Hey @SaschaEsser
If the customer is wanting the "gold standard" then they would want the NVIDIA H100, this is a quite expensive card, but it has the potential for really any deep learning tasks that you can come across, these are used by Google and other extremely large companies for their Deep Learning and AI tasks.
Another good option is the NVIDIA Tesla series, specifically the V100, I've used this in a few of my own deep learning projects, very reliable, and performs amazingly, but of course, it's affordable for an average researcher, so it has its limitations.
You can also use gaming graphics cards, specifically 3090, 4090, these are very powerful cards and can definitely hold their own, I've seen these used in deep learning tasks specifically.
If you are going to go with the H100, you will need a server system with a solid cooling setup, these will draw a lot of power, and with 80GB of VRAM they will be holding their weight nicely. These also do not have any video output, so they dedicate their work directly through the PCIE.
Hope that helps!
Cody
