Slashdot Mirror
Overclocking Pentium IIs
Dr. Damage writes "Ars Technica has just published an article entitled Clocking and locking the Pentium II. The article looks at exactly how PCs generate and control CPU clock speeds and how Intel locked the bus multiplier in the PII. It's also got some tech explanations and speculation about the possibilities for bus locking the PII/PIII architecture to prevent overclocking. "
How about overclocking a K7 ? :))
-kojak
I'm sick of this overclocking business.
If you want to overclock and can't find the info yourself, don't do it.
fix it please
You know, there is a reason they tell you not to overclock chips and put speed locks on them (and it's not just the fact that it voids your warranty). Anyone who has ever studied even rudimentery digital electronics will know that every circuit of this complexity (a CPU) has to have a certain speed determined at the time the CPU is made. Intel actually already pushes their chips to max safe speed so that they can make the most money.
I could go on and on about all the technical problems with overclocking (data corruption, data loss, component damage, etc), but I figure if you didn't listen to the above you probably won't listen to more technical reasons. Maybe you will listen to this: if you overclock your CPU, EVERY component in your computer no longer has a legal warranty. They have ways to find out that you overclocked your CPU, and if you overclock it even once, then you have just fucked yourself legally.
For example, let's say you overclock your computer for an hour or two to see if you can get Windows 98 to go any faster. Then you find out that Windows 98 sucks at any speed, so you set it back to the speed that it's rated for. Then your video card breaks. Not even from heat or anything, let's say it was broken from the factory. Sorry, you have to pay for a new one, because the warranty is void, because you overclocked it. You would deserve it too, for being such an idiot.
Already been done. You are an idiot for overclocking your chip, and now that you have, the warrenty on every component in your computer is void. Have a nice day.
Intel should just put the clock speed that the CPU is supposed to run embedded in the silicium so that people have a choice wether to overclock or not (try: "cat /proc/cpuinfo" to see what I mean). And make it easy for customers to detect wether or not they have been ripped off by the reseller!
But they won't do that because then the can't overclock them theirselves (like they have been doing with pentiums from day 1).
Here is how it wors: they produce a batch of Pentiums then test them at higher speeds than they have been designed for. If they run OK they are then sold at a slightly slower speed (safety-margin) instead of selling them at the speed they where designed for! the crooked resellers just narrow the safety-margin.
I sure hope AMD will not follow this stupid overclock-protection sceme (or is it scam?), but I guess that would be wishfull thinking!
Why don't they just but the crystal on the module?
1. You have no idea how FAB works. These CPUs are coming off the same line. a 333 and a 350 are the SAME DAMN chip, from different parts of the wafer. Speed ratings are statistically sorted, not test.
2. You are wrong: they cannot detect OC burn out. If you actually knew ANYTHING abtou electromigration, you'd know that one of the biggest problems in catching overclockers is that EM problems cannot be detected without serious investigation. Like Intel is going to spend 5x the cost of one CPU to see if you OC'd and fried yours.
3. Your video card manufacturer has no IDEA if you OC'd your processor. You must be a dolt. How do they know? Would you like to provide technical details?
"Intel actually already pushes their chips to max safe speed so that they can make the most money. "
Uhhh, no. Actually, there are many documented occasions on which Intel has UNDERMARKED CPUs because the demand for the low end chips exceeded stock.
Oh, and BTW, please enlighten us with this supposed method for determining whether components have lived in an overclocked system. Go ahead, I dare you to show me that you're not full of shit.
Apparently, you must know everything.
They build in a safety margin. There's a letter from some intel employee somewhere on http://www.sysopt.com, and he says that they test the chip at a much higher speed, and then rate it for a lower speed. Kind of a safety margin.
I have a P120 that runs at 208Mhz with a peltier cooler. It has been that way, running linux for about 2 years, with no reboot. I also have an amd 5x86 133 running at 160 for about 3 years. And neither one has had any problems at all. My friend also has his PP200 running at 266 with a big heatsink on it. It's worked perfectly for over 6 months.
