Detailed Results and Comments

Comment

• The Workstations are pretty much faster than PC's and Macs, as long as the age of the processors is comparable. Note a lot of the tested workstation are quite old, e.g. the machine with R4400, PA 7xxx CPUs and Alphas with clock speeds of 300 MHz and less. The R4400 and the PA 7100 were built at the same time as the Pentiums with at most 100 MHz.

  Modern workstations, such as the UltraSparcs and maybe the 500 MHz Alpha and the Origin 2000 are a lot faster, especially in FPU performance. But this is what you need with number-crunching: FPU-performance.
  

• Note that due to the preemptive multitasking of UNIX, the results may vary by a few percent from run to run. Therefore UNIX machines differing by less than 5% should be regarded as equally fast.

  With the Macs its different. I haven't found any way to directly measure the amount of CPU time used. However, the benchmarks take total control over the Mac due to the Mac's cooperative multitasking. and do not yield CPU time to other programs. The elapsed time is thus a good approximation in this case.
  Due to the (almost) total control of the machine, the results of the benchmark shouldn't vary by more than 1 %.
  

• To my surprise, there is no clear picture regarding speed loss if the amount of data is larger than the cache. However some basic things can be seen: if the amount of data is larger than the L2 cache, the speed of the computer drops. The severity of this speed hit depends on the size and speed of the cache as well as RAM speed.

  Surprisingly, larger caches don't help the Macs as much as they do other machines (see results for the PowerMac 8500 and G3).
  However, it must be noted that fluctuations in the results cause even larger fluctuations in the speed-loss value (the error is magnified if one divides one erroneous value by another).
  

• Compared to the PCs, the Macs are pretty much faster at the same clock speed as long as the CPUs are equivalent (i.e. PPC 603 vs. Pentium and PPC 604 and PPC 750 vs. PentiumPro and Pentium II). Even the PowerMac G3/266 is faster than a Pentium II with 300 MHz.

Mac-specific comments

• The speed gain from Libmoto is small with this benchmark (for both MacOS and Linux). The reason is that almost all FPU operations in the benchmark are simple FPU operations, such as multiplication and addition. Libmoto accelerates the execution of complex FPU operations such as Sin or exp().
  
• I found it interesting to see that the gnu compiler on Linux is faster than Code Warrior 9 on MacOS. (I checked Code Warrior Pro on a G3 once. It was even slower.)
  
• Tests by Sven Negrassus of MacInfo showed that this benchmark seem to be a kind of overclocking checker. It seems to be very sensitive to excessive overclocking. Hence it seems to even be a tool for checking the stability of your computer.

Wish list

• runs on Windows (of course someone needs to port it first).
  
• results for the new DEC Alpha AXP 21264
  
• results for the new Pentium II/Xeon
  
• results for other x86 CPUs, especially non-Intel ones (e.g. AMD K6-II)

The detailed results

Explaining the columns: