Epox 8KHA+ PCI Speeds

A look at PCI speeds above 133 MHz. – Joe

SUMMARY: Based on UXD’s Diagnostic Card, the Epox 8KHA+ does not enable PCI divisors above /4.

Josh was nice enough to send an email asking me to look into the issue of “hidden” /5 and /6 PCI divisors on the Epox 8KHA+. Some are claiming that, after flashing the 11/30/02 BIOS, they are able to run PCI in spec at speeds above 166 FSB – “in spec” meaning that the PCI/AGP bus runs at spec for both 166 and 200 FSB.

There are a number of links for interested readers:

• 3D Center – in German, but seems to be one of the original sources;
• Amdmb.com Forum – a thread in English with additional links and explanations.
• 94228 Chipset Specs – this is the clock generator which enables high PCI divisors.

To test this out, I used my UXD P.H.D. PCI Diagnostic Card which reads PCI bus speeds directly. This card mounts in a PCI slot; after the mobo completes BIOS, the UXD card takes over. Note there is no OS involvement – in running the tests below – the only disc was a floppy.

For these tests, I ran the most conservative memory settings and underclocked a T-Bird 1400 at high FSBs – I ran it at a 6 multiplier. The intent was to read PCI speeds in each BIOS at various FSBs to see what’s going on:

Aside from the downshift at 120 MHz, PCI speeds after 133 MHz keep on climbing. I could not get beyond 170 MHz, but the pattern is clear – if there was a /5 divisor, it would hit at 166 MHz.

As always, a sample of one, but for this particular board, there are no /5 or better divisors. We’ll continue to experiment, but unless the UXD Diagnostic Card is misreading PCIs (it did not for Iwill’s XP333), it does not look like enhanced divisors are enabled in the 11/30/01 (8khal00.bin) BIOS.

If I’m missing something in BIOS to enable these divisors, please drop me a line.
{mospagebreak}

Email Joe

This subject has been beat to death in the forums. By the end of the year this will be ancient history. There has been a lot of arguments on both sides. When Joe asked me to take a look at this issue I thought it might be fun and it was.

From almost the first day the first KT266 reference boards were released there was speculation about a 1/5 divider or multiplier depending on how you look at it. This speculation continued with the release of the KT266A.

Then there were those who claimed the chipset itself did not support it. And then another side said it did. You might think this would be easy to figure out but there is another problem. The only way to get the chipset data sheets is if you are an OEM and sign Non-Disclosure Agreements. It was one of the first things I checked.

Some of our community have made reference to an Anandtech article. In this article is the following excerpt, which I am using under copyright terms of fair use, for purpose of illustration:

“The BIOS, with release date 11/30/2001, provides the ability for you to practically push the FSB of the board to over 200MHz, which has been confirmed by a lot of sources. This is made possible by two very important features by EPoX, the 1/5 and 1/6 PCI bus divider and also the 100/133 memory bus divider. With these two settings, you can be running your FSB at 200MHz while maintaining the memory and PCI buses at 150MHz and 33MHz respectively. The memory could be compensated by the new PC2400 or PC2700 parts, and the PCI bus is still within spec.”

If you are a reviewer and running a web site, this is the one thing you don’t have. However this is the one thing I do have. I can start on a project and devote all my free time to it until I reach a conclusion one way or another.

It only takes a simple test to prove whether or not the dividers are there ’cause I did it with Joe’s XP333R 2.1.

What got my interest was the ICS 94228 PLL. I took a look at the data sheet and saw it was capable of some interesting combinations.

I started digging around the various corners, nooks, and crannies of the internet to see what I could find. I found a lot of discussion about the 8KHA+ and dividers.

I liked the conspiracy theory the best:

The dividers are there but EPoX has hidden them for good reason. If everyone knew they were there no one would buy the new KT333 boards.

Sounds good to me!!

Is there a way to unhide the hidden dividers?? I live far from civilization as you know it. While I was waiting for the board to get here, I discovered my Soltek board had an ICS 94228 PLL. This was my big chance to get a jump on the project.

I started looking around for a good free BIOS Editor to see if I couldn’t dig around and find something. I found myself wishing I had keep up my programing skills instead of working on my soldering skills.

