The core components of deep learning server are CPU, hard disk, memory and GPU. In particular, many deep learning depends on the large-scale data processing ability of GPU, which requires the computing power and quantity of CPU, and different data have different requirements for GPU memory.
At present, most of them are doing deep learning with RTX3090. The latest RTX4090 has been put on the market, and its single-precision operation ability is twice that of RTX3090. Both GPUs have 24G memory. For example, A 100 emphasizes double-precision computing power, with 40G and 80G video memory, while A6000' s single-precision computing power is similar to that of RTX3090, with 48G video memory, which can be selected as a reference.
Of course, the most important thing is the money in your pocket. The market price of A6000 is more than twice that of RTX, and A 100 is more than 100,000 recently. It is estimated that it will not be listed soon, and the price is out of stock. RTX3090/4090 is low in price and high in cost performance, which is why most people choose them for deep learning. This is the choice of the market.