In the Beginning

For years our processors had been crippled by a major factor which had remained overlooked, not the clock speed at which the processors ran at, but the speed at which the rest of the system could transmit data, commonly known as the Bus Speed. 

Now used as a term loosely associated with the speed your memory operates at, a system's bus speed dramatically affects virtually every single component in your computer, from your Video Card to, in the case of Socket-7 systems, your Level 2 Cache Memory. 

Since we're dealing with Socket-7 motherboards here we'll leave out the specifics from other architectures (specifically the Slot-1 design of Intel's Pentium II), but in general, your L2 Cache operates at your system's bus speed.  When Intel announced their 586 line of processors, which due to a flaw in the design and a denied request for exclusive use of the name 586 was eventually renamed to the Pentium Processor, they also announced a new interface for the processors.  Migrating from the Socket-3 design of the time which dominated the 486 market, Intel chose to use yet another flimsy Socket interface, dubbed Socket-5.  Motherboards using this Socket would also have to boast support for higher Bus Speeds, out of all of the supported Bus Speeds which were originally available upon the release of the first Pentium Processor, the 66MHz bus speed seemed to quickly make itself the standard...a standard which the industry has tried to shake time and time again over the past 18 months. 

With the advent of the Intel 430VX Chipset manufacturers began pushing the limits of the 66MHz Bus Speed to new extremes, some of the first VX based motherboards not only took advantage of the 66MHz bus speed which was officially supported by Intel.  These motherboards also enabled user's to Overclock their systems (or increase the speed of their computers past the specified limit) by taking advantage of the newly introduced 75MHz bus speed, a bus frequency which was previously only reserved to non-Intel chipsets and the motherboards that housed them.  However for the first time, an Intel chipset supported a bus speed, albeit unofficially, that allowed the industry to really have some fun and stay competitive with the rest of the market.  VLSI and VIA already had chipsets out or preliminary designs for motherboards which would support this 75MHz bus speed mainly for use with the Cyrix 6x86-PR/200+ (75MHz x 2.0 - officially) so in order to remain competitive it was almost necessary for an Intel chipset to be used on a motherboard that would support this higher frequency. 

Soon after these VX boards began popping up with support for the 75MHz bus speed, HX boards seemed to be doing the same...eventually an 83.3MHz bus speed setting was discovered on a few boards here and there, leading to the overclocking revolution you see unfold before your eyes every time you pay just about any hardware site a visit.  Throughout all of this however, Intel has still refrained from upping the 66MHz bus speed specification to even 75MHz, they were determined that the 66MHz bus speed wasn't enough of a limiting factor to change it.   It was up to the users to seek out the motherboards that supported higher bus speeds to make their newly built systems operate like they "should," and it was up to the manufacturers to produce motherboards that either had hidden undocumented settings or unofficial settings for the 68, 75 and 83MHz bus speeds (the 68MHz bus speed is referred to as the Turbo Frequency of the 66MHz bus speed, used Internally by manufacturers for testing purposes...however it makes a sweet little overclocking tool if you don't want to risk much).  The irony of this situation was that most motherboards that supported the 75/83MHz bus speeds and were reliable at those bus speeds were based on Intel Chipsets, not solutions from other companies.

Broadening their Horizons

In May of 1997, when Intel finally released the true successor to the Pentium Processor (the Pentium MMX was simply an enhanced Pentium in my mind), the Pentium II, it became quite obvious what sort of a limiting factor our friendly 66MHz bus speed turned out to be.  The actual clock speed of a CPU is derived not by a simple game of point and choose, but by an even more simple equation: CPU Clock Speed = Bus Speed x Clock Multiplier.  The Pentium II, starting at 233MHz at its release and now up to speeds in excess of 300MHz required that a clock multiplier of at least 3.5x be used in cooperation with the 66MHz bus to obtain the high clock speeds the processors ran at.  The 300MHz Pentium II operating at a 66MHz bus speed requires a 4.5x clock multiplier, and with Intel planning a 400MHz Pentium II later in 1998 and eventually Pentium II processors in excess of 500MHz maintaining this silly bus speed would eventually pose quite a few engineering problems to motherboard manufacturers.  The solution?  Increase the bus speed.  But to what speed?   75 and 83.3MHz are nice for overclocking, but if you're going to make a major change such as this one you go for the gold, the mighty 1-0-0. 

