Maybe cpu burn-in is true

[Mixxen]

Well, I've had my cel 533 @ 822 for about 6 months now. I remember trying it at 896 when I first recieved it, and it wasen't able to hit 896. Now, after building my friends 900 tbird system, and becoming a bit jealous...I tried my cel at 896, and it's stable. go figure.

 

[jinsonxu]

Try this thread:
See pm's post

 

[Mixxen]

Oh well...maybe it's something else. Who knows..and again, go figure. Electrophysics is not my major.

 

[pm]

I can't explain why it seems to work for some people.

One time about a year ago, I got really curious about the fact that people on the Net reported situations like you mentioned, Mixxen. So I wandered down a couple of floors to the process gurus and they said that the idea that silicon could get faster with time was preposterous. No one had ever heard of such a thing - and, mind you, these are the gurus at Intel who do all the process work - and then proceeded to give me a half-hour lecture on hot electron gate impact ionization, electromigration, pMOS bias temperature instability and a whole lot of other reliability issues. But one suggested pulling a report on the pre-production burn-in yields and pre and post speed bins at Intel for a processor that was, at the time, the most widely produced at Intel. So, with a little bit of begging (since I didn't really need this data to get my job done) I got a report that showed the stastical variation of speed bins for parts going into burn-in and coming out for a processor that you all would recognize. We don't normally keep this data, but it's taken routinely during pre-production. So, I looked at this and, with a statistical sample size of well over 10,000 units (I can't remember the exact number... 16k maybe), virtually all parts got slower. There was a very small anomoly that did actually get faster, but this was a big time statistical outlier... like 3 sigma out (which means a very small amount for those of you who forgot their statistics).

Very rarely do I pretend to have all the answers (on this BBS or in life). I'm just a low-level engineer who does circuit design at Intel and I'm no expert or guru. But on this one issue I have done enough research to qualify for a Master's degree on this subject. Silicon does not statistically get faster with time. Whatever effect you guys are seeing is not an effect of the silicon - unless you all happen to be lotto winners as well.

Since burn-in is designed to destroy parts, I highly recommend that you do not do it. Increasing the voltage of a processor well beyond spec is the best way that I can think of (short of using a hammer) to kill a microprocessor in the fastest time possible. It's your CPU and you can do with it what you want, but this is my advice.

Patrick Mahoney
IA64 Microprocessor Design
Intel Corp.
pmahoney@mipos2.intel.com

 

[Mixxen]

Hmmm...it's probably the Maxtor drive I bought a couple of months ago, doh, I should have oc'ed the cel when I installed the new Maxtor. Doesn't Intel use overclocking methods to achieve more yeilds on the higher end P3's. I'm not too sure, but aren't the high end P3's using 1.75V instead of 1.65V?

Hey, BTW pm, do you happen to know any good links on the net where I can obtain information for the Intel x86 architecture. I'm doing a report for class...and you seem like the right person to ask .

 

[pac1085]

My Celeron 566 seemed to do the exact opposite. When I first got it, it would do 952 and 935 stable, but after a few weeks, it would only do 850, and 893 if I was lucky. I can understand what PM is saying about silicon getting slower as it grows older because of this. My AMD cpu is still fairly new, and im currently running it at 1GHz (2.10v), and I will see what happens with this chip...

 

[pm]

Mixxen

Do you want 32-bit x86 architecture info - like the instruction set, the way the registers are set up, etc? Or do you want microarchitectural details on a specific processor?

I'll assume the first. Call 1-800-879-4683 and
order "Intel Architecture Software Developer's Manual vol. 1: Basic Architecture" Reference Number: 243190. It should be free. If it isn't then say that you were referred to the number by an Intel employee and that this is for a school project. Call me if you have any trouble: 970-898-3262

This an excellent book for intro to the x86 architecture. It's fairly easy to read and it's very complete.

Edit: I tried calling and they are sold out. Mixxen, email me.

 

[pac1085]

I want a copy of that book It would be very intersting to read...

 

[pm]

You'll have to stand in line, Vortex.

Seriously, it is a good book - especially if you are into x86 assembly. I thought about it over lunch and decided we must have an online version, and there is one. IA32 Software Developer's Manual vol.1. It's Adobe Acrobat's .PDF and it's 2.5MB. Vol1. is the intro - covers the basics of the architecture. Vol 2. is the instruction set reference. Vol. 3 is the system programming guide.

Mixxen: email me your address and I'll snail mail you my copy... I'm IA64 nowadays anyway. pmahoney@mipos2.intel.com

 

[nateholtrop]

atleat you can hit 800 I can't w/ my cel 533 I can get it to 760 but I think that is because of my ram. I have generic mixed and match pc133 ram I got a stick of NEC 64Mb ram and some generic crap. oh well.

 

[Mixxen]

Awesome...there's a lot there to read though, in one week.

I scanned through the pdf's and there's a lot of info on the instruction set (which I do need BTW). But I also need info on things like the Pipeline Stucture, Superscalar, Cache, ect.

Thanks a bunch for your help! Very cool

 

[pm]

Ah. So you want microarchitectural details then.

Architecture is the instruction set. It doesn't change much. The last major change was with the 386 when we went from 16-bit to 32-bit. There have been minor changes since with MMX and SSE and a few minor instruction changes between generations, but nothing major.

Microarchitecture is the actual implementation of the instruction set into a microprocessor. 386, 486, Pentium, etc. Which specific microarchitectures are you interested in? I should be able to turn up a few articles that have flow charts, details and pipeline charts.

 

[qiu]

Hello ICSVortex...

You're gonna get it burned allright @2.10 V No question about that...

 

[pac1085]

Hey, 2.10 volts OWNS! lol,I doubt I'll toast the chip, even if i do, it was only 50 bucks. I dont care about how long the chip will last, because I'll sell it off to somone in a few weeks when its obsolete

 

[Mixxen]

Well, anyone of the Pentium I, II or III would be nice, if they are sort of similar then one should be great. I have some manuals on the 8086 and 386, so I proabaly don't need any of those. Do you know if there is info on the IA32...just to see how it differs from the previous x86 lineup.

Hey pm, thanks a bunch again!!!

 

[pm]

Mixxen: I presume by IA32 - you mean IA64... right? Or are you really interested in how IA32 differs from the old 286 16-bit x86? There's not really that big a difference. There are plenty of little differences and they tend to add up to a big difference over all. Still, that first reference that I posted should mention the differences between 16-bit code operation and 32-bit operation.

If you mean IA64, this is a huge difference. It would take a while to discuss.

If you want you are welcome to take this offline by emailing me.

I'll see what I can turn up re: simple explanations of the various microarchitectures you mentioned.