Louka, That was my first reaction too, but he does go into a lot of depth, including preconditioning and use of CG to solve nonlinear problems. And he doesn't leave you with a long list of undefined terms to look up!

Most intriguing for me was that CG is used to tackle really large relaxation problems. How large, I wonder? (If you tell me to think big, I'm just likely to do it, so what's a realistic target?) And how much sparsity does it take for this method to be a good choice? The nonlinear use seemed less appealing because you have to keep restarting it. But maybe if one doesn't have any other good choices, those restarts are a small price to pay?