The FSB Survey

.

Introduction

First, I’d like to thank the almost 175 people who took the time to report on their machines. Without you, this wouldn’t have been possible.

The Myth of Uniformity

Many people think that a product either should or should not work under certain conditions. They believe this is the way it ought to be; they want consistency, in large part because it makes their lives simpler.

Unfortunately, life is not so simple. This survey shows conclusively that you can’t think that way and get the expected results. You can’t buy a BX board, the latest RAM, the proper latest cB0 stepping Coppermine, a GeForce card, and be guaranteed 150Mhz. You certainly improve your odds quite a bit over that of reaching that goal with older equipment, but that’s it.

So this survey is not going to give you a sure-fire guarantee of putting together a system that reaches 150Mhz. (Well, actually it does recommend a very likely way, but if you dream of 1Ghz, you won’t like it.) It doesn’t exist. Sometimes it will work. Sometimes it won’t. If it worked for you, great, but you aren’t everybody. There’s no guarantees, just likelihoods and probabilties.

This is also why most surveys run by websites are useless. No website gets crates of a product to test. If you think someone testing a couple video cards guarantees you 150Mhz, it doesn’t.

Methodology and Aims

Skip if you’re not a statistics buff

We asked for people to tell us how they did, they told us, we tabulated them. How valid is this?

No doubt there will be a few of you who will holler that this survey is not “statistically valid,” and come up with all kinds of reasons.

If my goal was to show that there was a 26% chance you’ll reach 150Mhz FSB, plus or minus 3%, you’d be right.

But I’m not out to do that. I’m out to get a rough idea as to success and failure rates. I put up percentages, but they aren’t meant to be taken literally. If I say 25% of something does something, a perfect test might reveal the 25% to be really 15% or 35%. For our purposes, though, it doesn’t matter. We just want a rough idea, and for that, you need relatively little statistical data.

This is also true when a certain result almost always happens. If something happens over 90% of the time; it doesn’t take much to show statistically that it does indeed happen just about all the time (again, if you’re not concerned about exactitude).

A more valid argument is that results can vary considerably based on the expertise of the user. We were able to extract data from the survey, and it indeed shows you can increase your chances of success with the proper choice of equipment. However, it also shows that there is no guarantee of success.

What I did

I took the surveys and tabulated them on an Excel spreadsheet. Then, based on what you said, plus my knowledge, I tried to assess the most probable reason why the machine couldn’t go any further, based on the equipment in the machine. If you reached 150Mhz, that was considered a “success” (even though you might have wanted more). If there was no particular indication a person tried to go further with the machine, that was considered a reason.

After doing all this, I then sorted them a variety of ways and tabulated the speed ranges by category of motherboard, model of motherboard, and type of video card. Other factors were tabulated as part of the “reason the machine didn’t go further.

General conclusions:

  • There was a limited selection of equipment used; most people tended to use just a handful of items.
  • About 90% of those surveyed reached 133Mhz. The number drops to about half reaching 140Mhz, and between a quarter to a fifth reaching 150Mhz. Only one BX motherboard (the Abit BF6) significantly varied from this pattern.
  • On the whole, BX boards do better than Via boards, but, outside of the BF6, not tremendously better. BX board users were far more likely than Via users to be satisfied with a 133Mhz or sub-140Mhz result, probably due to usually having older equipment than Via, along with fear of overclocking the AGP bus. Via board users tend to have newer equipment, and pushed more. Via boards also seem to have more problems than BX boards. Anecdotal comments by users who had used both BX and Via overwhelmingly preferred the BX board.
  • There is an AGP bus speed effect on BX boards, which seems to affect about 20% of BX board users. Unfortunately, most of this occurred at the lower speed ranges.
  • Newer, high-quality equipment improves the odds of success, but does not ensure it. If you carefully select new equipment, you can increase the odds in your favor. If you choose one particular set and settle for a few less Mhz, you can do even better than that (see below).
  • Older equipment can usually reach 133Mhz, but faces increasing difficulties getting past 140Mhz.
  • The only real dog that emerged among motherboards (though with few respondents) were the Tyan Via boards (few Via boards were represented in this survey).
  • For video cards, Voodoo AGP cards were somewhat more likely to cause problems than other cards. Matrox cards were less likely to reach 150Mhz than others, nor did NVidia TNT2 cards. GeForce cards tended to do better, but not to a marked degree.

  • It is probably more realistic, especially if you are trying to overclock with the 650E or better, to first aim at 133Mhz, and consider anything above that gravy.

Results

Most survey respondents owned BX boards, over 70% of them. Via board users accounted for a bit over 25% of the respondents.

Almost two-thirds of BX board respondents used one of four boards: the Abit BE6-2 and BF6, the Asus P3B-F, and the Soyo 6BA-IV. Close to 60% of Via users used the P3V4X, and about 15%
used the Soyo 6/7VCA.

About 70% of CPUs tested were Coppermine cA2 steppings, and about 75% of the processors had a rated speed of 600Mhz or less.

Over 40% used a GeForce card, about 20% used a TNT2 card, and about 15% some form of Voodoo 3 card.

