AFAIK, the architecture of the replay system doesnt take in account the real-time physics, but only its graphical representation. You can add fidelity, and the graphical representation of the in-game physics will look more close to the real-time physics, but stll the replay is a different animal. Noel pointed that the size of iRacing replays are 5x the size of rF2 replays, that happens because in iRacing the replays store the physics data of the player. Now, imagine how big might be a server replay, if the architecture was based on the storage of physics data ... 20 clients in a server= 100x (5x per 20 clients) the size of a replay in rF2? A rF2 server replay, weighting say 100 MB, might be about 10 GB? And these calculations are based on iRacing ... have you seen how intensive is the CPU load of tgm? Ie, the huge amount of data required for the physics math of rF2 tires? Here we cant give evidence, but i would not be surprised if rF "physical data" were 10x or more, compared to 5x of iRacing. A server replay that now is 100 MB might weight 20 GB?