Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
Check out the new SourceForge HTML5 internet speed test! No Flash necessary and runs on all devices. ×
AMD

Paint Yourself An Athlon MP 150

SNIa asks: "How many people are checking the prices of AMD chips after seeing this?" and points out this article at HardwareZone.com about modifying Athlon XP processors to perform like MPs. No guarantees, except a voided warranty.
This discussion has been archived. No new comments can be posted.

Paint Yourself An Athlon MP

Comments Filter:
  • sort of (Score:3, Informative)

    by GigsVT ( 208848 ) on Monday March 18, 2002 @06:27AM (#3180295) Journal
    . However, do note that connecting these pads together does not automatically modify the CPU into an Athlon MP. It stays as an Athlon XP.

    It doesn't change the performance of the chip, only allows you to run two of the newer XPs in a dual motherboard.
    • What's the diff? (Score:2, Interesting)

      by BLKMGK ( 34057 )
      If the XP and MP cores are the same then what exactly is the difference? The BIOS says XP instead of MP? AMD maybe did some testing? Perhaps the cut trace is the result of a failed test? Or more likely a marketing gimmick like the locked multipliers onthe Intel CPUs? "Hey Bernie, we're out of 500mhz CPUs could ya' downgrade some of the 1gig units?" As yield increases on established CPU lines this really does happen if there's a shortage of slower CPUs. I see no reason to believe that AMD isn't just as smart both with clock speeds and the SMP ability. As fast and as cheap as their CPUs are I've seen fit to go ahead and buy the higher speed units and not monkey with the multipliers but to save $80+ I'll be more than happy to mod an XP CPU's SMP jumper :-)

      If AMD doesn't like this then they're going to have to be MUCH clearer about what's different even if it's just to say that it's added testing involved. It's not like the CPUs come with a report card telling you what failed and what didn't ;-)

      Benchmarks man, we need benchmarks and lot's of them. I'm sure it won't be long before Annand, HardOCP, or heaven forbid Tom's (gak) get onboard with this and benchmark us to death. :-)
      • "Hey Bernie, we're out of 500mhz CPUs could ya' downgrade some of the 1gig units?"

        That Bernie guy is just the kind of asshole that would go along with it too...

      • Re:What's the diff? (Score:3, Interesting)

        by saider ( 177166 )
        If I recall, microprocessors for a given line are all produced using the same process. Because the process is not exact, each microprocessor has slight variations from the one next to it. These variations have an impact on performance.

        These variations are typically the reason for different clock speeds. The more variations there are and the farther they are from the spec, the less likely the component will work as designed. However, by slowing down the clock, you can get the component to work reliably. Also, microprocessors have redundant components in their design so that if one is not fabricated properly, the microcode will use one of the redundant blocks.

        Simply put, AMD doesn't say they will build 700 XP 1400s and 300 MP 1400s. Rather they schedule a production run on a given line for 1000 chips knowing that about 30% of the chips will be suitable for multiprocessor use. If there is only demand for 100 MPs, then they will simply label 200 of the chips that passed the MP test as XP in order to satisfy the demand. The chips that came out perfectly are the high clock speed/large feature count chips. If features fail or the chip cannot work at a given clock speed, the requirements are dropped to a lower level and the processor is labelled as such. Also not that as time goes on and the process is refined, more "perfect chips" are made and so more chips of a higher clock speed can be made. This is why processors start life with a slow clock and slowly get better as the manufacturing process is tweaked.

        So, when you put a XP into a MP system, you are taking some chances.

        1) Your chip is simply a relabelled MP that works fine in both roles and was relabelled to meet XP demand.

        2) Your chip is a failed MP that was labelled as XP because not all the MP parts work right.

        Either way, you have little to loose by trying it as an MP. I would use a reversible techique, such as soldering copper foil to the pads, so that the mod can be undone if the processor is a #2.

