Yeah you could, simplified, add each entity's looping function pointer to a list, and run them all through the main loop.

I think it's very odd that 400 cube model entities act way more slow than 1 entity containing 400 cube models. One could test it dynamically by creating some cubes through DirectX and register it in one entity. Would be interesting. If performance is indeed way better, then one could write some easy functions to merge entity contents into one entity, of course including a vertex manager .