Last Pieces of the Puzzle
The prime culprit in the multiplier problem faced by many of you with high-speed TBirds and KT133A has always been “does the processor initialize at the default multiplier?” We mentioned that
about two months ago.
We tried getting some official confirmation back then, but didn’t get any and then started seeing a number of cases where people were able to run CPUs when they shouldn’t have been able to, given what we knew at the time.
Anandtech has an article up in which the final pieces of the puzzle came together and confirmed some of our suspicions.
The Via reference design initializes the CPU at the default multiplier times the applicable FSB until the BIOS takes over (along with the BIOS multiplier).
However, that’s not the only way you can do this.
A motherboard can also be constructed to either initially boot using jumpered connections, or (in jumperless mode) by using a 100Mhz FSB by default for those initial cycles.
We congratulate Anandtech for coming up with those last pieces of the puzzle. We also congratulate them for providing an explanation as to why certain non-destructive modifications serve to lower the multiplier enough in certain cases to allow those with 1200Mhz chips to boot with mobos based on the Via reference design. If you’re in that situation, certainly read the article.
However, we absolutely disagree that the one easy answer to this is just to buy one of the three motherboards listed in the article or any other that does not follow the Via reference design. Not when there is an obvious alternative for those of you who haven’t bought a CPU yet.
Buy a 133Mhz FSB TBird.
The price difference between a 100Mhz and 133Mhz FSB CPU has skidded in the last couple weeks. At the 1.2Ghz level, it’s now only $10-15.
Unlike the Intel CPUs, there is no reason why the 100Mhz FSB TBirds are any better overclockers than the 133Mhz FSB chips. Remember, you can change the multiplier on TBirds, and you can’t on Intel chips. Not being able to change the multiplier was the only reason why Intel E chips were better than EB chips.
As we’ve stated before, we’re concerned about quality control issues on Abit boards. We’re concerned about the A7V133’s ability to handle Palominos.
What we’re most concerned about, though, is that you might think you have no choice but to buy these boards if you want to buy a high-end TBird and a KT133A and don’t want to futz with your CPU. That’s not so.
Yes, this will cost you a little more, but this also will give you the full range of KT133A boards to choose from.
You make the choice, just know that you have one.
Why Did It Work For Some People?
We saw a lot of indicators which pointed to the CPU initiating at default multiplier times FSB. However, there were some that worked fine.
What happened in those cases?
1) TBirds often can initially post at much higher speeds than that at which they can regularly operate. Since this problem first emerged, I’ve had a number of people report to me that they’ve been able to get POSTS at up to 1700Mhz with their TBirds. Chips couldn’t actually fully boot at anywhere near those speeds, but they could last long enough for the BIOS to take over.
Some chips may be able to do it sometimes and not others, that’s why I suspect some of the earlier tricks I suggested like cold boots appeared to “work” sometimes.
2) Being right for the wrong reason is better than being wrong for the right reason The AMD tech docs make reference to essentially how the Via reference design worked. I didn’t see a hint of an alternative, so I was fairly skeptical about the claims in the forums that there was no problem with motherboards that used a jumperless design (especially after reading about a few that did, but see below).
I didn’t think a jumperless design could obviate the default multiplier on initialization, and in fact, it doesn’t. I did think (to myself) that if there was something to this, it would have to be because the motherboard initiated with a default FSB of 100Mhz (which in fact turns out to be the case).
What I’m wondering about is how the fully jumpered mobos override the CPU initialization.
3) There are still other causes. A lousy pencil job will still leave you with a CPU that won’t do what you want when you change multipliers, even if you have one of these mobos. At least some users of the KT7A report some problems shifting multipliers (doesn’t seem to be a problem among A7V133 users).
So if this happens to you, that’s the first place for you to check.