intel 440BX chipset motherboard roundup!

[chrisNova777]

https://www.anandtech.com/show/556

he Intel 440BX chipset has been with us ever since it was introduced in May of 1998.  This is quite unusual for a Slot-1 chipset since the first two chipsets for the Pentium II never lasted more than a few months.  The first Pentium II chipset, the 440FX, lasted only a few months, from the introduction of the Pentium II until August of 1997 when the 440LX chipset made its debut.  The latter managed to stay alive for 7 months before being replaced by the 440BX chipset.  So why is it that the BX chipset has been around for an incredible 24 months and is still being used by motherboard manufacturers?

In order to answer that question you have to go back to the theory that necessity is the mother of invention.

Intel needed a chipset for the Pentium II and they needed it at the release of the CPU in May and not a few months later.  What they ended up doing was taking the 440FX chipset, otherwise known as the Natoma, which was used on entry-level Pentium Pro motherboards and presented it to motherboard manufacturers as a Pentium II solution as well (because of the fact that the Pentium II used the same bus as the Pentium Pro).

While the Pentium II was gaining momentum, Intel was working on implementing a "new" graphics bus into their next 440 chipset, which ended up being the 440LX, the world's first AGP enabled chipset.

The upgrade to the BX came about because Intel felt the need to leave the limiting 66MHz FSB of the Pentium II 333/300/266/233 behind and replace it with a faster 100MHz FSB frequency.  This increase in FSB frequency would not only lower the clock multiplier of future Pentium II CPUs but it primarily offered a higher bandwidth data path from the CPU to the chipset and to the memory.  With the AGP bus taking up to 533MB/s of system bus/memory bandwidth, the 533MB/s of available memory bandwidth on the 66MHz 440LX chipset could potentially become a limiting factor; by increasing the system bus and memory bus operating frequency to 100MHz, the amount of available memory bandwidth also increased to 800MB/s.

The next step in the evolution of Intel chipsets came with what was then known as the Camino chipset, the successor to the popular BX.  It was originally thought that the Camino chipset, now known as the i820, would offer everything the BX chipset had to offer while adding 133MHz FSB support as well as Ultra DMA 66 support.  As more information was released, it quickly became known that the i820 would support a brand new type of memory, RDRAM, but at the same time, the chipset would be able to work alongside SDRAM.

While all of these rumors ended up coming true in one sense or another – i820 did add 133MHz FSB/Ultra DMA 66 support and it did support SDRAM in addition to RDRAM (however only if a Memory Translator Hub was implemented on the motherboard) – the fact of the matter was that the i820 as a platform was not affordable enough (thanks to the high price associated with RDRAM), didn't offer a large enough performance improvement over the 440BX, and its even poorer performance and possibly instability when used in conjunction with SDRAM and an MTH made this chipset a highly undesirable solution, even by motherboard manufacturers.

This put motherboard manufacturers in an interesting situation.  Intel was obviously pressuring them to promote and sell as many i820 motherboards as possible yet they couldn't since there wasn't a great enough demand for them.  VIA began offering their Apollo Pro 133A chipset to motherboard manufacturers that needed 133MHz FSB support without having to move to i820.  Unfortunately, in the usual manner of VIA chipsets, the Apollo Pro 133A's overall performance was not the best, and in a head-to-head comparison, at the 100MHz FSB the Intel 440BX chipset would actually pull out ahead.

The only real advantage VIA's 133A offered over the 440BX was that it officially supported the 133MHz FSB frequency, and with the 1/2 AGP multiplier, that would allow the AGP bus to operate within spec when the FSB is set to 133MHz (133/2 = 66MHz = AGP spec clock speed).

Even the first BX motherboards ever released featured unofficial support for the 133MHz FSB setting, but there was a lack of memory that could run at that frequency (as memory was just starting to ship as PC100 compatible) as well as a lack of AGP video cards that were capable of running at 89MHz, which is what the AGP bus would operate at when the FSB was raised to 133MHz (133 * 2/3 = 89MHz = 35% over 66MHz spec).

BX at 133MHz
As we mentioned in our RDRAM Performance article, the BX chipset is now capable of running, albeit unofficially, at the 133MHz FSB frequency.  What has changed since May 1998 that allows for a BX motherboard to run at 133MHz?

