Plaform specific notes

[rwalle]

Memory requirement:

For "Problem (distributed_communication_single_node): Distributed Communication", it would be nice to put a reminder to do a calculation of the memory requirement before running the benchmark. On Linux, CPU code that uses more memory than available will be quickly killed, but it could cause the entire system to freeze and then reboot on Mac.

Other Linux-only requirements:

• nvcc backend is only available on Linux. So for parallelism part, if someone wants to run their code with GPU on Windows (e.g. gaming laptop/desktop), it won't work. They could try their luck with WSL2 though
• triton is only available on Linux as well. There are unofficial Windows builds however

While some problems can be solved with CPU and gloo backend, GPU is a hard requirement for others.

It might be a good idea to mention this in README or pdf to clarify these requirements, so that anyone (including non Stanford students) could set up their environment early (e.g. leasing a GPU).

[marcelroed]

Thank you! This is great feedback. For assignment 2 I think the only reasonable case is to require GPUs for almost all of the tasks. I think requiring Linux or WSL for any GPU tasks is a sensible decision. I will need to do some additional verification on WSL to make sure it works for everything.

I'll add some disclaimers/suggestions for what environments we support based on this.

Let me know if you have suggestions for making it more granular or anything else.