ScienceMark 2.0, Pifast
Starting off the benchmarking run by looking at memory analysis. There should be no differences between the Athlon 64 Clawhammer models here. The speed of L2 cache isn't measured.
That proves to be the case. The single-channel Athlon 64 CPUs are remarkably efficient when it comes to bandwidth, according to ScienceMark 2.0's measurement. 3GB/s out of a potential 3.2GB/s isn't bad. The Pentium 4 CPUs show an expected similarity between results, but only manage to extract 4.3GB/s out of a potential 6.4GB/s. They, of course, are being run on an Intel dual-channel chipset.
Latency shows a similar picture. The Model 3000+ is no doubt helped by the excellent on-die memory controller, so it more or less matches the Model 3200+ and 3400+'s results. The Pentium 4s lag behind. That's what happens when the controller hub is located on the Northbridge.
Our memory investigation hasn't revealed much. It shouldn't, unless we're testing for L2 cache size. Pifast, our first practical test, calculates the constant to 10-million places. Memory bandwidth and clock are important naturally, but how important is cache size to this activity?. The answer, with respect to the Athlon 64 architecture, is not very much. The Model 3000+ trails the 3200+ by less than a second, but it trails the 3400+ by a considerably larger margin. Remember that clock speed for both the 3000+ and 3200+ models is the same; 2GHz. At this very early juncture, it seems as if AMD has been generous with the 3000+ rating. Notice the almost linear difference between the Pentium 4s. That's as it should be. Models are segregated on clock speed alone.
