Intel won the run and is the first to ship a chipset that is running at 100 MHz system or front side bus clock. People who want to have the fastest Pentium II CPUs will now have to take advantage of this chipset, because 333 MHz is as far as Intel will let you get in a board with 440LX chipset. 350, 400 and 450 MHz will require 100 MHz FSB and so 440BX. 100 MHz FSB alone are hardly giving any performance increase over 66 MHz, as you know from my article about the performance differences between 66 to 100 MHz FSB Slot 1 systems, but 350 or 400 MHz CPU clock will make sure that a BX system will indeed be faster than even an LX chipset with a 333 MHz CPU.
Ram Timing Negligible
You certainly remember the fact that the higher data bandwidth of main memory at 100 MHz FSB is only responsible for a mere 3% performance increase over 66 MHz FSB in most software, which is due to the L2 cache of a Pentium II CPU, that runs always at half the CPU clock, unchanged by the FSB. This may be surprising to a lot of you, because things are so different in Socket 7 systems. However, the reason for this difference is the direct relation between bus clock and L2 cache speed in Socket 7 systems. Now although it is surprising that a 50% bandwidth increase from 528 MB/s to 800 MB/s has only a little 3% impact on overall system performance, there is a good side to it as well. In the past I advised all people who wanted to get the most out of their system to tune their memory timings in the BIOS setup. This could improve performance particularly in Socket 7 systems, but even Slot 1 systems with 440LX chipset could notice a difference. This time has changed with 440BX and the 100 MHz FSB. The data bandwidth is now so high, that even if it was double it would have hardly any impact on overall system performance anymore. Thus little differences in memory timing have no influence on system performance at all anymore, if the data bandwidth is 830 MB/s or only 760 MB/s, the difference in performance is minimal. The consequence is that many motherboard manufacturers removed the memory timing options from the BIOS setup, but made sure that every SDRAM is running safely and reliably instead. Others cut down the timing adjustments to only one and the ones who still offer 4 different adjustments for memory timing must be asked if they want to impress or confuse the user. In this case you are still not doing anything wrong when choosing the slowest and therefore safest settings, the system will run the same as if you would use the fastest but most dangerous settings.
It Doesn't Have To Be PC100 SDRAM !
The memory issue is anyway one of the most important ones with 440BX. 100 MHz bus are asking 50% more performance of the SDRAM modules and there is a good chance that you can throw away the PC66 SDRAM you bought for your Socket 7 or 440LX board. However, I was very surprised that although some boards wouldn't run some PC100 DIMMs, almost all ran some high quality Samsung PC66 DIMM without the slightest problems. Thus you have a chance of being able to use your old PC66 SDRAM. As long as you bought very high quality. Quality is not the only thing that is required, one other important requirement is the serial presence detect SPD-EEPROM on the DIMM, which holds some important information about the setup and hold cycle times of the module and which is particularly responsible for the auto adjustment of the RAM timing in the board. Intel's PC100 and PC66 specification allows 9 ns tAC for PC66 modules, but only 6ns for PC100. It is possible that PC66 SDRAM already applies to 6ns time for getting the data ready, but this is of course pure luck, although certainly worth a try in case you own PC66 SDRAM. My Samsung PC66 SDRAM ran fine in all boards that would let you use PC66 modules even up to 112 MHz and it's now used in my very own system at 100 MHz FSB, the system where I am just now writing this article. Unfortunately are there quite a few BX boards that wouldn't neccesarily run each PC100 DIMM without problems, the Toshiba PC100 SDRAM I used for testing had problems in three boards, all of them were dual CPU boards. The boards from Intel and Siemens wouldn't let you run any SDRAM that doesn't say 'I am PC100' in the SDP-EEPROM. The Intel board tells you this in particular whilst booting before it simply halts, the Siemens board just keeps a black screen.
