I’m seeing a lot of the following going around:
People have a KT333, or maybe a select few KT400 motherboards.
They buy an AMD processor with a default multiplier of 13X or more.
Right now, the conventional wisdom is to lower the multiplier and increase the FSB by as much as possible.
They eventually find out that if they want a multiplier of 12.5X or less, they’re going to have to either unlock the L3 bridge on the CPU or do a little wire trick.
Some start wailing and gnashing their teeth at this point, because they don’t want to do anything that might leave evidence of overclocking, or because it’s too hard to do.
Others do what they have to do to unlock the lower multipliers, then find out they can’t reach the desired speed at 12.5X, want to go back to 13X or more, and find out they’ll have to undo their work to get that.
Then they start wailing and gnashing their teeth.
As stated above, the conventional wisdom on AthlonXPs has been “Lower the multiplier, increase the FSB.”
Up to now, this has been a good thing to do. For instance, if you have a processor capable of 1800MHz, 10 X 180MHz gives you more performance than 13.5 X 133MHz, mostly because it widens the FSB and allows for more memory throughput.
However, this only works all the time when there is no problem increasing the FSB to the desired level. There comes a point where you just can’t do that. Nobody runs an Athlon XP at 6 X 300MHz simply because you can’t reach 300MHz FSB with any current motherboard.
When FSB becomes a bottleneck, as processors get faster and faster, you get less of an advantage from lowering the multiplier, simply because you can’t increase FSB all that much.
If, for instance, you have a processor running at 13 X 190MHz, and the highest FSB your equipment can stand is 200MHz FSB; the most you’ll be able to do by lowering the multiplier is 12.5 X 200.
If your processor can do 2470 MHz but not 2500 MHz, the most you’ll be able to do is 12.5 X 197 MHz, and you’ll actually lose a few total MHz in the process. Nor is there much performance gain between 190 and 197/200, certainly a lot less than by going from 133 to 180.
The Mobo Bottlenecks
Both the KT333 and KT400 have a maximum PCI divisor of /5. That means you start overclocking your PCI/AGP devices, and in particular your hard drive when you go past 166MHz. Hard drives in particular can’t be overclocked like CPUs.
Unless you’re lucky, that means you probably can’t expect 200MHz FSB or more from those motherboards. Some will. At best a lot, at worst, most, won’t.
What’s happening with Via boards right now is the following:
People are either:
Or . . .
Remember, there is no PCI/AGP lock on these motherboards like there is for the nForce/nForce2 boards.
Owners of nForce2 boards shouldn’t think the sky is the limit, either. Having a PCI/AGP lock frees you from the PCI bottleneck, but increase the FSB much beyond 200MHz, and eventually you’ll run into another bottleneck: memory. You’ll no doubt be able to run a higher FSB speed than you could with a Via board, but you just run into different wall sooner or later.
The Lazy Man’s Approach To Overclocking
A lot of people reading this will look at the steps mentioned above and say, “So what? Big deal.” This article is not meant for you, yet.
However, for many people, this is a big deal. They don’t want to do all that work, then find that they have to undo it.
Call such people lazy if you like, but with TBredBs becoming available, automatically unlocking no longer gets you a nice guaranteed performance boost.
So if going through the experiences mentioned above sends shivers down your spine, we suggest you do the following:
First, don’t do anything to the processor at first. Just install, test, and see how many MHz you can get using the higher multipliers.
Once you’ve done that, then decide if you want to go with a lower multiplier. The higher you can go with a “high” multiplier, the less likely and more risky going to a lower multiplier will do you any good when you have a Via mobo.
As a very simple, rough guide for people with a TBredB and a current Via board:
If you can get 2300MHz with, say a 13X multiplier out of a Via board and you really dread the thought of doing and maybe undoing those things, leave well enough alone. You’ll lose a bit, but probably not more than a few percentage points.
If you can get 2400MHz with a 13X multiplier out of a Via board, and you don’t like the thought of doing and maybe undoing these things, leave well enough alone. You’ll lose even less than person number one.
If you can get 2500MHz with a 13X multiplier out of a Via board, you probably should leave well enough alone even if you like the thought of doing and undoing these things.
In all three cases, you then need to ask yourself the following:
“Is gaining a few percentage points in performance, maybe, worth unlocking the CPU, hoping my hard drives and other equipment are up to a good deal of overclocking, and having to undo the work (and maybe end up with a scrambled hard drive) if things don’t work out?”
The higher the MHz, the riskier it gets for you to get any appreciable performance increase out of unlocking, simply because you’ll have to push your FSB over 200MHz to get it, and past that, the prospect of hard drives scrambling and other PCI devices refusing to work becomes quite real.
If the answer is “No,” (and if you’re getting 2400MHz or more; the more likely it is the answer should be “No,” then settle for what you get, or get an nForce2 board.
To The More Committed
As I write this, I know that the hardcore overclockers are expressing sentiments along the lines of “How gay” (if not quite as kindly).
The reality is most overclockers aren’t hardcore. If anything, the trend is towards more people overclocking, with more of them willing to take less and less risk.
However, if Athlons get much faster, it will make less and less sense for even the hardcore folks to automatically go the lower Athlon multiplier range.
If Athlons get to 2600 or 2700MHz, unless you’re downright lucky with the other equipment you have, reducing the multiplier to 12.5 or less isn’t going to do you much if any good, either with any mobo that has a maximum PCI divisor of /5.
Even if you have an nForce2 board, memory bottlenecks will keep you from doing, say, 10 X 260MHz.
Mobos and memory will prove to be the bottlenecks for Athlon overclocking in 2003.
Keep in mind though, that if getting the last 2% or so of overall performance out of your machine isn’t worth much effort, that’s the level of payback we’re talking about.
It’s your machine and effort, you decide.