Why would they care for a few hundred MBs when the games are in the 10s of GBs?
CPU cache space for code is much smaller than GPU memory for models (and the former is more important for performance since many CPU operations like pipeline parallelism are latency bound, not compute bound).
alt Hacker News
CPU cache space for code is much smaller than GPU memory for models (and the former is more important for performance since many CPU operations like pipeline parallelism are latency bound, not compute bound).