Now, to show you how important the bus speed really is, I took the opportunity given to me by the recent acquisition of an Asus P/I-P55T2P4 board to do just that. Fortunately, you can run this board at six different bus speeds: 50 MHz, 55 MHz, 60 MHz, 66 MHz, 75 MHz and 83 MHz.

Since there are 4 different multiplier settings of a P54C (1.5, 2, 2.5, 3), I had 24 different CPU Speed/Bus Speed combinations to test. Unfortunately, my P166 would not run any faster than 208 MHz. At 208 MHz, I could not even run Winstone 96 without crashing (though I tried about 25 times with 25 different settings). Hence, the combinations were reduced to 22 under DOS and 21 under Windows 95.

I decided to use two well known benchmarks.

• For DOS, which nowadays only seems to be used by gamers, I chose the Quake 1.06 Timedemo Benchmark . I actually fell in love with this benchmark because it revealed a very nice surprise to me, which will be extremely interesting to you as well. It's also very easy to execute and does not take long to run.
• For Windows 95 I chose the well-respected Winstone 96, since I just could not get my hands on the new Winstone 32. The practical thing about Winstone 96 is that it's also used for the P-rating system. Hence I was able to make my own P-rating for the different configurations. The big downside of this benchmark is the time it takes to complete. Together with the defragmentation of the HDD before each test run, it took between 20 and 40 minutes to run, which adds up to about 12 hours of non-stop testing.

Let's start with the Quake benchmark.

Quake 1.06 Timedemo Benchmark Results, Grouped By Multiplier

Grouped by Multiplier

Multiplier x1.5,25,Frame Rate (fps)75,8.7,blue82.5,9.6,blue90,10.4,blue100,11.5,blue113,13,blue125,14.4,blue

I am presenting you this kind of grouping to show you that - amazingly - within the same multiplier setting, the Quake frame rate is directly proportional to the CPU speed and hence a linear function. The gradient is different from multiplier to multiplier and of course depends on your system. This opens up a completely new way of interpreting the results of my survey - but more about this soon on my Quake Survey Page. It gives you the chance to actually calculate the expected frame rate for each CPU/bus speed, once you have one result with the corresponding multiplier.