Jumperless Becomes Fashionable
Two things are obviously becoming pretty common now. 15 of the 20 boards in this test were equipped with a hardware monitor that would inform you about CPU temperature, power supply voltages and fan speeds and 8 boards can call themselves 'jumperless', offering the CPU multiplier and FSB adjustment in the BIOS rather than the fiddely jumper job. This is certainly a satisfying situation and I hope that jumpers will finally become history. There are board manfacturers who say that their customers prefer a hardware over a software adjustment, but these ones I may ask why they can't at least use tidy dip switches rather than messy jumpers.
All 20 boards ran with all AGP graphics cards I used, which are no less than a card with NVIDIA's RIVA128, 3DLabs Permedia2, Intel's i740, Matrox MGA-2064, Matrox MGA-G100, Matrox MGA-G200 and ATI's Rage Pro (Turbo). What I found out though was that some boards get very touchy with the used SDRAM when using a RIVA, i740 or MGA-G200 card and these boards got a fail with these RAM in case the instabilities happened.
The 'B21-Issue' - Automatic FSB Adjustments
One important thing which sets BX boards apart from each other is the way the FSB is adjusted. Intel's specification says that the board should check the state of CPU contact 'B21' and adjust the FSB to 66 or 100 MHz accordingly. 9 of the 20 boards apply to exactly this spec, 4 boards would let you either choose between 66, 68, 75 or 83 MHz in case you've got a 66 MHz FSB CPU or between 100, 102, 112, 133 for 100 MHz FSB CPUs. 7 board would leave the adjustments entirely up to you, so you can choose between 66 up to 133 MHz FSB, regardless which CPU is plugged in. In case you still should want to run at 100 MHz FSB or above with the boards that read out 'B21' you need to do the 'B21-procedure' I described in my article . The 7 boards that are not applying to Intel's spec are listed under my 'overclocking recommendations'.
AMI WinBIOS Is Big Disappointment
The compatibility test brought it to light that BIOS isn't BIOS. AMI's WinBIOS showed that not every BIOS manufacturer is able to see the sign of the times. Whilst every board with Award BIOS is able to distribute any available IRQ to any PCI and ISA slot, AMI's WinBIOS is still in the BIOS middle age and assigns a fixed IRQ to each PCI slot, making my compatibility test a real painful experience. Thus all boards with this wonderful AMI WinBIOS failed the compatibility test and only one could make it into the recommendation list because the manufacturer (Tyan) had done some serious work against this occurance and because this one board has already everything onboard, making the compatibility test almost obsolete. However, not all boards with Award BIOS were the same as well. Asus, AOpen and Chaintech are some typical examples for companies who put a lot of effort in their own BIOS development, resulting in the most bug free Award BIOS available. Particularly Asus claims to have 40% own code in the BIOS and this is noticable when you plug in different PCI or ISA cards. Elitegroup is also using Award BIOS, but for some crazy reason the BIOS wouldn't allow you to disable the IRQ for the USB. This resulted in a compatibility test failure, because there is simply no sense in USB sitting on an IRQ when USB isn't used at all. Siemens showed how cool a BIOS can really be, since the D1064 with Phoenix BIOS was the best product in the compatibility test. This will certainly be due to a lot of BIOS development work within Siemens. Intel's SE440BX is also using Phoenix BIOS, but it didn't look quite as good in the compatibility test at all, especially when enabling the onboard sound it was a huge pain getting other cards to work.
ISA PnP Network Card Caused A Lot Of Problems
There was an interesting occurance using an ISA PnP network card. In no less than 11 cases the card was reported as running fine in the Windows 95 system manager, but it was still impossible to connect to any network. This problem could be solved as soon as the used IRQ was manually changed to IRQ5. Any IRQ higher than 7 would show this described inability to connect to a network and that without the slightest error message. Microstar's MS-6119 is the only board with this failure that made it into the recommendations list and I hope that MSI will sort out this problem asap. Chaintech showed that it can be done. After I reported this error to Chaintech it took 2 days and they supplied me with a BIOS that didn't have this problem anymore.
The performance of the boards was within a 3% range, which I consider as negligible. Thus even the 'slowest' board certainly offers satisfying performance.