I disagree Leadworks. In a test with 10,000 particles where about 80% were off-screen it was significantly beneficial to only draw the particles that were on-screen and to not even visit the ones that were off-screen (except to move them). It probably depends on what you're using the particles for, how many you have, how many go offscreen and under what conditions. As usual there's probably no single approach that works best for all situations, because as usual a specific hard-coded single-solution approach will work best in a specific situation.