Rating CPUs

[pjungwirth]

Hello all! This is my first post to this forum, so let me begin by saying what a great place it is for information! Thanks to Charles for putting this site together.

So here's my question. Does anyone know how a company rates the speed of cpus? They just have to test them, right? That means they see how high they go before they fail? Now, if overclocking a cpu can ruin the chip even for lower speeds, can't this test also?

I don't really trust the idea of overclocking, but I was just wondering: hasn't every chip been "overclocked" once? One argument in favor of not overclocking is security of mind that a chip has never been run beyond its specifications, but *hasn't* it, really?

Or am I mistaken about how a company rates its cpus? If anyone has any information on this, I'm very curious to know.

Thanks,

Paul
~{:-)

[bassvax]

Try www.cpuscorecard.com (http://www.cpuscorecard.com) excellent comparison site.

------------------
Tango down!

[pjungwirth]

Sorry, I guess I wasn't clear in my first post. I don't mean how does a company (or someone else) benchmark a cpu that runs at a set MHz. I mean how does a vendor test a cpu to determine how to clock it? How does a vendor decide "This is a 700 MHz and this is a 900 MHz." Do they just keep clocking it higher until it won't run?

--Paul
~{:-)

[tjaymadison]

Since the etched silicon wafers that eventually become the 'guts' of a CPU
are produced in batches, I'd assume the testing would be done by sampling,
rather than individually. The samples are very probably maxed out in order
to determine the most reliable top speed with an acceptable margin of error.

Those tested samples are then most likely kept for archival reasons, and
not sold as either OEM or retail products. If a CPU intended for sale gets
any tests at all, they are bench tests for pass/fail, not for performance.

There are just too many DOA chips to believe they are tested individually.

------------------
"I am not able to rightly apprehend the kind of confusion of ideas that could provoke such a question."
-- Charles Babbage, mathematician, computer pioneer, analytical engine designer (1791-1871)
-- (Question: 'Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?')

"Just because I don't care, doesn't mean I don't understand."
-- Homer Simpson

[This message has been edited by tjaymadison (edited 06-06-2001).]

[AwARe]

Hello pjungwirth,

Its been a while since I really looked into any of this kind of stuff (so it might be a little different now), but if I remember correctly......

Basically a CPU is "designed" to run at a certain specification, in other words they know ahead of time "before" they start popping them out of an assembly line, what the expected mhz, voltage, temp, output, etc etc etc are going to be for any given CPU. Once the CPU is put into production, the resulting CPU's are then tested on a sample basis, whereas they grab a couple off the line and see if they are living up to specs. I would imagine during this testing they outright burn a few of them up. These CPU's that are tested are most likely "Not" sold.

This is how you end up with a CPU that can be OC'd. It would be imposible for them to be sure that "every" CPU off the line will run at the maximum that lets say the best one that came down the line could. This means they would "Have To" allow for some room for lower performing chips off the line. They also cant very well rate them to run at "max output" as this would greatly reduce the "lifespan" of a chip. (and they are getting alot better at knowing just what a chip is gonna do way in advance)

They have to find a balance, between the percentage of chips and how they perform off the line, and at the same time trying to get the chips to run at the fastest speed they can posibly sell them at, and also rate them at speeds that give them a reasonable lifspan, so as to protect their reputation as a reliable chip. Some CPU's at times have far outperformed their design specs and others have also left a bit to be desired.<- Some CPU's are better OC'ers then others.

Some chips have not lived up to expectations right off the line, and it has been necessary for the rated speed to be slightly lower then what the chip was designed for. For the most part, if a chip "outperforms" what it was designed for, it is still sold at the speed at which they designed it to be sold at. For example the celery 300a that everyone I know that bought one, ran it at 450, and a great majority of the chips they sold were capable of 450 with very little trouble, but they never repackaged them to be sold as 450. (At least I dont think they did http://www.PCGuide.com/ubb/smile.gif )

OC'ing a chip also for the most part is going to take a bit of trial and error. Sometimes it requires some extra cooling, and alot of times it also requires a bit of close monitoring, or even more. It really wouldnt be practicle for them to sell a chip that required all this "extra work".

Think about this, if an office manager had to personally verify the speeds and cooling requirements of "each" of the computers he just purchased for his employees, and the employees all had to be held responsible for monitoring the tempature of their CPU's at any given time, well that kind of speaks for itself.

Another thing to concider is the fact that even though a CPU will boot up at a certain speed, doesnt mean that it's not going to overheat while its under heavy use. This is another problem all together.

There is also the issue that even though some CPU's will report a higher "mhz" this doesnt necessarily mean its going to give you all that much of an improved performance. In alot of cases people are just putting their CPU's under stress, and burning them up, for no good reason at all. All this excessive heat and wasted money, just so they can tell their friends they have a 14ghz CPU heh.

