Here's another one for you, just to prove the antiquity of the argument. In 1981, IBM had a range of 3033 models - the 3033U, 3033N and 3033S.
The 3033U was rated at 4.7 MIPS and the 3033S was rated at 2.2 MIPS. The _ONLY_ hardware difference was the cache size - only 512KB on the 3033S. If you wrote a synthetic kernel benchmark (the Finanzamt Charlottenburg in Berlin did this) that could be contained entirely within 512 bytes, the performance of the 3033U and 3033S was identical. Biut when you tried a real workload, the 3033S was less than half as fast. -- Phil Payne http://www.isham-research.com +44 7785 302 803 +49 173 6242039