        The reason that AMD does not publicly say this is that it is very difficult to explain this to the average, or even technically inclined Joe and they get little benefit from doing so. It would be a disaster if AMD said to TechieJoe that MP and XP were the same thing and said to AOLJoe that MP and XP are different. Just stick to the story that the processors are different, because they usually are.
        • Re:What's the diff? (Score:2, Interesting)

          by Saurentine ( 9540 )
          Simply put, AMD doesn't say they will build 700 XP 1400s and 300 MP 1400s. Rather they schedule a production run on a given line for 1000 chips knowing that about 30% of the chips will be suitable for multiprocessor use. If there is only demand for 100 MPs, then they will simply label 200 of the chips that passed the MP test as XP in order to satisfy the demand.



          This was the case with Intel 486DX and SX processors, but there is no evidence (not even anecdotal) that AMD does MP/XP selection in this manner. Why not? AMD has been getting great processor yields lately. They'd probably only save a few chips per thousand from the dustbin that could have been XP. And because they're not the leading company in the market, it's better to toss the chips that fail the MP test in case there's a further flaw that economical high speed testing won't detect.

          Intel did the 486 SX/DX thing mainly because of tremendously large demand for processors that commanded $395-$1000 each. Sure, they kept doing it later as a convenience, but it was the high demand/high price/low yield situation that led them to do it in the first place.

          Frankly, I would have been shocked if someone said there was a real difference in the cores (even before this revelation; it's the packaging where the differentiation is made), and I'd be incredulous if they were grading out reject MPs as XPs. There just isn't enough incentive to do so.

    • only allows you to run two of the newer XPs in a dual motherboard...

      Any K7 processor (Durons included) can be used in an SMP setup. The only difference (IIRC) in the MP is the data prefetch unit, which is modified for better performance in multiprocessor situations. There is nothing newsworthy described in the article, and I'm wondering how in bloody hell did it get on the front page.

      • It's entirely possible that the MP data prefetch unit is the exact same unit XP's have..

        possibly with slightly different microcode, but I'd be VERY surprised if AMD were fabbing two chips based on the exact same core with slight differences throughout.
        • I agree, it's much more expensive to produce two unique product lines to do such similar tasks, than to just disable a feature of one and leave it on another.
          I know for a fact (My mother is a Xerox technician), that the only difference between the 6135 (135ppm), and the 6180 (180ppm) printers is a software switch that sets the effective speed of the printer (They sometimes need to set it to the higher rating (or the lower rating, in the case of the 6180) for testing purposes.
          It was a MUCH less expensive way than producing 2 distinct product lines with such a subtle difference.
      • Exactly. I guess enough people don't know the difference between the XP and the MP. Yes, they both use the Palomino core, which is SMP enabled already. The MP is just tweaked to perform better in an SMP environment. People have been using XPs and Durons in MP motherboards for quite awhile now. No news here, and in most cases there is nothing to "mod".
    • Agree. I have a renderfarm (6 computers) based on XP cpus, the only risk taken by using XPs in MP motherboards is that you could get a BIOS update one day that will "disable" XPs. I've bought 1 unit to test it before hand to be sure everything was stable, and that I wouldn't need bios upgrade afterwards (since it's always doing the same thing), it was okay, and still is today.

      Even with older BIOS I would get "2 MP found" in the system (tyan tiger MP), with the 2.03 bios and after, its correctly identifying MPs from XP.

      There's probably something with MPX motherboards BIOS that is disabling XPs.

      Performance-wise, I've noticed a subtle difference when I put 2 ram stick versus 4 ram sticks (must be something that has to do with how memory is accessed), but that's about it, MP and XP in rendering don't seem to have a difference (I have 2 MP for my server, I didn't want to take any risks on that :) ).
  • "However, we cannot guarantee if it would run stable as some users reported problems with dual XP systems."

    I think this is the understantment of the century, ... can anyone he frying ...
    • Re:Understatment (Score:1, Insightful)

      by Anonymous Coward
      It's not like you're overclocking the system - if the core logically supports being run in an MP configuration, there is no reason why there should be problems otherwise.

      However, if the processor didn't actually support MP (doesn't apply in this case), you are guaranteed to have problems. But those problems should be quite severe (cache coherency problems? IPIs don't work? It'll never even boot successfully).