A Discriminating Speed Limit: 66/100

To the surprise of all of their critics, Intel made a huge jump in their specification for the Bus Speed of the next generation of Pentium II processors, from 66MHz to a whopping 100MHz.  However with all the hype surrounding this decision, they failed to explain what would happen to the hundreds of thousands of Socket-7 motherboard/processor owners; would they be left in the dark?  Forever doomed by the 66MHz bus speed?  Unfortunately, yes, from Intel's standpoint that is.  Their position on this is very understandable, they have this hot new processor that they want the masses to buy, so what better way to promote it than supporting it with a bus speed never before achieved by a manufacturer in the PC industry?   This way, Intel kills the Socket-7 market, and gains complete control of the Slot-1 Industry since they are currently the sole proprietors of the Slot-1 architecture (although that could change very soon).

The Catch?  Three Letters

No plan is flawless, although Intel's marketing strategy seemed to be, there still remained a tiny glitch, Advanced Micro Devices, AMD for short.  If Intel could pull off a 100MHz bus speed who is to say that AMD can't?  No one, at least AMD didn't think so.  Shortly after the announcement of their hit K6 Processor, AMD released the first specs of their next generation K6 chips, including the AMD K6-3D.  What made the K6-3D so special was it included support for both the 66MHz bus speed, and the 100MHz bus speed while maintaining compatibility with the Socket-7 architecture.  As Cyrix did with their 6x86-PR/200+ AMD managed to breathe more life into the Socket-7 market by announcing a processor that would be at home with a 100MHz bus speed.  And as expected, the first few chipset manufacturers to announce compatibility with this bus speed were non-Intel chipset manufacturers, specifically VIA and ALi.  Although everyone expected the industry giant VIA to produce a chipset with support for the 100MHz bus speed, no one would've guessed (well, most people didn't know at the time) that Acer Labs Incorporated (ALi) was working very closely with AMD to develop a chipset that would accent the release of the K6-3D with support for the 100MHz bus.  When it was finally made public that ALi was working on a chipset design (Aladdin V) with 100MHz bus support, the race was on to get to 100MHz.  VIA announced their MVP3, while SiS quietly worked on their 5591, soon enough we'll see quite a few chipsets which would support this speed...however even while benchmarks were floating around about how well the K6 performed with the ALi Aladdin V Chipset no one bothered to ask if Intel had changed their minds about supporting the 100MHz bus speed with their TX Chipset, the most popular Socket-7 chipset out today...

Breaking the Speed Limit

What has been keeping us from the 100MHz bus speed?  Two things, our memory and our PCI peripherals.  The Random Access Memory (RAM) of your system operates AT your system's Bus Speed, and while most SDRAM and even EDO DRAM modules have no problems running at speeds up to 83.3MHz, when reaching 100MHz things get a bit tricky.   With the quality of the SDRAM floating around the market increasing dramatically as the release of the first 100MHz Bus Speed Socket-7 motherboards approaches the problem of getting RAM to work at that high of a bus speed is slowly deflating itself to more of a smaller issue than a major worry. 

The other half of the story is the PCI Bus, and the components that rest on it.  Unlike the memory bus, the PCI bus operates at 1/2 the Bus Speed, meaning for a 66MHz bus speed your PCI peripherals will operate at 33MHz which is perfectly fine.  When running at a 75MHz bus speed however, things get a bit more tricky, since half of 75MHz is 37.5MHz, a 13% increase in the PCI bus speed.  A great deal of PCI cards work flawlessly at this speed, however a number still won't function properly, taking the bus speed up to 83MHz also increases the PCI bus speed to 41.6MHz which even fewer PCI cards will function at.  As you might be able to guess, at 100MHz, the PCI bus operates at 50MHz, a full 51% increase over the specification for most PCI cards.  While the PCI 2.1 specification theoretically supports speeds up to 66MHz, very few cards actually stand by that theoretical limit, in fact, most manufacturers just assume 33MHz is the limit. 

So how do we get past this barrier?  The answer is quite simple, instead of dividing the Bus Speed by 2 in order to obtain the PCI bus speed, use a 1/3 divisor (100MHz x 1/3 = 33.3MHz) which results in a PCI bus of, yep you guessed it, 33MHz.  Problem solved...not really though, it is still the responsibility of the manufacturer to provide us with this setting, however you can rest assured that most motherboards that officially support the 100MHz bus speed will come with this feature.

What about Intel?

Earlier we were posed with the question, "Has Intel had changed their minds about supporting the 100MHz bus speed with their TX Chipset..?"  The answer...no.  However if we remember back to the days of the VX chipset, Intel didn't support the 75MHz bus speed yet it did appear on motherboards based on their chipsets.  The same situation exists here, while searching for a hidden 83.3MHz bus speed setting on an Intel TX based motherboard I stumbled across a 100MHz bus speed setting, here's where things get interesting.

