Original Link: https://www.anandtech.com/show/574

BX-133 Video Guide

by Matthew Witheiler on July 5, 2000 5:01 AM EST


Introduction

Since its introduction nearly three years ago, Intel's 440BX chipset has made a significant impact on the motherboard market. In an era of constantly evolving technology, the 440BX has become the mainstay of motherboard producers and consumers alike. Even the i820 chipset, thought to be a replacement for the aging 440BX, ended up being devastated by the little chipset that could: the BX. Plagued with an RDRAM-only memory system after the SDRAM enabling MTH failed, i820 motherboards were quickly forgotten by both consumers and manufacturers, only to be replaced by long lasting BX motherboards.

While the longevity of the 440BX cannot be questioned, the sub par 100 MHz front side bus speed of the BX chipset proved to be a performance limiting factor to professionals and overclockers alike. On the professional side of things, the 100 MHz FSB provides a bottleneck for essentially all system operations. For overclockers, the 100 MHz FSB limit prevented any overclocking of the clock locked Pentium chips.

Luckily, even early BX based motherboards provided a solution to this problem. By setting a few jumpers, or in some cases just pushing a few keys, 440BX based motherboards were able to surpass the Intel suggested speed limit and push the boundaries of FSB speeds. Many BX based motherboards allow for not only the coveted 133 MHz FSB speed, but even higher FSB speeds. The longevity of the 440BX only served to enhance this overclocking option, as good board designs became more and more refined.

It has been shown time and time again that a BX based motherboard running at a front side bus speed of 133 MHz produces the fastest possible motherboard solution. Even the recently announced i815 chipset from Intel proves to be no match for a BX at 133 MHz, even though it officially supports the 133 MHz FSB rating.

The problem with a 440BX running at 133 MHz is obviously not speed: there is plenty of that. Unfortunately, the problems arise from the interaction of components to the motherboard at this non-speced speed. Items such as PCI bus dividers, memory, and AGP dividers are all potential road blocks, preventing 133 MHz operation. Recent months have seen the PCI bus divider and memory problems vanish, leaving only the 2/3 AGP divider standing between you and system bliss. Fortunately, AnandTech is here to help you. To begin, let's investigate why some of the previous problems have disappeared.



PCI Bus Dividers- A problem no longer

There are three main external systems that interact with and are dependent upon the front side bus. These are the PCI clock, the AGP clock, the memory, and the processor. From the beginning, it was well known that the processor would not be a limitation to higher FSB speeds.

So, while the processor was left out of the equation for FSB overclocking success, 3 items still remained. Of these problems, the first one to be solved was the PC clock divider.

Ever since the introduction of the PCI bus standard, it has been known that cards are produced around a 33 MHz specification. Due to this fact, the FSB speed always has to be divided down in order to interface with the PCI bus. With the 66 MHz FSB speed that was predominate in the original Pentium era, the PCI bus divider was set to 1/2, reducing the FSB speed from 66 MHz to the 33 MHz needed for the PCI bus.

The 440BX brought along the 100 MHz FSB speed. The increase in FSB speed resulted in a divider increase for the PCI bus. No longer could the FSB be divided in half to provide a PCI bus frequency in spec. As a result, a 1/3 PCI bus divider was introduced. By dividing the FSB speed by 3, the PCI bus on a 100 MHz FSB system would be running at the desired 33 MHz..

It was thought that since the PCI divider was "set" to 1/3 on 440BX motherboards, running at a 133 MHz front side bus speed would result in numerous problems. It was assumed that the 44 MHz PCI clock that would result from a 133 MHz FSB would cause the malfunction of PCI cards, and in fact, this PCI bus speed did. It was not until the automatic 1/4 PCI bus divider was discovered by a hardware tweaker named Mark Booth that we came to realize that the "133 MHz FSB [is] a viable alternative to 100 MHz, 103 MHz, and 112 MHz front side busses." Armed with his oscilloscope, Mark showed that his ASUS P2B 440BX based motherboard would automatically choose the most desirable PCI bus divider in order to attempt to maintain the 33 MHz PCI bus standard. It was proven that at speeds up to 133 MHz, the motherboard chose a 1/3 divider (causing PCI bus problems) but at the 133 MHz speed the chipset would use the desired 1/4 clock divider.

