For projects that use Deep Learning, the memory size of the GPU on the PC is key.
If the memory size is small, it will be difficult to train large AI models.

To train images from a 5MP class camera, you need a GPU with at least 24GB memory.
So, if you really want to make a Deep Learning inspection system, RTX A6000 (48GB memory) would be necessary.

It is not as expensive as the A100 for generation AI, but the 1 million yen level is cost prohibitive for appearance inspection purposes.
Furthermore, the announcement of the end of production on this A6000. The successor (6000 ada) will be priced at the 500,000 yen level.

After all,
(1) Prepare a PC equipped with a GPU with large memory for Deep Learning training.
Optimize (downsize) the generated AI model for inference.
(2) The actual PC is used by loading the AI model optimized for inference.
Since the size of the model optimized for inference is small, a GPU with a small memory capacity can be used.
The following is a realistic use of the system.

In any case, if you want to do Deep Learning, it would be a good idea to buy RTX A6000 while you can afford it.