> I am assuming you have the necessary hardware to play around
Can you expand on that? Is it enough to have an nvidia graphic card that's like 5 year old, or do you need something more specific?
A 5-year old card, i.e. an NVIDIA Ampere RTX 30xx from 2020, is perfectly fine.
Even 7-year old cards, i.e. NVIDIA Turing RTX 20xx from 2018, are still acceptable.
Older GPUs than Turing should be avoided, because they lack many capabilities of the newer cards, e.g. "tensor cores", and their support in the newer CUDA toolkits will be deprecated in a not very distant future, but very slowly, so for now you can still create programs for Maxwell GPUs from 10 years ago.
Among the newer GPUs, the RTX 40xx SUPER series (i.e. the SUPER variants, not the original RTX 40xx series) has the best energy efficiency. The newest RTX 50xx GPUs have worse energy efficiency than RTX 40xx SUPER, so they achieve a somewhat higher performance only by consuming a disproportionately greater power. Instead of that, it is better to use multiple RTX 40xx SUPER.
I'm not a CUDA programmer, but AIUI:
- you will want to install the latest version of CUDA Toolkit (12.9.1)
- each version of CUDA Toolkit requires the card driver to be above a certain version (e.g. toolkit depends on driver version 576 or above)
- older cards often have recent drivers, e.g. the current version of CUDA Toolkit will work with a GTX 1080, as it has a recent (576.x) driver
each nVidia GPU has a certain Compute Capability (https://developer.nvidia.com/cuda-gpus).
Depending on the model and age of your GPU, it will have a certain capability that will be the hard ceiling for what you can program using CUDA
When you're just getting started and learning that won't matter though. Any Nvidia card from the last 10 years should be fine.
Recognizing that this won't result in any useful benchmarks, is there a way to emulate an nvidia gpu? In a docker container, for instance?