Just some quick test to let you guys know that synthetic benchmarks can be manipulated:
引用:
AMD Opteron(tm) Processor 144
CPU speed: 2646.28 MHz
CPU features: RDTSC, CMOV, Prefetch, 3DNow!, MMX, SSE, SSE2
L1 cache size: 64 KB
L2 cache size: 1024 KB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
L1 TLBS: 32
L2 TLBS: 512
Prime95 32-bit version 24.14, RdtscTiming=1
Best time for 512K FFT length: 17.604 ms.
Best time for 640K FFT length: 23.012 ms.
Best time for 768K FFT length: 27.951 ms.
Best time for 896K FFT length: 33.391 ms.
Best time for 1024K FFT length: 37.258 ms.
Best time for 1280K FFT length: 47.333 ms.
Best time for 1536K FFT length: 58.072 ms.
Best time for 1792K FFT length: 70.158 ms.
Best time for 2048K FFT length: 78.418 ms.
Best time for 2560K FFT length: 102.758 ms.
Best time for 3072K FFT length: 125.455 ms.
Above is a run of prime95 on my Opteron 144 Venus core.
It has a 1MB L2 cache.
By analysing the best time, you can see from 512K to 640K, the increased time is only 6ms. From 640K to 768K only 5ms, etc.
But when you compare 1024K and 1280K, you\'ll see the time increased for 10ms. From 1280K to 1536K, a 11ms margin, etc.
So, guess what?