      There is nothing physically special about an MP configuration, so unless the modification does something else as well as identify the processor as MP-capable, it shouldn't be dangerous.
  • by nzhavok ( 254960 ) on Monday March 18, 2002 @06:32AM (#3180311) Homepage
    that they would have this capability, although disabled. I mean they just cripple the chip so that they can get more cash out of the MP versions, smart move unless you wan't another fiasco like the celeron/pentium II.

    The bottom line is that only a few hobbiests and nerds are going to go to this length to get it working, certainly not many businesses are going to accept this. So AMD makes it difficult enough so that you can't do it at the flick of a switch, but you can with a little guts (and silver paint). Seems like a win/win situation to me.

    Also don't forget it may be easier just to use a couple of Durons to accomplish this task, dirt cheap too :)
    • To be honest, I imagine spending what you'd spend on 2 Durons on 1 Athlon XP would get you equal or better performance overall. Because Durons have a smaller L2 cache (64K vs. 256K), I would also imagine they need more memory bandwidth, leading to a not insignificant loss in scaling compaired to 2 Athlon XPs.
      • Yes you are almost definitly right! Also with the cost of some MP boards you could almost write the durons out of the equation, don'tknowwhat came over me.
    • The Celeron/Pentium III wasn't a fiasco. Just because some enthusiasts bought celerons and used them in dual-boards, does not mean most people did.
      • The Celeron/Pentium III wasn't a fiasco. Just because some enthusiasts bought celerons and used them in dual-boards, does not mean most people did.

        He said Pentium II(2) not III (3). A great many people bought much cheaper celerons rather than PII's and overclocked them, getting very nearly the same performance for about half the cost.
        • I don't have any inside information here, but it is important to note that you said "half the cost." not "half the profit margin for Intel."

          Clearly celeron yeild was much better at the time, so without Intel's internal numbers we can't know if this was good or bad for them.

          This is compounded by the fact that a significant percentage of those celeron sales were cutting into /AMD/'s market, not the pentium.

          -Peter
          • Clearly celeron yeild was much better at the time, so without Intel's internal numbers we can't know if this was good or bad for them.

            You have a valid point, but seeing how Intel about that time started taking steps to make overclocking more difficult than it had been previously, I'd say the internal numbers didn't indicate that this was a good thing.
      • I said Pentium II not III. In this case the 'fiasco' I was refering to was nothing to do with dual boards. I was referring to Intel creating such a good chip (Cel-A) at a low price which essentially cut their throat on the high-end version which was the Pentium II. People didn't see the point in buying Pentium IIs when they could get a similar performance celeron at the same price. Now if AMD hadn't disabled the MP ability on XP processors they could have a similar situation, i.e. a lot of people using XP instead of MP.
    • The other good thing about this approach is that corporate customers, who are sensitive to warranties and need to cover their asses, will never overclock these processors. So, effectively, they can offer a bit of extra power as a sweetener, but only to techfreaks.
    • >Doesn't really surprise me that they would have this >capability, although disabled. I mean they just >cripple the chip so that they can get more cash out >of the MP versions, smart move unless you wan't >another fiasco like the celeron/pentium II.

      Ah, yes.. harkens me back to the days of the ole Intel SX/DX, present-but-disabled math copro... the good ole' days.
  • by zensonic ( 82242 ) on Monday March 18, 2002 @06:35AM (#3180316) Homepage
    While serving the purpose of locking the cpu for ordinary people, these small bridges on top of the cpus have attracted all the HW geeks which have unlocked, overclocked, underclocked (to get a quiter system)these cpus.

    One of the biggest strength in the Athlon line (in my opinion) is that AMD have been able to sell locked cpus to 90% of the population but have made it possible for the last 10% to overclock these cpus. They have won both the normal consumers and the techfreaks in that respect.

