SIMClr does not respect number of workers

[Haydnspass]

🐛 Bug

The SIMClr example does not respect the number of workers or at least not as expected.

I suspect that this is due to the changed behaviour of pytorch 1.7 where the default num_threads is #physical processors whereas pytorch 1.6 had a default of 1.

The actual number of cores being used in this notebook is num_threads * num_workers, while this might be technically true it is highly unexpected (isn't it?).

To Reproduce

Steps to reproduce the behavior:

1. Install pytorch_lightning >1 with torch=1.7
2. Runt he SIMClr example with varying amount of workers
3. Monitor the CPU utilisation.

Expected behavior

Number of cores beuing utilised = number of workers

Environment

• PyTorch Version 1.7
• OS (e.g., Linux): Linux
• How you installed PyTorch (conda, pip, source): conda + pip
• Python version: 3.8
• CUDA/cuDNN version: 11
• GPU models and configuration: RTX 2080 TI

[github-actions]

Hi! thanks for your contribution!, great first issue!