Overclocking doesn't fuck shit up if you do it right. You seem to be using basic microsoft style FUD to steer people away from overclocking. Do you work for Intel by any chance?
And there is no way in hell they can find out you overclocked your cpu, unless you are a fucking idiot and you tell them you did.
What a moron.
Does anyone know how to bypass the multiplier lock? It would be nice if someone could get ahold
of the schematics that those guys in Germany used to do it.
Oh, stuff it. Your post is wrong from one end to another.
1. Intel makes the best chips they can. If the market needs more P2-350's then that is what the chips will be marked. Even if the silicon tested ok at 450.
2. Overclocking the _CPU_ itself (multiplier) has no effect the warranties of other components. Changing your system bus speed could affect other components, but it's not likely to affect your warranty.
3. "They have ways to find out that you overclocked". If you are stupid enough to melt something, perhaps. Hint: Overclocking does not magically change the components of your system.
4. "Your video card is broken from the factory, but you have no warranty because you overclocked once" Riiiight.
Next time, please post on a subject that you know something about. Save the FUD for your friends and family.
Oh, and BTW everybody, if you get any mail with the subject heading of "GOOD TIMES", DON'T OPEN IT! It will DELETE YOUR HARD DRIVE!
Whew, that was a close one.
You are such a fucking moron. You have no clue what the hell you are talking about. Your one of those pussies who drive 45 in a 55 MPH zone because it's "safer", and you wouldn't wanna miss a day of work at your beloved Intel.
Actually, given the FUD tactics in your letter, you should put your name and address at the top, call it a resume and send it off to Microsoft. Of course, that might mean you would have to take a plane for your interview, and we all know how dangerous that is.
Wellllllll It very much depends on how you overclock. For example, my K6-2 300 runs a 3x multiplier on a 100mhz FSB, if i increase the multiplier to 3.5x (which i HAVE its nice and stable and the cpu is cool to the touch even after hours of gaming) then my bus speed (PCI bus speed too) remain the same, hence no damage to my cards, and not 1 of my cards has been OC'd. Now on the other hand, my K6-200 is running on a 66mhz FSB. to overclock it, i would have to increase the bus speed to 75 or 83, both of which cause unhealthy bus speeds for your PCI devices (41.5mhz for 83 and 37mhz for 75) and since your pci devices ar ement to run at 33mhz well that could cause problems (read: why i did not do it). There is one more route i could have taken on my K6-2 was to drop the FSB to 90mhz and take the multiplier to 4x, but that proved no faster than 3.5x100.
Just my $0.02
What I really want to know is how to get the damned retail boxed processor heatsink off without munging the processor. I'd have no problem overclocking if I could slap on a decent aftermarket heatsink.
because I'd replace the crystal. It's a damn sight easier than circumventing the multiplier lock.
I'm sick of you idiots complaning.
If you don't like a subject, don't read it (there is nobody twisting your arm now is there?)
Go insult your momma!
>Do any of these semi-literate buffoons actually read the articles before they start whingeing about them?
Yes they actually do.
>This is a good, solid, well-written technical article that explains such things.
1 I wasn't talking 'bout the article
2 No it doesn't it merely excuses them.
>Anybody who knows enough about chip fab to make this type of judgement would have seen most of these issues addressed in the article.
I've seen them and still I think the points are valid, as far as I'm concerned the article just confirms them. If Intel can o/c their own CPU's why shouldn't we ourselves?
But I suppose you are just to lame to discuss things with arguments instead of insults
Uhm... With a 100Mhz FSB, PCI cards run at 33Mhz. With a 66 Mhz FSB, PCI cards run at 33Mhz.