    .... ofcourse it also helps that their cpus are affordable and that there are plenty of them!
  • What? (Score:3, Funny)

    by O2n ( 325189 ) on Monday March 18, 2002 @06:36AM (#3180318) Homepage
    On the second page:
    Approved by Jimmy Tang on Friday, 15th of March,2002 [...]
    Below is where you'll find the bride.


    Being Friday, maybe the thought of getting layed enhanced his sense of humor... :)
  • by Soft ( 266615 ) on Monday March 18, 2002 @06:39AM (#3180321)
    For a while I thought it was like the Pentium III [userfriendly.org]...
  • spoiler...
    now how do you get an athlon XP chip to act as a MP chip? just paint it up so it thinks it's an MP chip.
    • Re:spoiler .... (Score:2, Interesting)

      by ackthpt ( 218170 )
      Good question.

      I'm in the midst of building a dual CPU system and was interested to see reports of Durons and Athlon XP's running in dual motherboards. Certainly the builders think they are getting something like MP performance. Seems to me that successfully getting two CPU's to run on an Asus A7M266-D, Tyan S2466N or MSI K7D isn't necessarily the same as having two MP. I'd like to see benchmarks showing a negligible performance difference. All that money to get there, I wonder why some still try to skimp a few bucks.

      IMHO it's not too unlike hotrodders, many of whom knew tricks to get more power out of cars, but at what cost, i.e. stinky exhaust, short lived engine, failure to pass smog (in CA) Not quite the depth of knowledge, but lots of creative guessing.

      If painting, the stuff to use is in little bottles, sold at automotive stores, used to repair reat window defrosters.

  • by tapiwa ( 52055 ) on Monday March 18, 2002 @06:42AM (#3180329) Homepage
    This is taking overclocking to the next level. The kind of people who do this sort of thing have always had the "warranty voided" hanging over their heads.....

    What I would be keen on is for someone who has done this to tell us how stable their machine is.

    I am yet to read a article on overclocking etc that does not say that some users have experienced problems... ... I think tech writers use this spiel instead of the less friendly "we are not going to be held responsible if you fsck up"

    • You could always cut and scrape the paint with an exacto knife, the process is reversible.
    • by Space cowboy ( 13680 ) on Monday March 18, 2002 @08:44AM (#3180568) Journal
      I have a dual Tyan m/b with two XP chips in it. It runs hot, but it's been calculating SETI@Home more or less 200% of the time since it was installed, about 3 months ago. No problems, and I think it would have found them by now :-)

      Pretty soon it's going to be taken off SETI and put onto the real task, but I thought S@H would be a good reliability burn-in (not to mention adding 6-8 S@H units per day to my total :-)

      Simon
      • I have a dual Tyan m/b with two XP chips in it. It runs hot, but it's been calculating SETI@Home more or less 200% of the time since it was installed, about 3 months ago. No problems, and I think it would have found them by now :-)

        Yeah, if a dual XP system can't find them aliens then nothing can!

  • This proves little. (Score:4, Interesting)

    by Niadh ( 468443 ) on Monday March 18, 2002 @06:44AM (#3180332) Homepage
    The article never got out of POST... It's a nice though that just because you fooled the motherboard into thinking the CPU's have MP support they really do. This article proves nothing to me. Not until I see some *nix (or NT I guess) boot up and utilize both CPU's I'll just assume it's wishful thinking. Like setting your 500MHz celeron to 3 GHz and watching it post for about 3 seconds right before the chip burns :p

    Can anyone link to a better site? one that does actual so it working?
  • .. of people who paint up thier cps to pretend to be "other" cpu's. not my style.
  • by Anonymous Coward on Monday March 18, 2002 @06:52AM (#3180344)
    I have worked in AMD's R&D department for the last six years. I was pretty involved in development of the XP and MP series in particular. Let me make clear that modding XP chips to work as MPs is a bad idea and will most likely result in one or two cooked CPUs within a month.

    Yes, we know about the trace. We put it there to differentiate between the XP chips which contained our hacks to make it play nice in multiprocessing systems (aka the Athlon MP), and the chips not containing those hacks (aka the Athlon XP). Think of it like this: converting your car to diesel would give a great increase in fuel mileage, but it cannot be achieved by simply adding a "DIESEL FUEL ONLY" decal by the gas tank.