For one thing, motherboard manufacturers have been tweaking their designs quite a bit over the past two years.  The BX motherboard platform in general is at the point where you shouldn't have to worry too much about variations in performance or stability when going from one motherboard to the next.  This perfection of the motherboard design, especially from the companies that have had quite a few BX boards (i.e. ASUS, ABIT…) has made their reliability at higher FSB frequencies much more of a reality and less of a dream.

Secondly, the PC133 memory standard has been completed and implemented by VIA as well as refined by Intel.  There is finally memory available that was designed with a 133MHz operating frequency in mind, and not too long ago Micron began shipping their –7E parts, which are officially rated at 133MHz CAS2, which provides for an additional 5 – 10% performance improvement over 133MHz CAS3.


Click to Enlarge

Finally, video card manufacturers have been designing their video cards to operate at a greater range of frequencies that remain outside of the 66MHz AGP specification.

A combination of all three of these factors has made it possible to run BX motherboards at the 133MHz frequency, even without the presence of a 1/2 AGP clock divider (there would have to be a revision of the BX North Bridge in order to add support for that divider).  Now, not all BX motherboards are capable of running at the 133MHz FSB, but there is definitely a high chance of it working on the latest BX motherboards, especially those produced by such companies as ABIT, ASUS, AOpen, and MSI, to name a few.

Ultra DMA 66 on a BX?
It doesn't make sense to buy a BX motherboard now with hopes of it remaining capable of running the latest processors in a few months (Willamette will use a completely different bus), but one thing a lot of potential BX motherboards owners were worried about was not having Ultra DMA 66 support on their BX motherboards.

As we proved in our Ultra DMA 33 vs. Ultra DMA 66 comparison, the Ultra DMA 66 specification does not provide any tangible performance benefits for today's hard drives but that is quickly changing.  The IBM Deskstar 75GXP is supposed to be able to provide performance that is limited by the Ultra DMA 33 specification, which could cause problems for BX motherboard users since their boards would be limiting their disk performance.

Last year, when it became clear that the BX chipset would be around for at least a little while longer while Intel readied the i820 chipset, motherboard manufacturers began adding external Ultra DMA 66 controllers to their motherboards.  At that time, there wasn't really a need for Ultra DMA 66 support since no hard drives could burst at above 33MB/s, but quite a few users went after the motherboards simply because they supported Ultra DMA 66.

There are a few options for users when it comes to having Ultra DMA 66 support on a BX motherboard.  Currently Promise, CMD and High Point manufacture controllers that are being used on BX motherboards in order to add Ultra DMA 66 support.  Moreover, if your motherboard doesn't feature either one of those on-board controllers, you can always purchase an add-on card that features either one of the controllers.

High Point HPT366


Used on: ABIT BE6-II & Soyo SY-6BA+IV

Promise PDC20262


Used on: Gigabyte GA-6BX7+ & Microstar BXMaster

CMD 648


Used on: ASUS CUBX

AGP 2X vs. AGP 4X
Another one of the paper advantages the i820 and Apollo Pro 133A chipsets hold over the old 440BX is their support for AGP 4X transfer modes which are theoretically twice as fast as the AGP 2X transfer rates supported by the BX chipset.

The 32-bit wide AGP bus, when operating in 2X mode allows for a peak transfer rate of 533MB/s.  The same AGP bus, when operating in 4X mode allows for a peak transfer rate of 1.06GB/s.  Going by those two numbers alone, you definitely see where AGP 4X can hold a performance improvement over AGP 2X, but if you take into account that the amount of available memory bandwidth on your graphics card is going to be between 3 – 5GB/s (2.7GB/s for GeForce and 5.3GB/s for GeForce 2), all of the sudden this 1.06GB/s of memory bandwidth offered by AGP 4X isn't all that great.

The performance hit you get when going from local memory on your graphics card to system memory via the AGP bus is so great that the difference between the AGP 4X transfer rates and AGP 2X transfer rates remains of very little significance.

