133A/DDR Challenges

Flexibility

Socket A overclocking has been pretty simple up to now, get the lead out, do the pencil job, and increase the multiplier. There’s been little room for FSB overclocking.

133A/DDR board owners will now be faced with an embarrassment of overclocking choices; while still being able to increase the multiplier, you can also increase FSB, or do both.

This will give overclockers the flexibility to prevent gross disappointment in overclocking attempts and makes AMD chips an even more desirable O/Cing option than Intel’s.

With Intel chips, being able to only change the FSB often left the overclocker with a “Boom or Bust” situation.

If the processor didn’t jump the first FSB hurdle (going from 66Mhz to 100Mhz for Celerons; 100Mhz to 133Mhz for Coppermines), the overclocker was left with less than desriable fallback positions which required serious FSB overclocking (explained below). If the rest of his equipment could not handle major FSB overclocking, the overclocker often found himself capable of only minor overclocking.

With AMD chips, overclockers faced with that problem can either lower the multiplier and keep the FSB, or increase the multiplier and lower the FSB.

Don’t get too wound up on memory and FSB speeds

These are nice-to-haves, and you should certainly look for them, but they don’t make a lot of difference in the grand scheme of things. If somebody gets a memory score 10% faster than yours, that doesn’t mean that machine is 10% faster, probably doesn’t mean the machine is 1% faster for most things.

The reason for this is that the Athlon CPUs (and most current applications) are designed to fetch data from main memory as little as possible. They are designed to have the necessary data already parked in the L1 or L2 cache (which is much, much faster than any RAM) almost all the time.

Overclocking hard drives/PCI devices also improves performance, but again, not a lot for overall system performance.

So if your machine can only do 140Mhz FSB, and your (friend, rival, Quake-blooded enemy) can do 155Mhz; it’s not that big a deal.

Bottlenecks

FSB Overclocking

Your hard drives and other PCI devices are meant to run at a speed of 33.33Mhz. The motherboard takes the overall FSB it is running at, and divides the clock to get to that 33.33Mhz. At 100Mhz, the divisor is /3; at 133Mhz, the divisor is /4.

Should you run at an FSB speed higher than 100Mhz or 133Mhz, the speed at which your devices will run will also be higher than spec speed.

Examples:

If you run at 150Mhz FSB, your devices run at 150Mhz/4 = 37.5Mhz.

If you run at 160Mhz FSB, your devices run at 160Mhz/4 = 40.0Mhz.

Effectively, you are not only overclocking the processor, you are also overclocking the hard drives and PCI devices.

So What, What’s A Measly 20% Overclock? My processor overclocks over 50%!

It’s the difference between the machine working or not.

You cannot overclock hard drives and PCI devices as much as a CPU. Also keep in mind that it’s not like the CPU, where you have to just get one item to overclock; ALL your hard drives/PCI devices have to be able to do it. It’s the difference between rolling one seven at the casino, and four or five in a row.

Well, you have better odds than that, but I’ve observed that roughly half the people trying to overclock their FSB by 20% succeed.

Some devices are notorious for not overclocking much, or sometimes even at all. SCSI/RAID controllers and network cards are often bad at this. Hard drives can be awfully finicky, too, and hard drives that don’t like this have a bad habit of scrambling themselves. So backup anything you don’t want to lose before trying this.

Unless you have a particularly finicky device, you usually should be able to get a 10-12% overclock without problems, but conwider anything above that a bonus rather than an entitlement.

I’ll Just Increase That PCI Divisor To /5

No, you won’t. Those divisors are built into the chips themselves. If they aren’t there, you can’t do it. No software tricks have ever been found for this.

Memory

Now we’re going to find out just how good your RAM is. This will separate the men from the boys.

If all you have is PC100 RAM, if it’s high-quality, you’ll probably get to 133Mhz and a little beyond. Don’t expect 150Mhz or better.

If it’s generic PC100, you may have some real problems getting to 133Mhz. Many new Coppermine owners found themselves buying new RAM because their generic PC100 couldn’t cut it.

If you have high-quality PC133 RAM, and your other devices don’t object to the procedings, you’ll probably get to around 150Mhz at CAS2, maybe into 160s at CAS3. If you’ve bought very high quality, CAS2 at 150Mhz RAM, odds are you’re not going to do much better, due to the FSB overclocking issues described above.

If you have generic PC133 RAM, you might get lucky. A few of these pieces do as well as the high-quality stuff. Most don’t. Also don’t be too shocked if the generic RAM works fine for a while, then dies.

If you find yourself with RAM that doesn’t quite cut it, you’re in a better position than Intel overclockers. Take what the memory will give you, and adjust your multiplier/FSB accordingly.

I’m Not Telling You Not To Do It

The main reason why we often emphasize the problems and pitfalls is because nobody else does (or deemphasizes it). We don’t like seeing people crash-and-burn when they don’t reach their targets, so we try to set realistic expectations. If you end up doing better, great. We’re not telling you not do it, just telling you what may happen and why.

Ignorance is not bliss when it comes to these things. These bottlenecks don’t need your permission to happen. If someone steps off a skyscraper, even if he were somehow innocent of the notion of gravity, he’s still going to fall and die.

Realistically, we expect most people to hit around 145Mhz-150Mhz with their systems. A substantial minority should do better than that, but the successes should really drop off at 160Mhz or more.

Just don’t kill yourself and spend a lot of money replacing parts when you’re only going to get tiny improvements out of them.

Email Ed


Be the first to comment

Leave a Reply