I did find some newer ones, but it didn’t take me to long to figure out I didn’t have the skills to alter the BIOS. The next time you complain about a BIOS, I suggest you try to edit one.

No, not yet. I remembered a program by Wolfram Podien. I’m sure some of you are familiar with CPUFSB – it’s also part of CPUCOOL. You can find his web site here.

This is a great tool for a number of things. You can use it as a diagnostic tool, which is what I wanted. You can also access undocumented settings in BIOS, which I wanted also. I used it with my K7S5A and knew it worked.

If you have a chip that can be fine tuned it is possible to adjust your fsb from windows without having to reboot all the time. You can register this program for about the price of a case fan which will ensure continued development. This is a valuable program for our community as it is updated on a regular basis.

I found this:

And I found this:

Is the conspiracy theory really true?? Are these the hidden dividers?? Is there a way to activate them??

The next thing I did was to check Joe’s data and I did this with a simple test. Joe found a divider at 120 FSB. I ran 3DMark at 119 and 121 FSB. I used SPD and let the computer do the thinking.{mospagebreak}

Default, aka John R. Abaray

Disclaimer:

I do not work for anyone or get paid for anything I write. Everything I write in articles or say in emails is Public Domain for the reader to do with as they wish.

This is what I got at 119 FSB:

This is what I got at 121 FSB:

This should be a No-Brainer for most of you: When the divider changes, the Bus is no longer overclocked and you see a big drop off in the 3DMark score.

The ICS94228 is capable of a divider at 166 FSB, so I ran a test at 165 and 167 FSB.

This is what I got at 165 FSB:

And at 167 FSB:

As you can see, there is no drop off.

You don’t have to believe me. You can try it for yourself.

The next capable divider of the ICS 94228 is at 200 FSB. There was no evidence of a divider here either.

I’ve read all the arguments and I ran tests with every available BIOS including the 8KHL00 which is supposed to have the dividers. I’ve turned over every rock and I could find no dividers, or proof of any dividers, over 1/4.

When I saw the dividers with CPU FSB, I even got a bit excited. Working with CPU FSB, I found you could only access what was in BIOS. Running tests with my K7S5A, I could access an undocumented 142 FSB along with others. With my Soltek board, I could underclock the bus.

I came to the conclusion if it were there I could access it with CPU FSB.

I have read all the illogical logic and all the irrational rationalizations applied to this issue. With all the talent of the Hardcore Elite in our community, if it were possible it would have been done and proven.

There is one other possibility: I could be part of the conspiracy.{mospagebreak}

Default, aka John R. Abaray

Disclaimer:

I do not work for anyone or get paid for anything I write. Everything I write in articles or say in emails is Public Domain for the reader to do with as they wish.

Just a short followup to answer several emails I received:

The 1/4 divider in the 8KHA+BIOS is at 120FSB, which you can see in the data sheet, is also used at 133FSB. My Soltek board with the ICS94228 is set up the same way, only a dip switch is used for 120 & 133 FSB with fine tune used for anything over.

Why they use the 1/4 at 120FSB I don’t know for sure. I have an idea, but there is no test I can do to prove it. I can only surmise there is a technical reason for this. The fine tune capabilities of the ICS 94228 are used for the rest of the FSB adjustments in BIOS.

You could send an email to one of the BIOS Engineers at Epox or one of the others companies using the ICS94228 in their boards. However, I doubt you would get an answer. They would have to write you a book to explain it.

This can be somewhat confusing if you just look at the capability of the chip. What the chip can do and what is used in BIOS are two different things.

When I was researching for the article, I came across the argument that it doesn’t matter what the chip can do, the KT266A can’t handle it. While it sounded reasonable, without the chipset data sheets there is no way to tell. If it could be done, it would have been a selling point hard to pass up by the board makers.

Some readers seem to think there is a divider at 160FSB. You can see from the data sheet there is no divider at 160FSB. It’s easy to see how someone could assume there was one. They looked at 150 and 170, then assumed there was something in between – you can see from the sheet there is nothing there. What you see is what you get. I’ve given you a simple test you can use to determine where the dividers are.

Default, aka John R. Abaray

Disclaimer:

I do not work for anyone or get paid for anything I write. Everything I write in articles or say in emails is Public Domain for the reader to do with as they wish.