NVIDIA has been using color space compression in VRAM for over 15 years, even the FX-series had it at a 4:1 ratio. Anandtech discusses it with every new GPU generation, but here's their
Pascal overview &
Turing overview for memory compression. NVIDIA has been improving it aggressively since the 700s/900s as a way to sidestep bandwidth limitations. My point is, Ampere is not the first GPU with strong hardware memory compression. A100 has hardware tensor compression, but that doesn't seem to be included in GA102.