    I don't want to sound elitist and say "Leave the chip-mods to the pros", but sometimes chip mods have an amateurish feel to them. Please, don't wreck your systems. Leave the traces on the XP chips alone.

    (posting anonymously to protect my job)
    • do those "hacks" you mention actual warrant the 80 to 100+ dollars cost increase AMD puts on the MP series?
      • by Anonymous Coward
        Better question: does dual-cpu justify an 80 to 100+ dollar increase in price? If so, buy it. If not, don't. The market will determine the price - econ 101.
      • by Anonymous Coward
        If you fucking people keep penny-pinching AMD it will go out of business. Then bend over because all you'll have left are $800 Intel P4's rammed up your ass with the next new proprietary crap memory system and yet-another new CPU interface to kill off competition. FUCK INTEL. LONG LIVE AMD!
      • No, but the R&D that went into developing those hacks needs a return on investment. Because most people don't use MP systems it isn't fair to pass that return onto the normal customers. So the multi-processor customers pay a bit extra to finance multi-processor research at AMD.
    • by tempmpi ( 233132 ) on Monday March 18, 2002 @07:28AM (#3180401)
      Yes, we know about the trace. We put it there to differentiate between the XP chips which contained our hacks to make it play nice in multiprocessing systems (aka the Athlon MP), and the chips not containing those hacks (aka the Athlon XP).

      AMD already said that the only difference between Athlon MP and XP is that the MP cpus passed the multiprocessor test. I think a chip mod like that is more like over clocking: AMD doesn't guarantee that it will work, but there is a good chance that it will work.
      How many Athlon MPs does AMD sell compared to the Athlon XPs ? 0.1% ? Many more cpus should pass the test and they will be sold as XP cpus.
      • I'm not sure about this. I have an athlon XP 1700+, but if I ask my BIOS (Epox 8KHA+) whether the chip is MP capable or not, it replies "it is" and POST now says Athlon MP 1700+. I have never seen an Athlon MP 1700+ advertised so I must assume that many of the XP chips out there are MP capable, just like many chips can run much faster than the number printed on it says.
    • Nice troll, dude. Hats off to ya.

      People: If both MP and XP have the same core (very likely), they will perform equally well in the MP configuration.

      Fried CPU because of this? Don't make me laugh. You have worked for 6 years for AMD the same way I worked for NASA in the Apollo Project.

      duh.
      • and you make a good point I didn't catch earlier.... If what this guy is saying is true than the MP CPUs have a different core? Like the Celerons right? Umm, yeah I don't think so. I seem to recall the Duron core is the same too with some memory cells lasered? Why would the XP CPUs that were produced awhile back be SMP enabled and not the ones being produced now? It simply makes more sense that AMD has cut a trace in order to pull more money from MP labeled CPUs. As others have stated - businesses won't do this, only enthusiasts who won't give a fig for the warranty. AMD gets money fom the majority of their targetted audience with this and the enthusiest enjoys it too - smart! This same sort of thing happened with the Celeron until the bastards at Intel cut an internal trace on the CPUs. I'm STILL pissed about that! ocked multipliers? Intel can BITE ME.

        Benchmarks are what we need to see in order to confirm this. That will be icing on the cake and prove whether or not this is worth doing. A comparison with Duron and MP SMP setups running about the same speed on the same board would be VERY helpful. For that matter do it on a couple of boards so I know which performs the best - I want one NOW! ;-)
      • Fried CPU because of this? Don't make me laugh.

        That's exactly what I also thought. This is not overclocking.

        You have worked for 6 years for AMD the same way I worked for NASA in the Apollo Project.

        Maybe he's an AMD stockholder and tries to push up AMD's revenue ? ;-)

    • Somehow I don't think so. It might not work right, it might crash, but burn? Umm, how exactly is that going to happen? As for protecting your job - perhaps posting in something other than your actual name? You didn't actually use your name for your account did you?