In the end, if you want to spend some time and see just what you can get out of your CPU, go for it!. Its alot like a mechanic that fine tunes a race car, he puts in a bigger radiator, a different radiator cap, and a new thermostat(cooling fan), better spark plugs (Ram), and plays around with the adjustments (BIOS settings etc etc). In the end he has it running much faster and better then it did when it rolled of the showroom floor. Keep in mind just like some cars make better race cars, the same goes for CPU's, they are all different.

Theres not a thing wrong with "Responsible" OC'ing.....just remember its not about "mhz"............Sometimes a chip running at a lower overall mhz with a different multiplier, will outperform a chip running at a higher mhz......(The motherboard and video card etc, are also in most cases alot more important then what mhz the CPU is running at)

EDIT: In the "old days" we used to have alot more control over all the related settings, and just about "any" CPU would run at about 30% OC'd. OC'ing was a fairly simple matter back then hehe (Not like it is today), but that seems to be changing, and that 30% used to mean alot more then it does now. Also in the "War Of The Mhz", that 30% is dwindling in some cases.

When you used to OC a 166 to 208 or a 200 to 250 you could really "see the difference"........Not so much anymore...

Hope you find this Helpful...

[This message has been edited by AwARe (edited 06-07-2001).]

[yawningdog]

Hi.
A special fixture tests each of the chips while still on the wafer and the bad ones are marked. All dies are then cut from the wafer with a laser or a diamond saw.
After the chips are bonded and packaged, final testing is done to determine speed and other such parameters. Different chips from the same wafer often run at different speeds. Maximum successful speed noted during testing determines the speed rating.
As the efficiency and proficiency of a chip assembly line increase, the line is able to yield a greater percentage of the higher clock speeds. In fact, if a line turns out too many high speed chips and cannot fill an order for lower speed chips, sometimes they simply re-label some of the faster ones to fill the order. http://www.PCGuide.com/ubb/wink.gif

------------------
Make everything as simple as possible, but no simpler.

[AwARe]

heya dog,

Do you have a good link to a site that explains the current process used fairly well? Sounds as if you may have been doing some reading hehe.....

Like I said earlier its been a "long time" since I checked into any of this kinda stuff really.............As far as I new from previous stuff I read about it...I knew they got dif speed CPU's from the same wafer, but as far as I knew before, they didnt really check each chip. They checked "different zones" in the wafer and divided it into sections, cut the chips, completed the mfg process and then did sample tests on the final product.......

Like i said in my earlier post "so it might be a little different now"....I checked several of the semi-conductor foundries sites for specific current info, but they dont really seem to offer much. They only say that they "offer" tested, non-tested, sorted, non-sorted etc etc, but it doesnt really say anything about if intel for example is actually paying to have the wafer tested in this manner......

If you have a good link Id appreciate the read.........Thanks

[yawningdog]

I used to work in the semiconductor industry for a company called Varian Ion Implant Systems in Gloucester Ma. We made the E-1000 which is what intel was using at the time to build the pentium.

Sorry I can't give you a link, but my text was paraphrased from a book called "Buying and Upgrading PC's - 12th edition" by Scott Mueller. You'll get some idea of how much reading I've been up to when you see the book.

They have to test each chip because in the early stages of a production run, chip yield from each wafer can be as low as 50%, and nobody likes to pay >$300.00 for a souvenier. http://www.PCGuide.com/ubb/mad.gif Manufacturers are really uptight about what information they give out, such as yield rates, because competition is fierce, albeit narrow. A skilled marketing department can turn such information into a train wreck and just about ruin a products marketability

------------------
Make everything as simple as possible, but no simpler.

[This message has been edited by yawningdog (edited 06-08-2001).]

[AwARe]

heh yeah I can see where that would be a probelm and why I cant really seem to get a clear answer...........

I dont think I have time right now to read a book, was really hoping for a web page or 2 hehe, but thanks for the info http://www.PCGuide.com/ubb/smile.gif

Maybe after I figure out where Im gonna be living in 3 weeks, find a second job, and my kids graduate school Ill check it out heh

I actually have that book lmao if ill ever read it I have no idea.
You read that thing? Honestly Im kinda scared to even look to see how many pages it is haha. My full version dictionary is thinner http://www.PCGuide.com/ubb/eek.gif

[This message has been edited by AwARe (edited 06-08-2001).]

[bassvax]

Hey AwARe gotta chime in here...I have that book and it is very much worth owning and reading...the lingo is understandable and the guy is a genius. Great leisure reading as well *believe it or not*. I have lost the desire to read books...used to read Agatha Christie, Patricia Cornwell and a few other choice works, but...I can still get into that PC book for some reason..maybe the phase of learning about puters is what lets me...Oh well

------------------
Tango down!