The Game Continues

So which motherboard would be the first Socket-7 motherboard available for sale with unofficial support for the 100MHz bus speed?  The Tyan Titan Turbo ATX-2. Tyan has always been a manufacturer who has relied on their sheer quality and reliability to get sales, none of their motherboards (until now) even supported the 75MHz bus speed, much less the 83MHz bus speed.  In any case, while searching for the 83.3MHz setting on the Titan Turbo ATX-2 I managed to stumble across this little setting:

JP23 1-2 = OFF
JP23 3-4 = ON
JP31 = ON

Upon setting the clock multiplier to 2.5x and using the above bus speed setting I noticed that the BIOS detected the processor I had installed (AMD K6) as being a 250MHz processor, a few benchmarks and tests later I confirmed that the system did indeed boot at 100MHz x 2.5 for a clock speed of 250MHz.   I should mention this now, that the Tyan Titan Turbo ATX-2 is very shaky at the 100MHz bus speed setting, and by no means do I expect you to be able to run a system for even limited amounts of time at this setting.  I tried a total of 5 different SDRAM combinations: 2 x 32MB Corsair SDRAM, 2 x 32MB SmarTech SDRAM; 2 x 32MB Advanced Megatrends SDRAM; 1 x 64MB Corsair ECC SDRAM; 1 x 64MB Advanced Megatrends SDRAM.   Out of all of those test modules, the SmarTech, Advanced Megatrends, and Corsair (64MB Chip only) would actually allow the system to boot, and only the Advanced Megatrends SDRAM (64MB Chip only) would allow me to get some results out of the test system.

In order to increase stability the BIOS Memory timings had to be configured at the slowest possible settings otherwise Windows would come back with a handful of errors.  The Hard Disk used in the test System (WD Caviar) was set to a PIO Mode 2 instead of the standard PIO Mode 4 setting to avoid disk corruption.   The Matrox Millennium being used with the test system had no problem running using a 50MHz PCI bus frequency however, and it provided a nice little performance boost as well.  The AMD K6 233 used in the tests was one of the first K6/233's to make it out into the retail market, and is therefore a very horrid overclocker, at 250MHz the system was too erratic to benchmark, and at 300MHz the system wouldn't boot.

What about the Pentium MMX and Cyrix 6x86MX?

Why did I chose to use the AMD K6 in the tests instead of the Pentium MMX or Cyrix 6x86MX?  Well, first of all, the Cyrix 6x86MX wouldn't even boot at the 100MHz bus speed setting.  At 100 x 2.0 the system returned a dead screen (no video) and obviously at all higher speeds the system couldn't manage to do more than make a few noises.  The Intel Pentium MMX went a bit further than that, the processor would boot at 100 x 2.0 as well as 100 x 2.5 (100 x 3.0 wouldn't boot at all, which is funny since 83.3 x 3.5 works perfectly fine) however Windows 95 failed to load with a Pentium MMX installed using the 100MHz bus speed.

The conclusion?  As they stand today, the Pentium MMX and the 6x86MX are not designed for use with the 100MHz bus, and you shouldn't expect to be able to use the processors at bus speeds that great unless either of the manufacturers plan some significant changes to their chips, for now, the only x86 processors which will run reliably using the 100MHz bus speed are the AMD K6 and the Pentium II.  Now that we've settled that, let's move onto the test...

What you've all Been Waiting for: The Test

Normally when testing a motherboard I run the test system through a few passes of Winstone 97 and Winstone 98, however while using the 100MHz bus speed on the Tyan Titan Turbo ATX-2 most tests failed, Windows 95's registry would either become corrupt during the testing process or the system would hang.  Therefore I had to become a bit more creative with the tests, meaning I ran the portions of Winstone which would run (none of Winstone 98 would complete) and compared those scores.  The test system ran without the Intel Bus Master IDE Drivers installed, the system seemed to be a bit more stable with the Microsoft DMA Drivers instead of Intel's BMIDE Drivers. 

Here are the exact specifications on how I tested:

How I Tested

• Each benchmark was run a minimum of 2 times and a maximum of 5 times, if the motherboard failed to complete a single test within the 5 allocated test runs the OS/Software was re-installed on a freshly formatted Hard Drive and the BIOS settings were adjusted to prevent the test from failing again.  All such encounters were noted at the exact time of their occurrence.

