Fermi GPU:
Released in 2010, NVIDIA engineers set out to design a new GPU architecture. The architecture defines a GPU's building blocks, how they're connected, and how they work.
Named after the Italian nuclear physicist Enrico Fermi, Fermi is to incorporate full geometry processing to the GPU, enabling a key DirectX 11 technique called tessellation with displacement mapping.
Kepler GPU:
Released in 2012, NVIDIA unveiled its plans to join the cloud gaming space. They’re developed specifically for the GRID and efficient in HPC setup. It has Graphics Processing Clusters (or GPCs), each of which contained a raster engine and four Streaming Multiprocessor (or SM) units. GPU Boost for automatic overclocking.
Maxwell GPU:
Released in 2014, Maxwell-based GPU has new Dynamic Super Resolution (or DSR) that’s designed to put 4K-quality content on lower-resolution screens. Maximum Ram will 12gb (Titan X or Quadro M6000) whereas Pascal 32GB of video memory. Quad SLI in Maxwell upto 4 Graphics card whereas Pascal upto 8 Graphics Card.
Pascal GPU:
Released in 2016, Pascal is designed for the best scalability and Pascal will be priced at a premium because it is a more expensive product as HBM2 and NVLink included.
HBM2 for Bandwidth and NVLink to improve unified virtual addressing scheme across the GPUs and between CPUs and GPUs. So it has capable of allowing data to move 5% to 12% faster across CPUs and GPUs compared to PCI-Express.
Conclusion:
Even after a new technology is introduced, sometimes the old one can be had at a much cheaper price and therefore continues to be a good price/performer since there is no end of life.
25 Times More Performance from Fermi to Kepler GPU
35 Times More Performance from Kepler to Maxwell GPU
10 Times More performance from Maxwell to Pascal GPU
Also, when compare the three generations of Graphics card on their raw performance alone, the Pascal does better than the Kepler or Maxwell cards .
Architecture: Fermi Kepler Maxwell Pascal
GPU Design: SM SMX SMM SMP
MaxVRAM: 1.5GB GDDR5 6GB GDDR5 12 GB GDDR5 16/32 GB HBM2
Max Bandwidth: 192 GB/s 336 GB/s 336 GB/s 1 TB/s
NVIDIA Volta GPUs, successors to Pascal Architecture.