Use bigger machines to build?

[leofang]

Suggested by @carterbox:

we can sign this feedstock up for the openstack server. Those runners have up to 64GB of RAM.
https://conda-forge.org/docs/maintainer/knowledge_base/#packages-that-require-a-gpu-or-long-running-builds

We recently run into OOM errors when building cuda.bindings. This could help.

[carterbox]

The default value of CUDA_PYTHON_PARALLEL_LEVEL is not 1, it's 0 or num_threads / 2. We should try setting it to 1.

https://github.com/search?q=repo%3ANVIDIA%2Fcuda-python%20CUDA_PYTHON_PARALLEL_LEVEL&type=code