Another thing to take into account is that, since the BX chipset only supports an AGP to FSB ratio of 2/3 or 1/1, at 133MHz FSB the AGP bus will be running at 89MHz which is a full 33% over the 66MHz specification.  This also translates into a higher transfer rate across the AGP bus since the operating frequency of the bus is higher.  More specifically, at 89MHz, you get something along the lines of an AGP 3X transfer rate although a bit slower than what that would actually be (since 100MHz AGP would theoretically be equal to AGP 3X).  The actual peak transfer rate across the AGP bus then becomes around 712MB/s which is a 34% increase over the 533MB/s of AGP 2X.

In order to prove that the difference between AGP 4X and AGP 2X is negligible, we naturally ran a set of benchmarks comparing the two.  In order to eliminate all potential bottlenecks and truly compare AGP 4X and AGP 2X, we ran the benchmarks on an i820 platform with a GeForce 2 GTS.  For comparison's sake, we've included AGP 1X scores as well.



As you can see, in a normal gaming situation, there is very little difference between AGP 4X and AGP 2X.



Even in a memory intensive situation such as Quaver, the difference is not that great. Although in this particular case, the GeForce's S3TC support as well as enhanced texture management routines that are a part of the 5.22 Detonator drivers increase the efficiency of the use of local graphics memory thus minimizing the need for AGP texturing.



In a high end test which is represented by SPECviewperf, the performance difference between AGP 2X and AGP 4X is negligable but there is definitely a huge difference present between AGP 1X and the latter two transfer modes

The Candidates

We rounded up a total of seven BX motherboards for this roundup:

ABIT BF6 https://www.anandtech.com/show/431


ABIT BE6-II https://www.anandtech.com/show/556/7


ASUS CUBX https://www.anandtech.com/show/556/9


Gigabyte GA-6BX7+ https://www.anandtech.com/show/556/10


Microstar BXMaster https://www.anandtech.com/show/556/11


Soyo SY-6BA+IV https://www.anandtech.com/show/556/12


AOpen AX6BXC Pro Gold https://www.anandtech.com/show/556/8

The first thing we noticed after running through all of the benchmarks and stability tests was that, overall, each one of the seven boards performed just about equally in terms of stability when running at 133MHz.

No board crashed more than three times during a 24-hour looped run of Content Creation Winstone 2000.  This is compared to the 6+ times that most average Apollo Pro 133A and VIA KX133 motherboards crash during the same 24-hour period.  The BX platform is definitely very refined, and even when overclocked, provided that you have properly selected your components (PCI cards don't really matter since you can run your PCI bus at 133MHz / 4 which keeps them in spec at 33MHz), your BX133 platform should be just as stable as any other 133MHz platform out there.

The highest we could push any of these boards reliably was around the 155MHz FSB.  The Soyo SY-6BA+IV was one of the only boards to run our 733MHz test chip at 155MHz x 5.5 reliably, even while running 3D games and applications.  But at 155MHz x 5.5, there was a noticeable drop in stability when compared to the SY-6BA+IV at the 133MHz setting.  We could've probably pushed the board even higher, but it lacked the FSB settings to go any higher.  Before you start asking, our ABIT BF6 was only able to get to around 150MHz before our benchmarks would no longer run reliably, so the 1MHz FSB increments above 150MHz weren't of much use.  We tested this using Micron –7E SDRAM, which is rated at 133MHz CAS2 and is the only currently available PC133 SDRAM capable of running at 133MHz CAS2.

Another issue we encountered was that on the ASUS CUBX, the CMD controller that provides for Ultra DMA 66 functionality required that we manually enable the Ultra DMA 66 setting, in spite of the fact that we were using Ultra DMA 66 drives and cables.

Regarding all of the boards that feature external Ultra DMA 66 controllers, if the drivers for those controllers were not installed properly or at all from the start, we noticed very erratic behavior under Windows often resulting in random lockups and failures to boot Windows properly, so make sure you get those drivers installed.

The HighPoint controller on the ABIT and Soyo motherboards was the only Ultra DMA 66 controller to come up as two devices under the SCSI devices section of Windows' Device Manager.  At the same time, the CMD controller on the ASUS CUBX was the only controller to come up properly as an IDE controller.  Those are just some of the odd quirks about working with these boards.