34 from start/resources/gpu-jobs.md.