O/C XP2800 to ???

Discussion in 'Benchmarks & Performance' started by scriptasylum, Aug 14, 2003.

  1. scriptasylum

    scriptasylum Moderator

    Messages:
    832
    Location:
    Des Moines,IA
    Hey, anyone out there know about how high I can push the FSB of my Barton without increasing the vcore? The memory should be able to handle 400 even at low latency, but I don't want to push the CPU that far too fast. Right now I've got the FSB set for 174 and the temps are about 42°C idle. I'm basically just looking at a ballpark figure that some people have done successfully.

    Another question: I've read some places that the 2800 is factory unlocked and some say it isn't. Is there a way to tell by a set of the bridges? I really think it's locked, because I don't have the mobo setting any particular multiplier (on auto) and it reads as 12.5. So, I figure something has to tell the mobo what multipier to use, or how would it know? Is the auto-unlock (if any) a feature of the nForce2 boards?
     
  2. jumpy

    jumpy Guest

    The cpu tells the board what multiplier to use so auto means it will set it to 12.5. Nforce boards should unlock most post tbred cpu's. Try changing it, it's the only way to find out :)
    According to www.cpudatabase.com the average o/c for that chip is 2348 Mhz. I would see if the multipliers work alright and then aim for 200fsb. I would think 200x11 would definitely be achievable on default volts.
    Good luck, don't put too many volts through that nice expensive cpu of yours :D
     
  3. Taurus

    Taurus hardware monkey

    Messages:
    3,206
    Location:
    Sacramento, CA
    i've built two nforce2/2500+ systems and one is doing 11*200 and another 10x200 (stock hsf on that one), both with stock voltage. 2800+ should be able to do 12x200 easily. start with 11x200 first to be safe and try moving the multiplier up in increments.

    i have a rev 1.0 nforce2 board so my fsb doesn't get much past it's 166. :(
     
  4. scriptasylum

    scriptasylum Moderator

    Messages:
    832
    Location:
    Des Moines,IA
    Ok, thanks.

    I did try it and went to 200*11, but am leary about setting the FSB so high from the CPU's spec all the time. It was running fine at around 47°C. My 3DMark01 score (for those who still take stock in it) was 17,571.

    If the 333FSB CPUs can so easily run at 400FSB, why doesn't AMD just sell it as such? Aside from thermal issues, what could theoretically happen to a CPU running the FSB so high out of spec? Since I'm running an nForce2 chipset, the AGP/PCI clocks are locked at their spec, so that isn't a concern.
     
  5. Taurus

    Taurus hardware monkey

    Messages:
    3,206
    Location:
    Sacramento, CA
    *shrug* why do athlons have overclocking headroom instead of being sold as their capable faster speed?

    it's just the way it is. i'm sure it has something do with the headroom. for instance, if you tried to push yours past 200, it probably wouldn't go far... and amd wants there to be that room there for each spec (fsb, core speed) to absolutely guarantee stablility and long life. so since yours didn't have eneough room above 200 to make amd happy, it was made a 166fsb cpu.

    this is just my theory. i seriously doubt you'll have any problems with your cpu running it at 200. especially since you're not upping the voltage. increase your multiplier some more and see how far you can go without touching the voltage. my friend's 2500+ is doing 11x200 at stock voltage. i know you can go higher.
    ;)
     
  6. scriptasylum

    scriptasylum Moderator

    Messages:
    832
    Location:
    Des Moines,IA
    I can see AMD giving their procs some headroom, but 33 MHz on the FSB? That is a 20% increase on FSB alone. I am currently running 200x11 now, but I don't think I'll go any higher just to be safe.

    BTW: In case anyone has the GA-NNXP or GA-7N400Pro, someone over at amdmd forums has made some new BIOS's based on the new ones posted by Gigabyte. They mostly address SATA and RAID issues, but there are some other enhancements as well.
     
  7. LeeJend

    LeeJend Moderator

    Messages:
    5,291
    Location:
    Fort Worth, TX
    This has been said before but...

    All processor chips come off the same assembly line! (Pretty much, if demand is very high there may be multiple FABs.) There is no "chip headroom". Speed grading is not designed in, it is the result of process control refinement issues in manufacturing. All Bartons were designed to be ~XP3200-3400 parts using a 0.13 micron process.