Sorry, but a lot of chips WILL run faster than the label says they will. Chip makers tweak the process to yield the fastest chips, once they have filled the quota for fast chips, all the rest get marked with the next slowest speed until all the speed grade quotas are filled. A 300mhz chip may not have been tested at final test for 450, but that doesnt mean that it wont pass the 450 testing. We did it all the time at Moto, I'm sure Intel and AMD do the same thing.
Intel could clock lock their chips if they really wanted. All it would take is a second PLL locking on the output of the multiplier and using a saw filter in the feedback loop. Saw filters are compact and very selective in the mid to high MHz range.
Thermal variation? Bullshit. That problem was solved nearly a hundred years ago. Put a small resistor under the filter to elevate its temp so that it lies near the middle of the chips operating temp. Build a small delay into the bootup sequence to give the filter temp time to settle.
Why doesn't Intel do it? 1)The last thing big companies want is to take product back. This is why they 'test to the max and then back down'. Any engineer with sense does this (Remember Scotty instructing LaForge on how to report repair time estimates to captains?), and Intel is a company of engineers. Always leave yourself wiggle room. It is not economically feasible to test every chip in a high production environment, so statistics are used. If you want to stay in business, don't ship a whole batch at the max speed of the tested chips.
2) Marketing. No one gets mad if their chip will run faster than what it is marked, and the chip will always be stable. But all hell breaks loose if the chip gets quirky at its rated speed and Intel gets a bad name (can anyone say, "Cyrix"). But if you've truly locked the chips, you can't just scratch off 333 and put 300. Hence the off chip solution shown in the article. Go ahead and make the chip, but we'll lock the speed in later.
BTW, just to be a stickler, the Pentium is more appropriately termed an IC (integrated circuit). A chip is a single element device such as a resistor or diode.
I just replaced my overclocked AMD 200. (OC to 233 oh boy what a difference) I replaced it with a 350 K6. I guess I could clock this guy up to 400 but why? It would be about as silly as OCing a 200 to 233 :).
As the MHz get higher the performance difference get's smaller. The 300a Cel's running at 450 are likely the peak of the OC advantage. It's a 50% increase in clock speed and a 50% increase in bus speed.
Not to bad but I have to think with out extreme measures taken to cool the CPU we won't be sqeezing significant performance out of OC'ed chips in the not so distant future.
Maybe this time next year a basic chip will be running 500+ MHz. A 50% increase in clock speed would require a 250 MHz jump! Ow. Get out the eggs dear. And how long will it be till the bus takes another 50% speed hike to 150 MHz?
LINSEX SUX!
supposidly, traces between core and L2 cache (even at 2-3cm lengths) begin to transmit the "funky monkey" at the high speeds (>300-MHz)
Of course, that is pure rumor at this point...
Ok I saw a couple of posts from people disbelieving that chips WILL melt down. I, yes I, have ruined a chip from excessive heat. It was a Cyrix system, I believe 5x86, running at the spec speed (not oc'd). However the on chip fan failed and I didn't know it so before long the system started crashing randomly all the time. Took me a little while to realize it and the chip never got back to normal stability.
I tossed it.
So the moral is yes, a chip can fail from too much heat. I have no experience with oc'ing though, so if you keep it nice and cool maybe it's not an issue at all, I don't rightly know.
I didn't want to create an account so I'm posting anonymously...
I was aware that it was more feasible to use a divider than a multiplier. However, I thought the explanation with the multiplier was a bit easier for non-EEs to comprehend, so I went that route, even though it was oversimplified and not as accurate. I didn't want anyone's eyes to glaze over reading a lengthy PLL explanation, because I didn't feel that the actual details of the PLL involved were as important to understanding what's going on. In fact, the actual details of the PLL aren't even publicly released as far as I know. But you pretty much know how it works, and the explanation I gave wasn't the best one.
This was an error in judgement on my part. After looking at the reader feedback, it seems that the divider issue is more important to the article than I thought it was. So I regret glossing over it now (my Inbox regrets it too), and I wish I had gone with the slightly more complicated, yet more accurate explanation.