It was later shown that it was not only Mark's ASUS P2B that had this property, but all newer 440BX based motherboards. With this discovery, we could rest assured that an out of spec PCI bus would not be a problem and that we could safely remove the PCI divider problem from our list of road blocks.



Memory- Another problem forgotten

The second item on the BX 133 problem list was the memory interface. Along with the introduction of the 440BX chipset, a new memory type was beginning to take form. PC100 SDRAM chips were necessary to work at the motherboard's stock 100 MHz front side bus. Since the memory interacts with the rest of the computer without the use of any divider, the memory actually runs at the FSB's speed. Although PC100 SDRAM modules had no problem running at the 100 MHz specified speed, problem began to arise when the FSB (and subsequently the memory) was pushed to 133 MHz.

No longer did the rated 100 MHz suffice for performance. Running the 100 MHz RAM at 133 MHz caused all sorts of problems. Since RAM quality and maximum speed varies by batch, some overclockers were left buying RAM module after RAM module until one DIMM performed properly at the 133 MHz speed. After a bit of time, some companies became known for producing SDRAM chips that could usually hit the 133 MHz limit.

As discussed in our PC133 Memory Roundup, it was VIA's Apollo Pro 133A chipset that originally called for a new memory standard to be produced. Since Intel refused to accept the need for PC133 SDRAM, due to the fact that all of the official 133 MHz FSB solutions used either RDRAM or PC100 SDRAM, VIA actually introduced the PC133 standard into the industry. By working with memory producers, PC133 SDRAM was quickly adopted, especially when Intel revised their roadmap to allow for the i815 to use PC133 SDRAM.

With the introduction of the PC133 standard, users could rest assured that their RAM would function correctly with a 133 MHz FSB speed. No longer would out of spec memory speeds cause system crashes or corruption, removing yet another problem from our potential BX-133 road blocks.

On a related note, we have heard of some problems running 440BX boards at 133 MHz with certain RAM configurations. In the lab we have duplicated these results by using two or more Corsair Infineon DIMMS as well as two or more Micron -7e DIMMS. In this test, we used Mushkin Nanya RAM, Memman Infineon RAM, and Corsair Toshiba RAM. Each stick worked fine by itself as well as in any combination with other sticks when at 133 MHz. When the FSB was pushed to 148 MHz, the 2 DIMM configuration worked fine however using all 3 DIMMs resulted in problems.



AGP Bus Divider- A lingering problem

So far, we have eliminated 2 problems out of 3 that prevent a 440BX motherboard from running smoothly at 133 MHz. All we are left with the AGP bus divider, a problem that still remains.

Like the PCI bus, the AGP bus does not operate at full front side bus speed. In a similar manner to the PCI bus's 33 MHz specification, AGP products are designed to interface with the computer via a 66 MHz AGP bus speed. As we noted before, the 440BX was designed around a 100 MHz FSB, a fact which produced all the aforementioned problems. Running at a 100 MHz FSB requires a 2/3 divider to achieve the desired 66 MHz AGP bus speed. Unlike the 1/3 PCI bus divider that was found to switch to a more favorable 1/4 divider when at 133 MHz, the AGP bus divider on the 440BX is set to an unchangeable 2/3. This means that when overclocking, the AGP bus jumps from a stable 66 MHz to an out of spec 88 MHz, speeds which can cause system crashes and failures due to video cards running out of specification. We can therefore see that really the only limiting factor to running a BX board at 133 MHz full time is the locked 2/3 AGP bus divider. Other 133 MHz FSB motherboard solutions, such as the i815 and the Apollo Pro 133A, provide a fix to this problem by using the proper 1/2 divider necessary.