      Seriously, I'm not trying to flame but claims that it will burn in a month or so sound awfully thin. Your going to have to either explain it a little better or point to some evidence of this occuring somewhere. Guess we'll know "in a month or so" one way or the other huh? BTW - all those neato' hacks made it into the Duron but not the XP?!

      One thing the article did NOT give and that I hope the next one will is BENCHMARKS. On the off chance that this guy isn't trolling benchmarks ought to tell us if both CPUs are actually being used by the OS and if those "hacks" are actually for real. Compare it to SMP Durons for instance - is it much faster? I'd also be interested in hearing more about what SMP MBs work best - I nearly bought the Tyan this weekend ;)
      • Weird shit happens when you start throwing multiprocessors made out of aluminum cans. It takes a very precise cache coherency protocol to make the whole system behave in a deterministic way. If one processor has a slight modification to the coherency protocol (say an extra delay cycle) that the other doesn't, then you're in for trouble. The chances might be one in a trillion that the differences in the protocol cause memory to become inconsistent, but consider the fact that millions of trials a second happen, it won't take long for one slight memory mistake (which will occur in the operating system during synchronization on a spinlock, most likely) to eventually bring the operating system to its knees. First it starts with two processors taking a spinlock simultaneously. Next they screw up an internal queue of processes, pages, or IO buffers. Next an IO buffer gets used by two processes simultaneously. Then suddenly they are sending two different devices corrupt data. Next thing you know some kernel driver somewhere is throwing a fit, and soon the kernel panics.

        Yes, having a processor, its memory, and its cache coherency protocol validated by the manufacturer is a good thing! It is not entirely overrated, and is quite crucial. Sure a hobbled-together SMP system might work for a couple days--until something funky sneaks into kernel memory because of an inconsistency in the coherency protocol.

        Now, about frying processors. You apparently aren't aware of the fact that nowadays the voltage settings of the processor are configured into the chip at manufacture time. They are chosen based on the clockspeed, intended heat dissipation, and overall, the requirements of the processor. Suppose you change the capabilities of the processor core without supplying the correct voltage. Suddenly the processor won't perform correctly because it doesn't have enough voltage to completely change internal states fast enough between cycles. Suddenly it has mysterious bugs.

        Or go the other way. Suppose you screw up and misconfigure the voltage, supplying too much. Power dissipation (and heat dissipation) are proportional to the square of the voltage. Suddenly a small increase in voltage results in a lot of waste heat. Heat that could fry the processor.

        Yes, bad things CAN AND DO happen when you start monkeying with those L2 bridges!

        • Quit fear mongering.

          Upping the voltage by a 10th or 20th of a volt, or overclocking by a bit, isn't going to kill anything. In fact, many motherboards do just that, they run various components with a bit more voltage to help stability. Sure, heat increases, but as long as it's withing spec you're fine. (Yes, some motherboards overclock by a bit. Check how fast a P4 CPU is running with CPUID, chances are it's not exactly at the rated speed.)

          Furthermore, CPUs don't need to be matched at all for doing MP with Athlons. At least one reviewer tested with mismatched CPUs when the chipset first came out and it worked just fine. (Contrast to the Intel SMP which sometimes needs chips of the same stepping.)
        • What you've said about cache and timings is spot on. However as others have pointed out AMD doesn't expect CPUs to be from the same stepping, lot, or anything else funky like you used to have to do with Intel. Hrm, does Intel still require this? It would seem that things have perhaps progressed? The systems don't appear to be quite so picky wouldn't you say? I'm also not yet sure that the MP CPUs undergoe any addtional testing over what the XP units do. Does anyone have hard evidence of this? An URL? I'd be interested in seeing the evidence as to what exactly is different. The fact that the original XP CPUs were shipped SMP enabled is still to be explained if this is the case. I'll grant that maybe MP CPUs get an extra test that the XPs don't but I'd also bet that many XPs would pass. Errors is one thing, burning quite another ;-)

          As for voltage and heat. Oddly enough I've yet to see a CPU modify the voltage when I've put it into the MB. Each processor line has it's own spec and the voltage DOES change to that level but not anything different WITHIN the CPU line. You make it sound as if each CPU is tested and a voltage assigned based on that testing - NOPE! Do XP and MP CPUs of the same speed\core require different voltages? I've honestly not checked but I'll bet they don't. If a CPU works fine at a particular voltage single I'm not convinced that it's going to need any more or less voltage in an SMP configuration. If anything I'd guess more based on what I've seen overclocking. Care to offer evidence?

