Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.
And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.
How to start multiple dask workers with 1 GPU each?
See original GitHub issueI have a compute node with setting: Virtual machine size: (24 cores, 448 GB RAM, 1344 GB disk) Processing unit: 4 GPUs
How to start multiple dask workers with 1 GPU each? For example, I’d like to start 4 dask workers with 1 GPU each and use all resources provided by node.
I have tried the following from document:
"dask-worker {scheduler} --resources GPU=1"
this starts 1 dask worker
"dask-worker {scheduler} --nprocs 4 --resources GPU=1"
this starts 4 dask worker, and client:
Client: 'tcp://10.0.0.6:8786' processes=4 threads=4, memory=73.43 GiB
But It looks like this doesn’t use all resources provided by node.
Issue Analytics
- State:
- Created 2 years ago
- Comments:15 (8 by maintainers)
Top Results From Across the Web
Top Related Medium Post
Top Related StackOverflow Question
Troubleshoot Live Code
Top Related Reddit Thread
Top Related Hackernoon Post
Top Related Tweet
Top Related Dev.to Post
Top Related Hashnode Post
Great! Glad things are working for you . In the future you might want to use something like:
client.wait_for_workers(n_workers=…)
I would recommend using dask-cuda – the
dask-cuda-workeris designed specifically for creating dask workers pinned to GPUs