AGP Bus Divider- A solution

Luckily for us, there is a way around this problem. Some have found that certain video cards perform correctly at this grossly out of spec speed, allowing them to run their BX systems at 133 MHz or beyond even with the 88 MHz AGP bus speed. By testing the collection of video cards we have in the lab, we could accurately predict what cards will work on the 88 MHz AGP bus. Be it if you currently use a 440BX running at 133 MHz and are looking to upgrade your video card or if you are looking to push your BX system up to the 133 MHz mark, the following results should help you achieve your goal.



The Test

In order to test if a card performs correctly in a BX system set to 133 MHz FSB, we used an AOpen AX6BC Pro as a standard BX motherboard. The motherboard was equipped with 128MB of PC133 RAM manufactured by Mushkin. Next, using a Pentium III 733 E, we upped the FSB frequencies to 133 MHz first and then to 148 MHz (88 MHz and 98 MHz AGP bus speeds respectively). Stability tests for the video cards were then performed by looping Quake III Arena for a total of 15 minutes. A successful completion of a test was noted by a flawless run of Quake III Arena for this time period. This test was performed on a total of 27 video cards that would most likely be used for gaming in an overclocked BX-133 system.

 

Windows 98 SE Test System

 

Hardware

CPU(s)

Intel Pentium III 750E
provided by Memman

Motherboard(s)
AOpen AX6BC Pro
Memory

128MB PC133 Crucial Technology SDRAM

Hard Drive

Westen Digital Caviar 205AA 20GB UDMA 33

CDROM

Acer 24x

Video Card(s)

Various, as noted in following sections

Software

Operating System

Windows 98 SE

Video Drivers

Various

 

Benchmarking Applications
Gaming

idSoftware Quake III Arena demo001.dm3



The Results- An overview

The first set of cards that we tested, NVIDIA GeForce 2 GTS based cards, gave us a glimpse of what overclocking the AGP bus does. The first thing that we noted during testing was that if a card was not going to pass the tests, it usually let us know. In a very similar way to how overclocking a video card's memory or core produces artifacts on the screen (as shown in our Overclocking the GeForce 2 GTS guide), running a card in an overclocked AGP bus slot produced artifacts as well. Cards that did not finish the test began to show line-like artifacts in the green square at the bottom right hand side of the screen when running a Quake III Arena demo. This normally solid green box began to get vertical black lines running through it before the card actually failed. This serves as a good reference point to see if your card is being pushed to far by an excessive AGP bus speed.

The second item noted, as shown in the following table, is that the two cards that had sideband addressing (SBA) enabled in the card's BIOS provided for the most problems. In fact, these two cards were the only ones that failed both the 133 and 148 MHz tests. Sideband addressing was actually one of the original features to be boasted in the AGP bus. Allowing a video card to add three special extensions on the sideband channels, sideband addressing theoretically allowed for more AGP bandwidth. This feature, which is turned on or off in the BIOS depending on the video card manufacturer's choice, does not add any noticeable speed to an AGP card. To prove this, we tested the Absolute Multimedia Outrageous GeForce 2 GTS in Quake III Arena Demo001. With SBA enabled, the default state of the card, a frame rate of 114.4 FPS was recorded. When SBA was disabled using PowerStrip, the frame rate in the the same benchmark was recorded to be 114.4 FPS: the exact same speed as with SBA on.

With these results, one can not question that SBA provides no noticeable speed increase. What it does appear to do, however, is limit the functionality of a video card in an overclocked AGP bus. Sideband addressing provides for a more complex AGP bus route, meaning that there is more opportunity for failure.

To check and see if sideband addressing was enabled for an individual card, we used En Tech's PowerStrip. Not only did PowerStrip allow us to see if SBA was enabled or not, it also allowed us to turn on or off this feature. By turning off SBA on the cards that had it enabled, we achieved a 100% success rate when running the cards in overclocked AGP bus', as shown in the following table.



The Results- NVIDIA Based Cards

GeForce 2 GTS based cards

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
Absolute Multimedia Outrageous GeForce 2 GTS
Enabled
Fail
Fail
Pass
Pass
Disabled
Pass
Pass

ELSA Gladiac

Disabled
Pass
Fail
Disabled
Pass
Pass
Enabled
Fail
Fail
Pass
Pass

Many of the GeForce 2 GTS cards tested did not enable SBA by default. This is most likely a product of evolution, where more and more manufacturers realized that it caused more problems than it solved. This can especially be seen when comparing these cards to the TNT2 Ultra cards, all of which enable SBA by default.

