Speeding Up The BX
First I ran all benchmarks at 100 MHz FSB, clocking Coppermine at 600, 650, 700, 750 and 800 MHz. As higher multipliers than x8.0 have never been specified (Intel is using x5 and x5.5 for x9 and x9.5 with their new Celerons), it wasn't possible to clock the CPU higher when only running 100 MHz FSB. I continued with runs at 133 MHz FSB, which - although being out of the chipset's specs - did not cause the slightest problem. The 133 MHz FSB tests were run with CPU clocks of 533, 600, 667, 733, 800, 866, 933 and 1000 MHz.
The most interesting benchmarks were run at 600, 675, 750, 825, 900 and 975 MHz, using a FSB speed of 150 MHz. I would have liked to include 1050 MHz as well, but our test sample did not want to run stable at this speed. Running a BX system at 150 MHz FSB pushes the AGP card and chipset far beyond their specifications as both are overclocked by 50%, resulting in 150 instead of 100 MHz for the BX-chipset and 100 instead of 66 MHz regarding AGP speed. I had to use a fan for the graphics card, since the Asus V6600 card became rather hot at 100 MHz AGP clock.
AGP At 100 MHz
When running a Pentium III at 100 MHz FSB, the BE6-II gives you two AGP clock choices: 2/3 FSB or 1/1 FSB. The normal setting to get 66 MHz is 2/3. Selecting 1/1 will clock the AGP at 100 MHz as well, resulting in a maximum bandwidth of 800 MByte/s instead of usual 533 MByte/s. I did not finish those test runs, because the results are pretty much the same as at 66 MHz AGP clock. WE received several mails where people held the overclocked AGP responsible for the superior performance of BX133 over i820/i840. This reasoning is illogical in the first place, since i820/i840 support AGP4x, resulting in an AGP-bandwidth that's still higher than AGP2x overclocked to 89 or even 100 MHz. My benchmarks prove that the higher AGP-clock is hardly able to improve any scores.
3D games today are not taking much advantage of AGP-bandwidths of more than AGP2x. The improved scores in 3D games of BX133 or BX150 systems are due to the improved performance of main memory, where the reduced latency of SDRAm clocked at 133 or even 150 MHz seems to be more important than the increased bandwidth. This is the reason why RDRAM-systems don't show the same high frame rates. RDRAM does improve bandwidth, but at the cost of latency.
SYSmark 2000 - Windows 98 SE
The test results using BAPCo's SYSmark 2000 show the following facts:
A Pentium III 600/150 is about 8% faster than a Pentium III 600/100 and 3% faster than a 600/133.
The Giga-Coppermine is clearly the fastest processor for Windows applications, followed by two clock speeds that make use of 150 MHz FSB.
Using 150 instead of 133 MHz FSB increases system performance to the level of the next faster 133 MHz CPU. Coppermine CPUs clocked at 100 MHz FSB are clearly the losers.