Apple G5 Dual 2.5GHz vs. Dual 2.0GHz (first revision)
Or ZNeK's machine vs. mine. Running them Skidmarks, the Dual 2.5 Ghz is 25% faster, big surpise
| Test | Dual 2GHz | Dual 2.5 GHz |
|---|---|---|
| ParseVid (INT) | 104.5% | 160.3% |
| MPEG (INT) | 148.1% | 183.6% |
| PixBlend (INT) | 148.9% | 186.1% |
| Ellipticrypt (INT) | 150.6% | 188.1% |
| Rijndael (INT) | 205.8% | 257.1% |
| Quicksort (INT) | 212.1% | 265.3% |
| MDCT (FP) | 361.0% | 451.3% |
| IntToFloat (FP) | 238.4% | 297.5% |
| Q3 (FP) | 180.0% | 225.1% |
| FFT (FP) | 383.4% | 481.7% |
| VolInt (FP) | 283.1% | 355.4% |
| Quant (VMX) | 213.3% | 268.0% |
| Galaxy (VMX) | 166.2% | 207.2% |
| IDCT (VMX) | 204.0% | 254.8% |
| BigMult (VMX) | 229.6% | 287.0% |
An interesting and mysterious fact is that measurements are very stable for FP and VMX tests, but vary a lot for INT tests. This is possibly the same weird problem I had measuring some other code (see Head Scratching). I have still no clue, what the reason could be, but I am sure that knowing the reason would be enlightening.