          Lastly, as someone has already pointed out - an additional bit of voltage is NOT going to fry the CPU. This bozo is trying to say that simply running the CPU SMP is going to fry it, the more I think about this the more I think he's trolling. I've overclocked CPUs starting WAY back with 4mhz XTs and I've yet to have one fry as a result. I've bumped voltage, I've massivly bumped FSB, I've water cooled, Peltier cooled, and done all sorts of things to eek performance out. CPUs are actually pretty durable and simply tasking them differently ala SMP or even something intensive like Distributed.net isn't going to burn a CPU. Put a decent heatsink on it and use your brain - it'll be fine. Distributed.net's software bumps my CPU temps up a max of 10degrees - no biggie. How much would SMP operation bump the CPU temp? Not more than that I'm sure.

          Theory is all well and good with this stuff but it seems that folks are running the XP CPUs SMP without reporting problems and in the end that's all the proof needed....

          P.S. For some real fun measure the voltage fom your off-shore power supply with a good meter. It can get pretty bad! Not long ago I had the fan in a powersupply die on me - voltages across the board shot up. 12volts was pushing 14 and 5volts was closer to 5.4. Nothing fried but the added heat did indeed overwhelm the cooling on that CPU and it locked up after only a few minutes of run time. Slapped in a new fan and all was good - no damage. What you set in the BIOS ain't always what the rest of the system sees (shiver).
        • Weird shit happens when you start throwing multiprocessors made out of aluminum cans.



          true but read on ...



          It takes a very precise cache coherency protocol to make the whole system behave in a deterministic way. If one processor has a slight modification to the coherency protocol (say an extra delay cycle) that the other doesn't, then you're in for trouble. ...... Next an IO buffer gets used by two processes simultaneously. Then suddenly they are sending two different devices corrupt data.



          This is true on a slot-1 style (ie P2/3) bus where both chips are on the same bus and snooping each other's transactions - Athlon uses the Alpha EV6 bus which doesn't work that way. Instead each CPU has it's own private bus to the north bridge which is responsible for managing cache coherency between CPUs. The CPUs simply respond to the cache commands from the bridge.



          Even a single CPU (on a non-MP chip set) has to honor the cache coherency protocol ... because DMA transfers from PCI (but not AGP) are required to be cache coherent. This requires at the very least cache-line shootdown/flush support on all CPUs.



          Having said that it's quite possible that there are cache transactions used in a real MP system that aren't needed for PCI support and AMD screwed up on their first silicon and didn't support all the cache transations correctly (not having an MP bridge around to test it on and all that) which might require a chip spin and an 'MP' version - but bug fixes like that are going to be in all their future parts (as a chip designer I know you don't touch stuff once it works right :-)

    • First of all, people can and have been running Athlon XP chips in a dual configuration successfully for a while now. What has happened is they started cutting a trace.

      Ok big deal. Why would your CPU fry when it worked just fine before with the trace not cut? All you are doing is reconnecting that trace.

      Sheesh. Sad to see FUD coming from AMD... :(

      • possibly because the mp configuration uses
        more logic circuits in the chip.
        more active circuits at a given clock rate
        equals more heat ?
        and more strange RF effects.
        strange RF effects can possibly induce more heat
        situations by interacting with other chip
        real estate ?

        i'll be honest and say i'm wild guessing.
        but those of you who yell "its pure bull"
        have yet to show me your high-RF circuit
        design credentials...

    • I too have worked in the R&D department at AMD. I was in charge of the toilets on the 2nd floor, does that make me a chip-mod expert too?
    • Let me make clear that modding XP chips to work as MPs is a bad idea and will most likely result in one or two cooked CPUs within a month.

