Introduction
Quite a lot of time has passed since my last update of the Overclocking Guide and now some really important things happened, which urged me to write something about this topic.
Although I never included it in my Overclocking Guide, the Pentium II is a wonderful CPU to overclock. All PII 266 CPUs that I came across worked just fine at 300 MHz and as long as you cooled the chip well enough, they even run at 337.5 MHZ, (4.5×75 MHz).
A simular thing used to be true for the Pentium MMX CPUs, even Pentium MMX 166 CPUs ran just fine at even 233 MHz after simply changing the multiplier.
The Time Has Changed
For a several weeks now it looks as if Intel has an excellent yield in the production of the Pentium MMX as well as of the Pentium II. This means that probably 90% of the produced chips are of top range spec, meaning that Intel almost only produces Pentium MMX 233 and Pentium II 300 CPUs. However, since a lot of people are still asking for Pentium MMX 166 and 200 CPUs as well as PII 233 and 266 parts, Intel has to mark a lot of faster parts slower than they are in reality. Now here the overclocking issue comes in. Intel is aware of the fact that people know about overclocking and they want to get top money for top performance. I never thought of having a reason to buy a PII 300, because the PII 266 was just as good. Lots of you advised friends to buy a Pentium MMX 166 instead of a 233, because it ran the same and indeed it was. Intel doesn’t want this, as everyone can imagine. Hence they did something that they’ve done once before in the past. I’m sure that lots of you still remember the Pentium 133 SY022 CPUs, which wouldn’t recognize any multiplier setting higher than x2. Intel disabled the higher multipliers by just not bonding one of the multiplier pins. That is the same they are doing now. Please understand that this is some kind of touchy thing when chip production starts. You can only test a CPU after it’s packaged, which means after bonding. As long as you are not producing many fast chips like e.g. a Pentium MMX 233, you are taking the risk of wasting some high end parts by bonding them to only 166 or 200 MHz versions. That’s why Intel never has this ‘overclock protection’ in newly released chips. However, when the yield of high end chips is really high, even higher than what the market wants, you can afford castrating fast parts to slow ones by just bonding it.
The result of this is that Pentium MMX 166, Pentium MMX 200, Pentium II 233 and Pentium II 266 CPUs cannot be overclocked via higher multiplier settings anymore. This seems to be the case for the last 4 to 6 weeks, which means that there are hardly any ‘overclockable’ Intel CPUs on any of the shop shelves anymore. Only a shop that’s hardly selling any and that’s got a CPU lying around for a long time might have one of the ‘gems’ left.
This restriction of the multiplier settings does of course not affect the overclocking via higher bus speeds. You can still run a PII 266 at 300 MHz, but it’ll have to be 75 MHz x 4, because 66 MHz x 4.5 won’t work anymore. You can even push it to 333 MHz, but this means using 83 MHz bus speed, which runs only in very few boards with very few RAM types and hardly ever with AGP graphic cards. Forget about 100 MHz in 440FX or 440 LX boards. I’ve tried it with the extremely stable LX6 from Abit and with the best memory I could get. 83 MHz is fine in the LX6, but forget the 100 MHz it offers. The same is valid for the latest versions of the Asus LX boards that also offer you 100 MHz bus speed — it just doesn’t work. 75 MHz are fine with Pentium II CPUs, 83 and more should rather be left alone, because the system gets too unstable and AGP graphic cards don’t like an AGP bus running at 83 MHz instead of 66. 100 Mhz should be left to the upcoming 440BX boards and the ‘Deschutes’ Pentium II CPUs that are designed for 100 MHz bus speed.
Socket 7 CPUs at 83 are easier to handle, so there’s still a chance of running a Pentium MMX166 CPU at up to 208 MHz and a PPMT 200 at up to 250 MHz, c’t Magazine may scream and shout, but ten thousands of people that run 83 MHz reliably and stably for months can’t just be wiped away and it even works in Germany, not only outside this wonderful country.
One last word to the castrated Pentium II 233 and 266 CPUs. This is the first Intel CPU which is ‘assembled’. This brings the thought near that the disabling of the higher multiplier was not done inside the actual chip package, but somewhere inside the cartridge. It could be possible that this castration can be ‘reversed’ after openng the cartridge and finding the right spot. If anyone of you is willing to try … let me know and I’ll publish it. It’s just a damn problem to open that %^&*$#@ cartridge …
And What about the Others … ?
Well, what about the AMD K6 and the Cyrix/IBM 6x86MX ? If you let me start with the 6x86MX, I have to tell you that especially Cyrix is selling their 6x86MX chips pretty close to the margins, as well known from the past already. They came up with the — with all due respect — crazy idea of selling parts with different specs under the same name. You can get PR166 CPUs at 2×66 and 2.5×60 MHz, PR200 CPUs at 2×75 or 2.5×66 MHz and PR233 CPUs at 2.5×75 or 2×83 MHz. These CPUs with the same P-rating are actually not performing equally, but who cares? Now I came across cases where you couldn’t even clock a PR200 CPU at 2×75 MHz to 2.5 x 66 MHz, which is slightly faster. Hence forget about overclocking unless you got a nice PR233 at 2.5 x 75 MHz, since then you might have a chance. IBM is unfortunately under the threat of conformity with Cyrix and rumors say that IBM is selling their chips the same way against their own believe … well, well.
This leaves AMD’s K6 as the only real overclocking champ and as long as your K6 is real, you’ve got a good chance to overclock which ever way you like, either with multiplier increase or with higher bus speeds or with both. However there’s a little problem. AMD doesn’t seem to care much about the fact that remarking their chips can be done by a six year old child. There’s only some paint printed on the aluminium heat sink attachment, which you can wipe off with several different chemicals. Printing a new marking on the ‘naked’ CPU is not too hard to do. This makes a K6 233 out of a K6 200 or even K6 166. Be careful with that, examine the printing very carefully. If it’s done well, you won’t have a chance though. The incidence of remarked K6 CPUs is cerainly higher than you think, and most reports about ‘unoverclockable’ K6 CPUs are due to remarked 166 version, marked to a K6 200.