The GeForce 2 GTS cards all functioned properly with SBA disabled. The one exception was the ELSA Gladiac which failed to run at our 98 MHz AGP bus speed.

GeForce 256 DDR based cards

Similar to the GeForce 2 GTS cards, the GeForce 256 DDR based cards had SBA disabled for the most part, as seen below.

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
Disabled
Pass
Pass
Disabled
Pass
Pass

ELSA Erazor X2

Disabled
Pass
Pass
Enabled
Pass
Pass

The one GeForce 256 DDR based card with SBA enabled, the Leadtek Winfast GeForce DDR Revision B, had no problem completing our tests at either 133 MHz or 148 MHz on the FSB. All the other cards had this feature disabled, once again most likely due to the fact that it seems to cause more harm than good.

GeForce 256 SDR based cards

As we progress back in time, we find that more and more cards with SBA enabled begin to appear. No longer are SBA cards a dying breed in GeForce 256 SDR based cards, but 50% of the cards tested had this feature enabled. Below are the results from our overclocking tests.

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
Disabled
Pass
Pass
Creative Labs 3D Blaster Annhilator
Enabled
Pass
Pass

ELSA Erazor X

Enabled
Pass
Pass
Disabled
Pass
Pass

As was the case with the GeForce 256 DDR, we find that SBA really does not seem to effect a GeForce 256 SDR based card's ability to perform in an overclocked AGP slot. Too bad this does not prove to be the case throughout all the video cards tested.

Quadro based cards

The Quadro, a professional version of the GeForce 256, is really only to be found in ELSA's Gloria II card. Targeted at the graphic designer, the Gloria II comes with SBA turned on. Let take a look at how it survived in the tests.

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
Enabled
Pass
Fail
Pass

As we saw and noted with the GeForce 2 GTS, the maximum bus speed that a card will run on seems to be dramatically increased by disabling SBA. In the case of the Gloria II, turning SBA off allowed for the card to run in our 98 MHz AGP bus.



The Results- NVIDIA Based Cards Continued

TNT2 M64 based cards

The only M64 based card that we have, the Guillemot Maxi Gamer Cougar, came with SBA disabled and had no problem performing in the overclocked systems.

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
Disabled
Pass
Pass

 

TNT2 Pro based cards

Once again, we only have one test card for this TNT2 core derivative The Gigabyte GA-660 provided this system. With SBA enabled, the card had some problems, as shown below.

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
Enabled
Fail
Fail
Pass
Pass

One more time we see how disabling SBA allows for reduced noise in the signal and thus higher bus speeds to be used at no loss of performance.

TNT2 Ultra based cards

Once again, as we turn back the clock we find that more and more cards come with SBA enabled. Take a look at what our TNT2 Ultra test showed.

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
Enabled
Fail
Fail
Pass
Pass
Enabled
Fail
Fail
Pass
Fail

Hercules Dynamite TNT2 Ultra

Enabled
Fail
Fail
Pass
Fail

Allowing SBA to be enabled in TNT2 Ultra systems seems to be disastrous. Every card tested had this feature enabled and every card failed our overclocking test. It was not until SBA was disabled did we begin to see some respectable performance, with some cards still not being able to work on our 98 MHz AGP bus.

The most likely cause of the high failure rate seen in these cards when compared to GeForce cards is due to the fact that these cards were produced somewhat early in the game. Time has only refined the video card manufacturing process and card manufacturers have become more aware of people using these cards at AGP bus speeds in excess of 66 MHz.

TNT Based cards

Reaching into the back of our shelves, we were able to find a card based on the final NVIDIA chipset tested: the Creative Labs Graphics Blaster TNT. Not surprisingly, this card had some problems running in our overclocked system.