The problem with an article like this is that you're writing to people with vastly different levels of experience. I usually target my articles at enthusiasts with at least some programming experience. So sometimes I err on the side of oversimplifying. But my usual strategy when doing an article like this, and it's worked very well so far, is to oversimplify in the first article, and then do a followup or two where I go more in-depth by answering all more technically savvy readers who called me on the oversimplifications.
So look for a followup where I answer some reader feedback on the PLL thing, and take care of a few other loose ends.
--Hellazon
um, despite the fact you seem to have become a ;-)
really popular guy with this posting, i still have
to add my comment. It seems totally stupid that
you can destroy other items in your PC by over
clocking - so lets say i have been running my 300
for a few months and i decide i want it up to a
333.... so i over clock it. Great. But alas, how
the hell does it break other components. If i was
to go buy a 333 it wouldnt break other components
so why should just o/c'ing a chip up to that speed
break components? (i get this funny feeling there
is somthing i am missing, but someone will slag me
off if i have
Get it right.
I own an old 9.54 MHz 8088 (by Siemens). I do believe Intel produced three 8088s - at 4.77, 7.15 and 9.54.
Thats an interesting point.... What if people actually need to underclock, but cant because the chip is locked. I can think of various situations where this may apply, such as building dedicated mp3 servers. You might want a low noise (sound decibel level) system with no fans. What you need is then a chip that runs cool and only needs a heatsink.
The other reason may to be to cut down on power use (car systems, etc...)
Call me crazy, but I'm hoping that you still can underclock if Intel gets sour & locks the chip.
Not that I'm not one for overclocking..heheh
I'm building a cryo-cooler on the side to see how far I can run a P166 to on a 83? Mhz bus...
Posted by Mr. Assembly:
The mips line of processors is going to die! SGI is going to move to intel, so your going to be pissing on your own computer anyway. Just be sure to have the paper towels handy to cleanup.
Posted by Mr. Assembly:
By the time that the components burn out if they do you're going to be tossing everything for an uprgade anyway. After three months your computer has devalued to nothing to even give this a passing thought. You know, I bet your the kind of guy that does'nt rip the tags off your pillows either, huh? I bet I'm right.
Posted by Tax_cheat_and_d_FBI_is_after_me:
I hate both INtel and Microsoft
their both make shit stuff,
G3 or any other RISC chips are way advance,
go 64bit UNIX !
"Intel has a pretty sane way of providing the consumer some protection against buying remarked, overclocked Xeons. Each Xeon has a processor information ROM (PIROM) built into the processor substrate. This PIROM is electrically programmed at the factory to hold details about the chip's core type, core stepping, etc. There are 16 bits at offset 16h in the PIROM that store the processors maximum core frequency (in MHz). So if you want to know what the speed your Xeon was rated at, just check that number.
It seems to me that if Intel is as serious about stopping illegal overclocking as they are about robbing every last shred of privacy from their customers, then they'd just do something like this
with the PIII, instead of that ridiculous CPU serial number"
--
Xenu loves you!
>Just let everyone burn up their chips trying to o/c above spec
I've heard of plenty of people trying to overclock chips, and never heard *anyone* say they burned out a chip doing so.
Do you just hate the thought that somebody out there may be getting the same performance as you for less?
Ooh, a sarcasm detector. Oh, that's a real useful invention.
>Intel actually already pushes their chips to max safe speed so that they can make the most money.
Just to explain a little economics, this isn't how things work.
Suppose Intel made one chip at one speed; how would they price it? Price it high, and only richer people buy it. Price it low, and those who would pay premium prices will pay the low price. It's much better to have more than one product, so you can sell the better product at the higher price and the worse one at a lower price -- you maximize your earnings. And if you only have one real product (how does a 350 mhz Pentium II made to 0.25 micron process differ from a 450 made on that same process?), simply sell some as a lesser product.
