RTX2080 Testing

From ScientificComputing
Revision as of 08:41, 30 April 2019 by Urbanb (talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Introduction

We have received four GPUs for evaluation:

  • 2 nVidia GeForce RTX 2080 and
  • 2 nVidia GeForce RTX 2080 Ti

All four GPUs are installed in a single node with two Intel Xeon Gold 6140 CPUs (2.3 GHz) and 256 GiB of RAM.

Initially only jobs with a runtime up to 24 hours can use these nodes.

Usage

Any RTX GPU

Use the -R rtx bsub option to request either of these GPUs:

bsub -R rtx ./my_gpu_program

Specific RTX GPU

Due to limitations of the batch system you can not request a specific one of these GPUs. Instead you need to request all four GPUs

bsub -R rtx -R "rusage[ngpus_excl_p=4]" ./my_gpu_program

and then programatically select the desired GPU using CUDA. The first two GPUs are 2080 Tis while the last two are the 2080 non-Tis.

For example to use only two RTX 2080 Ti GPUs, run:

bsub -R rtx -R "rusage[ngpus_excl_p=4]" "CUDA_VISIBLE_DEVICES=0,1 ./my_gpu_program"

To use only two RTX 2080 GPUs, run:

bsub -R rtx -R "rusage[ngpus_excl_p=4]" "CUDA_VISIBLE_DEVICES=2,3 ./my_gpu_program"