• Business Winstone 97 / Business Graphics Winmark 97 was run at each individually tested clock speed, if reliable scores were achieved with the first two test runs of the suite an average of the two was taken and recorded as the final score at that clock speed.  If the test system displayed erratic behavior while the tests were running or the results were incredibly low/high the tests were re-run up to 5 times and an average of all the test runs was taken and recorded at the final score at that clock speed

• After each motherboard was tested a complete format of the test hard drive was initiated and the OS/benchmarking software was re-installed afterwards a defragment was initiated using Windows 95's Disk Defragmentation Utility

• No foreign drivers were present in the test system other than those required for the system to function to the best of its ability

• All foreign installation files were moved to a separate partition during the test as to prevent them from effecting the test results

• All tests were conducted at 800 x 600 x 256 colors

Test Configuration

The Results

Here they are:

Test System 1 ran using the AMD K6/233 at 100MHz x 2.0

Test System 2 ran using the AMD K6/233 at 66MHz x 3.0

 

 

 

Interpreting the Results

In most situations, the 100MHz bus speed provided an increase in performance of about 10 - 12% at the same clock speed over the 66MHz bus speed.  This can be expected and is a considerable improvement when compared to the virtually non-existent increase the 100MHz bus provides when used with Pentium II systems.  The reason for this is quite obvious.  As mentioned earlier, Socket-7 systems keep their Level 2 Cache (Cache is basically ultra-fast memory where frequently used data can be stored for easy access) running at the bus speed your motherboard is set to.  In most cases, this speed is 66MHz.  By pushing the limits of your system using the 100MHz Bus Speed as tested here, your L2 cache can be accessed at a clock speed 50% greater than it could previously (100MHz vs 66MHz).

With the Pentium II's Slot-1 Architecture, things are a little different when it comes to how the L2 Cache speed is derived.  You may have heard of the term Dual Independent Bus (DIB for Short) in essence, this DIB architecture the Pentium II uses allows it to have its L2 Cache operate off the CPU, as to reduce costs, while maintaining a direct high speed connection to the L2 cache by keeping it on the CPU Cartridge, as to increase performance.  Therefore, the speed the L2 Cache operates at on a Pentium II system is equal to, not the Bus Speed, but one-half of the processor's clock speed.  It is for this reason that the Pentium II isn't affected too dramatically by higher bus speeds, because at 100 x 3.0 and 66 x 4.5, the L2 cache of the Pentium II still operates at 150MHz.  The only devices receiving the benefit of the 100MHz bus in that case are your main memory and any AGP/PCI devices that you may have. 

The best overall solution to extend the life of Socket-7 systems would be to place the L2 cache on the CPU or on a secondary Bus running at half the speed of the CPU (much like the Pentium II) while maintaining backwards compatibility with the Socket-7 architecture.  Although changes like this have been proposed for future processors from AMD and Cyrix, we have yet to see any concrete evidence that this has been done.  AMD claims that their K6-3D+ will have 256KB of on chip L2 cache running at clock speed, we will just have to wait for its release later this year to confirm that however.

With the results we have here, the applications that received the greatest boost in performance from the 100MHz bus speed were applications that made extensive use of a combination of raw processing power, heavy I/O and Disk Access, and graphics manipulation.  The Database 97 tests, Truespace Rendering Tests, and Quake 2 Gaming Tests were the ones the benefited most from the higher bus speed.  The Word Processing/Spreadsheet 97 were affected as well, just to a lesser degree, while the Disk Copy Time barely fluctuated at all averaging a 1.8% increase in performance, barely noticeable.  When the 100MHz bus does officially hit us, we can expect similar results in our real world usage of systems using the 100MHz bus speed.   The tricky part then will be buying SDRAM that will work at the higher bus speed...but is that the only problem?

About that L2 Cache...

Here's an interesting point often overlooked when considering the stability and reliability of a motherboard at the 100MHz bus speed, L2 cache.  Since the L2 cache is a component of the motherboard that will be running at this high frequency, it will take some high quality cache modules to keep up with the speed.  It is doubtful that we will have to worry much about the quality of the L2 cache used on motherboards that support this setting, since during the design phase of a motherboard the prototype is placed through a rigorous testing process in which the L2 cache remains enabled.  If the motherboard does officially support the 100MHz bus speed you can expect it to work properly at the bus speed, provided that you take the initiative to make sure that the components you use with the motherboard also function properly at the 100MHz bus speed.  Its not really good practice for a motherboard manufacturer to sell a motherboard that doesn't work at a supported setting, so I wouldn't worry too much about that happening.  The question here is, can your peripherals stand the heat at 100MHz?

Conclusion

In the next few months we'll see many motherboards appear with support, either officially or unofficially for the 100MHz Bus Speed.  Although you won't have to wait for the AMD K6-3D to take advantage of this bus speed, getting it to work with today's Socket-7 processors may be a bit more tricky than you would imagine.  While the 100MHz bus has been highly anticipated for quite some time, now that it is nearly here it is time to look beyond that...can you say 133MHz bus?  Who will be the first to once again break Intel's speed limit barrier, as fast as it may be, with another unofficial setting much like the 75MHz bus speed was in comparison to the 66MHz frequency?  The race is on once again...