This is by no means unique to Intel or even the computer hardware market. In software, Autodesk sells AutoCAD for ~$5000, AutoCAD Lt for ~$350. Does it cost Autodesk more to produce and support AutoCAD? Not really, not significantly; in fact their production costs would probably be lower with a single product. But there's no way they could sell AutoCAD for $5000 to some people and $350 to others (save things like academic sales.)
Nor for that matter is this unique to the computer industry. You'll find clothing companies have different lines, with trivial differences, but one has the marquee name and the other doesn't. Ditto shampoos, soaps, TVs, you name it.
Ooh, a sarcasm detector. Oh, that's a real useful invention.
Real men use MIPS. ;>
I digress, and agree.
If you're gonna use that Windoze shit, you need all the speed you can get. Sorry buddy.
"PIIs are already run hotter than any chip out there. Hell. I'll put money on the fact that in
a couple of years, people with PIIs will:
1. have to buy new motherboards to support the PIII or P4.
2. have to junk their system cause the chip is fried out. "
1. I read somewhere that the core voltage requirement of the PIII (really just a PII with KNI) is such that a PII mobo won't support it.
2. Hasn't this been the standard in the Intel world for years?
"I could put a rocket engine on a Yugo and it would go faster than any other car out there, but it'd still be a
p.o.s."
You can't polish a turd.
These are my friends, See how they glisten. See this one shine, how he smiles in the light.
" Intel actually already pushes their chips to max safe speed so that they can make the most money"
This is a false assumption.
Intel actually takes chips rated for 450, and marks them as 300, because they need to justify a lower price-point to compete with AMD on the low-end. That was the whole point of the Celeron line, if you recall the original Celeron, it had NO cache, and sucked. The new Celeron has on-die cache, and is no different from the PII you would spend $300 on.
If a smart enduser can figure out how to squeeze his cut-rate 300 to 450, and screw intel in the process, then more power to them.
These are my friends, See how they glisten. See this one shine, how he smiles in the light.
Um didn't SGI (PackardBell wannabes) just sell off their interest in MIPS?
Didn't SGI (Gateway wannabes) just introduce a new line of Intel-NT based workstations?
bub-bye IRIX.
bub-bye MIPS.
These are my friends, See how they glisten. See this one shine, how he smiles in the light.
The article explains why it's unfeasible for the CPU to have a frequency detector on-board. Re-read it.
--
Kevin Doherty
kdoherty+slashdot@jurai.net
Kevin Doherty
kdoherty+slashdot@jurai.net
Poor guy probably hasn't gotten his chip to OC.
Vermifax
Logout
to do this, i have opened the black plastic box, remove the PCB of my p2, then i drilled the 4 rivets of the metal plate... i then replace some thermal paste, reput PCB etc, and i put my big heatsink and fan on it!
also i never reput the black plastic box, my p2 is cooler without it... the retention is done by the heatsink
--
"Science will win because it works." - Stephen Hawking
Why couldn't Intel just create a pointer or CPU Hardware marker that when a processor is overclocked it will be marked (hardware) that it has been pushed passed it's spec. THat way when it was returned for warranty it would be checked and refused or vice-versa. If we keep burning them out we will also keep buying them. I currently have a Celeron 300A runing very nicely at 450Mhz@2.0V and have not even had a burp..
That darn Slashdot is so cool... Hey did you pay the phone *(#(Q%$#$ NO CARRIER
Why am I an Idiot? I got a Pentium II 450Mhz computer for around $400 less than buying the legit 450 PII. And by the way how did I void the warranty of the components in my computer...how the hell is the manufacturer of the rest of my stuuf going to know I overclocked it. Think about what you say before opening your mouth....
That darn Slashdot is so cool... Hey did you pay the phone *(#(Q%$#$ NO CARRIER
Some of this is true. At some point, Intel will mark chips at their maximum "tested" speed.