BX Boards

How well did they do?

  • Reached 133Mhz or better: 90%
  • Reached 140Mhz or better: 54%
  • Reached 150Mhz or better: 26%

Why didn’t they do better?

46% did not reach 140Mhz. Of that 46%:

  • 15% did not appear to make it primarily due to their CPU.
  • 15% did not appear to make it primarily due to their video.
  •   9% were satisifed with the speed they reached.
  •   5% did not appear to make it primarily due to their RAM.
  •   2% did not appear to make it due to their motherboard.

28% got to 140Mhz, but didn’t reach 150Mhz. Of that 28%:

  • 11% did not appear to make it primarily due to their CPU.
  •    7% did not appear to make it primarily due to their RAM.
  •    6% were satisfied with the speed they reached.
  •    4% did not appear to make it primarily due to their video.

So of the 74% that did not make 150Mhz:

  • 26% did not make it primarily due to their CPU.
  • 19% did not make it primarily due to their video.
  • 15% were satisfied with the speed reached.
  • 12% did not make it primarily due to their RAM.

Via boards

  • Reached 133Mhz or better: 83% (this goes to 88% if you drop the Tyan scores)
  • Reached 140Mhz or better: 54%
  • Reached 150Mhz or better: 19%

Why didn’t they do better?

46% did not reach 140Mhz. Of that 46%:

  • 19% did not appear to make it primarily due to their  motherboard.
  •    9% did not appear to make it primarily due to their RAM.
  •    9% did not appear to make it primarily due to their CPU.
  •    5% did not appear to make it due to their video.
  •    2% were satisfied with the speed they reached.

38% got to 140Mhz, but didn’t reach 150Mhz. Of that 28%

  • 24% did not appear to make it primarily due to their motherboard.
  •    9% did not appear to make it primarily due to their CPU.
  •    2% were satisfied with the speed they reached.
  •    2% did not appear to make it primarily due to their video.

So of the 81% that did not make 150Mhz:

  • 43% did not make it primarily due to their motherboard.
  • 18% did not make it primarily due to their CPU.
  •    9% did not make it primarily due to their RAM.
  •    7% did not make it primarily due to their video.
  •    5% were satisfied with the speed reached.

Analysis

Old vs. New

There’s one big factor which explains many of the differences between the findings, and which may make Via boards
look a bit better comparatively than BX boards than they ought. BX board users tend to use older equipment than Via users.
When I looked at a BX result, I almost always could find a likely cause for the board not running faster than it did in the subcomponents.
That doesn’t mean every BX board would run at 150Mhz without these problems; it’s the “just because I got cured of cancer doesn’t mean I won’t die of something else eventually.”
If that factor were fixed, it’s likely something else would come up, especially as you approached 150Mhz. So this makes BX boards probably look more reliable than they actually are.

With Via boards, though, I often couldn’t find any other apparent reason for it not running faster. They tended to have the newer RAM and video cards, so I couldn’t point blame at something else easily.

I suspect that if you equalized the equipment factor, you’d see a higher percentage of motherboard problems with the BX, but that the gap between the percentage of boards reaching 150Mhz or better would also increase.

Perhaps the best factor to consider is the opinion of those who used both BX and Via boards in their endeavors. About 90% of them found they did better with BX than with Via. At the end of this piece, I’m going to include observations from someone who builds a lot of systems, and whose observations pretty much correlate with what I found.

Video is the hobgoblin

You can get better RAM (though that doesn’t always work). You can match the CPU you buy to maximize the possibility of reaching 150Mhz (right now, that’s a cB0 600E).
But every once in a while, the video card comes out and bites you. Not all that often, but statistically enough to beware, and there’s no apparent way to eliminate the problem.

I get the sense this is mostly due to variation among components.

There are a few that are worse than others. The Voodoo 3 cards seem to cluster around the 133Mhz mark more than other modern cards, especially when there’s no apparent other reason for a system not to run faster. Some hit 150Mhz, but you
don’t see the continuum between 133 and 150 that you do with other cards.

Matrox cards don’t seem to have much problem with 133Mhz or even 140Mhz, but 150Mhz is pretty rare. The same could be said for TNT2 cards.

If I had to pick one type video card, it would be the Nvidia GeForce cards; but they’re no magic bullet, either. A small proportion of them die out at relatively low speeds, too.

Magic mobos?

With one exception, there were no significant differences in performance among the boards. Newer boards did better than older ones when reaching towards 150Mhz, but older mobos didn’t seem to have any bigger problems getting to 133Mhz than the newer ones.

The BF6 looked to be the exception. It reached 150Mhz FSB about 40% of the time, as opposed to the general BX average of about 25%.

For Via boards, we just didn’t get a terribly representative sample. Essentially, we got P3V4Xs, 6/7VCAs and Tyan boards.

We only got 5 Tyan boards, but the ones we got were awful. Only 40% of them even got to 133Mhz. Only a handful, to be sure, and maybe not representative, but it stuck out like a sore thumb amidst the data.