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
Enabled
Pass
Fail
Fail


The Results- ATI, Matrox, and S3

ATI cards

Unlike NVIDIA based cards, where multiple manufacturers make the cards, ATI's ability to produce their own products provided for standardization around the board. Take a look at the table below, noting that each card comes with SBA enabled.

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
Enabled
Pass
Pass
Enabled
Pass
Fail
Fail

All In Wonder 128 Pro

Enabled
Pass
Pass

It seems that ATI cards are a bit less picky when it comes to working in overclocked systems with SBA enabled. The only card which failed our 148 MHz FSB test was the Rage Fury MAXX. At this speed, one of the two processors onboard the MAXX appears to have failed, as only 1/2 of the Quake III Arena scenes were being rendered. This made for quite an interesting display.

Matrox cards

Although Matrox cards are not considered the ideal solution for gamers, many still hold on to them for their superior 2D quality. Let's take a look and see how these cards perform in overclocked systems.

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
Enabled
Pass
Pass
Enabled
Pass
Fail
Fail

G200

Enabled
Pass
Pass

It is nice to know that the most game worthy of the Matrox cards, the G400 Max, seems to have no problem running in overclocked systems.

S3 cards

Of all the cards tested, we experienced the worst results with the Savage 2000. This card refused to boot when at 148 MHz or 133 MHz This is one card you don't want to put in a BX-133 system.

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
?
Fail
Fail


The Results- 3dfx

3dfx's Voodoo line remains a popular video card solution for many people, including BX motherboard owners. To 3dfx's advantage, the Voodoo line has never fully implemented AGP texturing and thus has never been extremely dependent on the AGP bus. As a result of this, we found that each Voodoo card tested performed flawlessly in the 133 MHz FSB systems. Attempting to run the cards in the 148 MHz front side bus systems did not equate to the same success.

Card
SBA state
133 MHz
148 MHz
133 MHz SBA off
148 MHz SBA off
Enabled
Pass
Fail
Fail
Enabled
Pass
Fail
Fail

Voodoo3 3000

Enabled
Pass
Fail
Fail
Voodoo3 2000
Enabled
Pass
Fail
Fail

The Voodoo cards experienced problems running in the 98 MHz AGP slot. Disabling SBA, a procedure that solved some of the problems experienced on other systems, was not met with the same success as in previous systems. The Voodoo cards produced strange errors when at this AGP bus speed, usually not resulting in complete system failure as in previous cards but rather an unexpected break form the Quake III loop test. This might not be enough to stop the hard core overclocker from attempting high FSB settings, but it may deter those worried about system stability.

On a positive note, the Voodoo cards had no problem working at the 133 MHz FSB settings even with SBA enabled. While we have shown that SBA does not result in any performance increase, leaving it enabled prevents having to disable SBA at each system boot to ensure stability. Please note that the results in this test as well as the others performed are not necessarily indicative of all Voodoo cards out there. As with all overclocking ventures, your results may vary.



Conclusion

With the memory and PCI divider problems encountered by the pioneerers of the 133 MHz FSB well behind us, the only problem that has been standing in the way of the BX-133's success has been the AGP divider. AnandTech is here to remove that road block as well, clearing the path to success.

It seems that almost every newer card on the market today will function at a 133 MHz FSB or even a 148 MHz FSB. This is good news not only for overclockers who wish to verify compatibility before upgrading their video card but also for those thinking about a BX-133 conversion in the future.

For those with older video cards, there is a bit to be cautious about. Check out the above tables and see if your particular card will work with an overclocked AGP bus. In many cases, a card may begin to work simply by disabling the SBA function in PowerStrip under the diagnostics feature.

The best video card to get for an overclocked BX-133 system seems to be one with sideband addressing disabled. It is possible as well as feasible to have PowerStrip run every startup to disable SBA, however it is simpler and cleaner to use a card that has this feature disabled straight from the factory. The only card to stay very far away from is the Savage 2000, as it will not function at all in an overclocked system.

With the AGP divider problem no longer an issue, we have eliminated the final obstacle preventing a BX-133 system from becoming reality. Using this new found knowledge, nothing stands between your 440BX motherboard and a faster system.

Log in

Don't have an account? Sign up now