    How it works:
    Many silicon wafers (10-12 inch diameter) are cut from each boole of doped silicon. Many wafers are all then simultaneously impregnated by impurities and metal deposited on their surfaces in very large vacuum chambers to form transistors and conductors on the wafers. The wafers are then sliced and diced into individual chips.

    All the chips in this batch (and we're talking 10's of thousands) have essentially identical performance characteristics. Small differences come from where on the wafer a die was and where the wafer was sitting in the vacuum chamber.

    The die are then checked for gross functionality and the working ones are mounted on substrates. After mounting performance tests are run on each die and the parts are then graded for speed. (The parts that fail the lowest performance ratings get sold to digikey, radio shack and certain undisclosed OEMs.) Now comes the fun part. You get to slap a 300% higher price on the few parts that work the best.

    When you first start making a new chip the process isn't perfect so only a few die are the fastest. As you do more and more runs you figure out how to make all the chips come out the same as the best ones and maybe even make them better.

    Now comes the sad part. Engineering has now busted their asses for ~12-18 months and done their job. You are making 90% or more of your parts at the highest speed rating possible for the process and design, but marketing has their heads up their butts because they can't find enough fools out there to pay 300% more for a 25% speed increase.

    What is a CEO to do????? You can't sell all the chips for the same low price. Marketing advises you "that might wise up the speed freak fools to how stupid they are". So marketing tells engineering to make a substrate with little metal lines that can be burned through with a laser built into the test stand. The little metal lines lie to these strong, bold new chips and tell them that they are slow and stupid. You then sell these large quantities of misinformed chips, along with a few clunkers that were at the back of the line during manufacture, for the usual $100 and keep selling the rest of the identical chips to the fools for $400.


    Technical Notes:
    1). Intel has more respect for the inteligence of their customers and builds the metal lines into the die itself so they can't be defeated by super glue, conductive paint or pencil lead. You have to push the FSB instead which is a little trickier...

    2). As you improve processes you reach the underlying physical limitations of the particular chip design. i.e

    a). The capacitance of the transistor junctions and metal runs requires so much current to charge them that more heat goes into the die than can be removed using reasonable cooling methods. (Why the original Tbred & Tbred A were rapidly replaced by the Tbred B dies.)

    b). The time it takes the electric fields to move through some parts of the die are longer than it takes to move through other parts of the die and a race condition is created. The die starts doing things out of sequnce, gets confused and stops working.
    (Why intel is kicking AMD butt and AMD has to go to 64 bit data paths to keep up. Also why cryo cooling can push a given chip design only so far.)

    3). The "pretty much all from the same line" applies to using multiple production lines for die. This used to be common but with larger diameter wafers, smaller die, faster new product introductions it is becoming "not cost effective" to spend all the money needed to rediscover all the tricky little fabrication details that make a perfect part every time on adifferent production line.

    Ok, I vented. I feel much better now. :D

    Edit - Mommy read this and said I can't skip my medication anymore...
     
  8. jumpy

    jumpy Guest

    Whoaa, long rant there Leejend :D
    Everything you said there rings true. Speed binning is why you get some insane overclocks from the latest manufacture date low end chips. AMD and intel keep refining there process so more of their cpu's perform higher, they then downgrade them when the market calls for it.

    There we go, summed it up in two sentences ;)

    EDIT: fixed the name up :p
     
  9. Taurus

    Taurus hardware monkey

    Messages:
    3,206
    Location:
    Sacramento, CA
    nice, leejend.

    all the more reason why scriptasylum should push his proc higher. :D it won't blow up or anything. once it reaches it's limit, it will just be unstable... then you back it down one notch.
     
  10. scriptasylum

    scriptasylum Moderator

    Messages:
    832
    Location:
    Des Moines,IA
    Ok, thanks. (I guess)

    I have it stable at 200*12.5, but the temps are getting up there at 51°C, so I backed it down to 11.5 to get a more decent temp of 46°C... which is wierd, because I had it at 200*11.5 before and the temps were higher, now they went down at a higher clock?? Oh well.
     
  11. LeeJend

    LeeJend Moderator

    Messages:
    5,291
    Location:
    Fort Worth, TX
    Funny you should mention that. My CPU temps dropped since I first installed it too. Weird, I'd expect temp's to go up from dust build up and the thermal paste drying out.

    Oh well, not gonna complain about good luck. :)