kouohhashi

When it comes to inference using cloud services, Tesla T4 GPUs are often used, but they are not cheap. It seems that creating a server room with A4000 workstations or RTX 3060 notes might be more cost-effective, but this could potentially violate Nvidia's terms and conditions.

On the other hand, there are cloud services that advertise that using the A4000 for inference is acceptable. Does this mean that while support from Nvidia might not be available, it is implicitly tolerated by Nvidia?

We see many hot SaaS applications every day. like loom, MilkVideo, and Descript. But I'm curious how they set their prices and why does it make sense? 8$/month or 20$/m seems too cheap. If they use Google's speech to test API, probably they can not make any money. even if they have their own API, there are another costs like AWS and others cloud services.