The email keeps coming. Here’s an example:
“The Pentium boxed slot models, where Intel installed the heat sink and fan, seem to run cool and rock solid at 15% overclocked so the 850 should be able to do a gig no sweat.”
There’s two underlying assumptions here:
Sorry for the 2X4, but I’m getting a LOT of these. A lot of people are buying 800 and 850 chips, and if our database is any indicator, most aren’t reaching it. At least not with air-cooling, they aren’t.
Lots of people are doing it with Peltiers, but you gotta have a Peltier to get Peltier results.
There are two issues here:
- PCI Speed: All the devices you attack to the PCI slots (along with your hard drives) run at PCI speed. Normal PCI speed is 33Mhz.
This is much slower than the front side bus of the motherboard. So what the motherboard does is essentially divide the FSB speed by a number, and the result is the speed at which it expects the PCI devices/hard drive to operate.
Normally, the “PCI divisor” when the motherboard is running at 66Mhz is 2, at 100Mhz it is 3, and at 133Mhz, it is 4. (No current motherboard has a PCI divisor greater than 4).
A little division shows you that this keeps those PCI divisors/motherboards running at spec:
So if you take a processor that normally runs at 66Mhz, and overclock the FSB to 100Mhz, or take a processor that normally runs at 100Mhz, and run it at 133Mhz, you see that it makes no difference to those devices. They still run at the same speed. This is good.
It’s when you run them at speeds inbetween that you can have problems.
Let’s say to reach your goal, you need to run at 120Mhz. On many (not all) motherboards, you’ll still get a PCI devisor of /3. In that case:
That isn’t 33Mhz. In this case, not only are you overclocking the processor, you are also overclocking your PCI devices and hard drive.
So what’s a measly 20%?
Very often, the difference between your machine working and not.
Many devices have no problem being overclocked 20%. Many do. SCSI/IDE RAID controllers often are finicky. Network cards often are finicky. Hard drives often are finicky (and sometimes trash themselves when you try).
Running your PCI devices/hard drives faster than they can tolerate will stop you dead in your overclocking tracks just as fast and finally as inadequate cooling or a CPU that’s not just up to it.
Unless you know from personal experience beforehand that all your devices will handle overclocking (and no, somebody else having done it means nothing), your initial goals should not take it as a given that they will overclock successfully.
So what should I do?
You should plan your CPU/mobo purchase around the PCI speeds you are likely to require. The best and safest way to do this is to buy equipment that will let you reach your initial goal with an overclocking jump that leaves PCI speed alone (i.e., 66 to 100Mhz or 100 to 133Mhz). If you can coax a bit more out of your machine than that after you’ve reached your
basic goal, great, consider that gravy.
Unfortunately, 1Ghz is still a pretty iffy target for any PIII. Some do, some don’t. So buy a PIII 750 and expecting it to hit 1Ghz is fine from the PCI standpoint, but the processor might not be able to do it.
You need fallbacks.
You should buy a motherboard that lets you change the PCI divisor at lower speeds than the “standard.” In the case of the Celeron, an option of a PCI divisor of /3 at 90Mhz is good. For a PIII, an option of a PCI divisor of /4 at 124Mhz is the minimum you should accept, and a bit lower than that would be even better.
The ideal is a stable motherboard that lets you dial in any FSB and lets you set any divisor you like at those speeds, but you can have too much of a good thing, too.
If you have a PCI divisor of /4 at 124Mhz, what you are doing is slightly underclocking the PCI devices/hard drives; 31Mhz rather than 33Mhz. You don’t want to stray too much under the norm, either, some devices don’t work when they get underclocked too much, either. Using a PCI divisor of /4 at FSB speeds of less than 120Mhz probably isn’t a good idea.
If you’re buying a new system, and you buy a motherboard that gives PCI divisor flexibility, you can either buy a 700E, assume you’ll get 933Mhz, and hope your devices tolerate enough of an overclock to reach 1Ghz, or you can buy a 750E, accept 930Mhz as the minimum at 124Mhz, and hope the CPU can run at 1Ghz at 133Mhz.
- Buying a faster chip doesn’t make it more likely to reach 1Ghz: In fact, due to the PCI issue, it makes it somewhat less likely.
All these chips get made together. Usually, the manufacturing process is good enough that a big majority (like 80% or better) of chips will go as fast as the fastest chip. That’s not quite the case with Intel right now; it’s probably more like 50/50 with the cC0 chips. I’d say 80% or more will get reasonably close to 1Ghz; but a lot won’t quite make it. (Getting past 1.05Ghz with aircooling looks quite unlikely, though.)
It’s not like Intel (or AMD) tests these things speed grade by speed grade. Right now, since both are pushing the limits a bit with their fastest chips, they probably test enough chips at the highest speed to get their quota for those chips, then test the rest at a somewhat lower speed, and if they pass that, grade them based on what the market demands.
Understand that the less testing the CPU manufacturers have to do, the better. You only have to test more when your yields aren’t too hot. That’s a real pain in a highly automated environment; you need a lot more testing equipment. It’s far better to get high yields and downgrade chips than to have to test and test just to meet quotas.
So, (with the exception of a few high-grade chips, which isn’t the case here), buying a more expensive processor doesn’t get you a better processor. That’s certainly the case with an 800 or 850E.
More negative vibes! You make Kurt Cobain look like a bundle of optimism! I prefer to think positively!
In a sentence, I see the failures. The lousiest part of being involved with this website is getting emails from people who thought positively, and still failed. They failed, not because they didn’t think positively enough, but
because they thought too positively and expected too much from what they bought. And it doesn’t work the way they wanted it to. And they come looking for a magic fix. And I don’t have one. And since I can’t BS them and look at myself
in the mirror in the morning; I have to tell them that it isn’t going to happen.
You have to do that ten times a week, and you’d do what you could to minimize this, too.
Gamble For The Gravy, Not The Groceries
To be a good gambler, you have to at least first realize that what you’re doing is gambling. Then you need to realize that when you gamble, not only can you win, but you can lose, too. If you realize you are gambling when it comes to this, and can take losing in stride, no problem. But many don’t.
It’s better to set realistic goals, meet them, then gamble for the gravy. If you miss, you still met your initial goal. Don’t gamble for the groceries. You miss that, you’re left with only disappointment.
Personally, I’ve never been blessed with a wonder chip, and I’ve probably overclocked more CPUs than most of you. What I’ve done is look at the stats, figured out the odds and decided whether or not the close-to-sure thing was good enough. I went for that (so far have gotten it every time), then tried for more
(and have had a pretty lousy luck of the draw so far). But that’s OK because I got most of the improvement and was satisfied with that.
Works for me, at least.