:)
But think about it.. after Intel gets to 0.25 micron, and they're making Xeons and PIIIs... do you REALLY think that they're still having trouble getting Celerons to be stable at 300?
Part of the reason they make different speeds at different prices is just that - they need multiple price points. It's marketing. Far enough into production of a chip, odds are most EVERYTHING runs at top speed, but they mark it "300" because they need something they can sell for under $100.
I'm also curious to know how someone will find out I overclocked a machine... sheesh.
So do you like working for Intel? How long have you been there?
The PLL is not complete. Yes, there can be a multipler that is on the output of the vco. Generally, there is ALSO a multiplier/divider on the feedback loop leading from the VCO to the comparator. Setting this to some frequency division, and you have an output clock that is higher than the input clock.
(yes, that is right. You divide to increase the clock).
Clock dividers are a lot easier to design, so I wonder why this article used a clock multiplier.
It takes a real man to underclock them.
I've got a K6-2 350Mhz running at 300Mhz.
Call me crazy.
My studio - www.graylands.ca
Well, actually, I believe they did this so that they could drive the TV output of the same clock crystal as the processor . . .
Any sufficiently advanced technology is indistinguishable from a rigged demo
--Andy Finkel (J. Klass?)
My warantees, my hardware, my problem, not yours.
'nuff said
0 1 - just my two bits
Thank you for educating me FUDmeister.
I guess I should buy a Packard Bell and never, ever, break that little hologram sticker on the case. It voids your warranty, you know. And I certainly wouldn't want to do something that the manufacturer didn't intend.
Gasp! I've heard that some really bad people out there actually RECOMPILE their operating systems. Those idiots! Don't they know that if they make a mistake, their computers will blow up and they won't be able to use chatrooms any more??
-- What you do today will cost you a day of your life.
(The watercooling is ugly; it involves plastic sandwich bags filled with water and closed off with a rubbish bag tie (you know, the little metal wires embedded in papaer). I used several layers of saran wrap to protect the thin water filled bags against the sharp edges of the electronic componentry on the option adapters above the CPU. This should give you an idea of just how old these option adapters are; they still have soldered pins sticking through, unlike newer (i.e. almost anything made since 1988) PC boards where the pins don't reach through the whole plastic card.)
One disadvantage is that my IBM Portable PS/2 is no longer portable at all. It is rather a quivering mass of digital protoplasm on a lab shelf.
Cheers,
Joshua (who is looking for an 80 megahertz oscillator. Gee, if I can ``underclock'' 33mHz, why not 40mHz?)
--jon. Postel is dead. May we all mourn his, and our, loss.
*eep!* argh, you dirty mouthed people actually got me post! I'm so embarressed.
What kind of people are you? I certainly don't go around calling people a 'dolt' or a 'stupid fucker' esp in public just because they happened to be mistaken on something. C'mon people, we geeks can be more civilized than that.
deimos@kfa.cx
HOLY COW! I posted again! /me hides in the corner chargrined...
PIIs are already run hotter than any chip out there. Hell. I'll put money on the fact that in
a couple of years, people with PIIs will:
1. have to buy new motherboards to support the PIII or P4.
2. have to junk their system cause the chip is fried out.
okay folks:
I could put a rocket engine on a Yugo and it would go faster than any other car out there, but it'd still be a p.o.s.
I'm afraid. I'm afraid, Dave. Dave, my mind is going. I can feel it. I can feel it. My mind is going.
Just let everyone burn up their chips trying to o/c above spec and that way, they will buy more intel chips ... o/c TOO much, burn it up...buy more intel chips...etc etc
But it will suffer from a major "meltdown" when clocked too high (and therefore run too hot). I know people who have wrecked chips, and even motherboards, doing this.
And Intel, along with keeping the unscrupulous resellers at bay, locked its chips to keep anyone in the know from telling the cattle that the Pxxx's were already out for cheap.