The appropriate test(s) depend on what you do with it. If you play Quake all the time, a certain type of test woulld be appropriate. If you work on CAD, etc another type of test would be appropriate.
Use benchmarks that doesn't involve graphics, sound e.tc.
Games are bad benchmarks, and so is many synthetic benchmarks too. No benchmark is perfect.
Sisoft sandra is rather good on showing peak performance. Sysmark shows the average performance better.
Trying and running real applications is usually the best way to compare performance but doesn't give you any comparative numbers.