For the rest, they tended to trail the BX boards, but not by a lot. Again, I believe a completely level playing field would widen the gap, but on the whole, the Via boards tended to get to 140 or a little above about as regularly as the BX boards.

Does memory matter?

  • If you have PC100 memory, you may make 133Mhz, you may not. An occasional stick of very high-quality PC100 can reach 150Mhz, but don’t count on it.
  • If you buy generic PC133, occasionally, you won’t even make 133Mhz. Usually, you will, but they start failing as you approach 150Mhz.
  • High-quality PC133 does best of all, but it’s no guarantee.

Do cB0s matter?

Yes, they do, but not consistently. Many people bought CPUs that just weren’t very likely to run at 150Mhz anyway. Even the initial buyers of 500E and 550Es often found that 150Mhz wasn’t attainable.

While we didn’t get a lot of data on the higher speed cB0 stepping chips, what I saw indicated that 150Mhz was a rarity for those chips. For the 700E, not even 133Mhz is a given, though most make it.

Is there a magic bullet to get to 150Mhz?

Well, sort of. 🙂

If you want to have a very high (though not guaranteed, maybe around 80% or a little better) shot at 150Mhz, the system that seems to do it (or at least gets very close) is the following:

  1. A high quality BX board
  2. A cB0 stepping 600E chip
  3. High-quality PC133 RAM
  4. A GeForce card

Of course, you’ll only get 900Mhz even if you reach 150Mhz, but that will probably be slightly faster than a 700E running at 933Mhz at 133Mhz FSB.

But I want to hit 1Ghz!!!!!

You do? Then start reading all the water-cooling and Peltier articles. 🙂

Seriously, until we see the cC0 stepping in September or so, the odds are against you doing it with air-cooling, perhaps 5:1 or 10:1 against. Might not be too good with the cC0 stepping, either.

We’ll know within a month how the socket A Thunderbirds/Duron do for 1Ghz, but even there, those articles I told you to read won’t go to waste.

Finally, an email (I’ll leave the name out until the person says it’s OK to use it).

Hello Ed,

I have built lots of BX/Coppermine machines. The store I work
for has sold 300+ in the last four months (we sell 1500/year), and
right now it’s all 810 and BX-mostly BX. I also overclock like the
dickens. I will use my personal machine to answer your survey
questions, and will also speak of the 10+ BX machines with FSB
in excess of 135MHz that I have built for test, personal, employee,
and sales machines.

1. What processor were you using?

I currently use a SL44Y week 15 S1 600e cB0. I have used many S1 550e’s, OEM

2. Is (was) it an older cA2 or newer cB0 stepping chip?

My current 600e is cB0, all but one of my 550e’s were cA2, and
my one cB0 550e was not the best of the 550’s.

3. What motherboard are (were) you using?

I use the ASUS P3B-F for it’s rock solid stability at 150FSB when
used in conjunction with a GF video card (again, I use ASUS). I have
also used the BF6 ABIT when FSB in excess of the 150 max of the
ASUS was desired.

4. What kind of memory did you get?

I have used 7.5ns PC133 Infineon (any) for most of my high FSB BX
machines, set to 2-3-2-8, but have recently switched to 7ns Vitalic PC
133 set to 2-2-2-8.

5. What video card are (were) you using?

It is my opinion the nearly any quality GeForce video card will tolerate
FSB in excess of 135 on a BX, but I have used primarily ASUS GF’s
to insure the stability and overclockability of my machines. I currently
have the V6800 pure in my machine at 145/333. I look forward to
the V7700 pure GTS card arriving at my doorstop tomorrow morning,
in a UPS truck along with two SL3XK cB0 650’s that may replace the
SL44Y at the house. (if they’ll do 975, that is)

6. What FSB speed are (were) you able to reach stably with this setup?

I have in my personal machine, with the SL44Y 600e, 150FSB at 2-2-2-8
for 900MHz on chip with the stock 1.65V and a Cofan KC500 heatsink w/
6300 rpm Top Motor 40mm fan. Diode temp is 41C.

I have run the ABIT BF6 at 160FSB, 3-3-3-10, with S1 550e’s for 880MHz
at 1.8V with CF500. Hotter, around 46C.

The P3B-F, V6800, and Infineon PC133 is good for 150FSB every single
time, as long as you set the ras-cas delay to 3 (2-3-2-8), and have a cpu
that will stand the implied on chip rate. Slot1 550e at 825 usually goes,
and cB0 600e at 900 is a reality. I will see tomorrow if the cB0 650 can
hack 975.

If there is a weakness to the BX, it would be the multiple dimms may hurt
stability at FSB over 135. I use a single 128MB dimm for all my machines,
and would have to caution the owner that FSB above 140 with multiple dimms
might prove troublesome.

Regards,

Steve Ahrens

PS- I have used a P3V4X, and it was not very stable at all at FSB above 146,
with the same components that worked fine on P3B-F’s and the sorry P3C2K
at 150FSB. Besides, with Sandra numbers for the 150FSB cas2 BX like

PASS
1 2 3
474/521 483/544 507/546

I don’t worry to much about getting the VIA stable.

Email Ed


Be the first to comment

Leave a Reply