      I have yet to see a CPU overheating problem that cannot be solved with gratuitious amounts of copper and a few pet-inhaling fans, or liquid nitrogen if it comes to that.

      If you're trolling, you still have a lot to learn. If you really worked at AMD R&D, you have a lot to explain.
  • What with a new version of SuSE and now this.
    Looks like it's time for a new rig. I'll get:
    • Two XPs
    • SuSE 8.0
    • And now floppy drive [slashdot.org]
    • Clear [clearpc.ca] case
    • Perhaps a liquid cooling system.
      What would be really cool is if you had clear tubes
      and a special die in the water that changed colors by temperature.
    • I'm definitely getting those drive windows. [slashdot.org]
    • What would be really cool is if you had clear tubes and a special die in the water that changed colors by temperature
      The Cray 2 had a fancy see-through waterfall cooling unit [tmeg.com], you can see a picture [cray.com] at Cray's site too. The cooling liquid is fluorinert, which (IIRC) is used as a substitute for human plasma in surgery. The whole circuitry is immersed in the liquid (which is non-conductive); there's non of the heatsink and radiator business that happens with watercooled CPUs.

      Every now and then a used Cray turns up on Ebay, so it might be worth a look. It'd save on heating bills, and make a good conversation piece, too!

    • Yes - please do this happend. I dont want a freeken FAN on my computer - software alone is enought of a problem. Silent cooling should be a no brainer. -Wrong spelling ? as a foreigner who cares !
  • by Andorion ( 526481 ) on Monday March 18, 2002 @06:58AM (#3180353)
    This could come in very useful when you're trying to MP-enable your chip.... or maybe you can unlock the multiplier and MP enable it at the same time? =) BWAHAHA.... Where else, but Tom's [tomshardware.com]? =)

    -Berj
  • I dont see how this is much diffrent
  • by nookieman ( 548184 ) <kielNO@SPAMnookie.dk> on Monday March 18, 2002 @07:24AM (#3180394)
    You want me to put make-up on my CPU? What's next? a pink leotard and a tutu?
    • by Anonymous Coward
      Does that mean MP is simply a XP in drag?

      *ducks*
      • It has always been this way.

        MP is XP with more testings.

        Just like Xeons to the P4s, the core is the same, but Intel does more testing to the CPU and varous power protection (which is why there are more pins).
        • Re:Tarted up CPU (Score:2, Informative)

          by afidel ( 530433 )
          Actually the Xeons also have from 2-8X the L2/3 cache when compared to the P4. This makes a huge difference in some apps where dataset size is more important than raw crunching power. In fact on some database app's the old pentium pro's with 2MB l2 cache were faster than a normal p3's at 550 even though the ppro's were running at a slothlike 166MHz.
  • I'm not touching it! (Score:2, Interesting)

    by mnordstr ( 472213 )
    The last thing you want to do is play around with your XPs! They are the most fragile pices of anything I've ever seen. If you however decide to play around with them, remember that the processor die is like a small ant, put a little pressure on it and it will die. One has to be really careful when attaching the heat sink. And never ever try to run it even for a fraction of a second without a heat sink and a fan, it will die immediately, probably destroying your motherboard.

    Let this be a warning!
    Be careful when playing around with an Athlon XP!
  • Here's How (Score:2, Informative)

    by the_radix ( 454343 )

    Since the article is more or less slashdotted, here is how to modify the chip to work on a multiprocessor board:

    Complete the topmost L5 bridge if it is cut. They use conductive paint. Now the XP is detected as an MP for the multiprocessor boards.

    Of course, the stability of this is questionable, and no one except yourself should take responsibility for it.

    • Or does it still post as an XP? If it comes up as an MP then this is all the more telling IMO. I think this is a great hack and as soon as SMP AMD boards become a little more common I predict we'll see lot's of SMP AMD systems running around.

      I've just about flushed all of the Intel systems off my home network and replaced them with AMD. A couple of old dualie Celeron are left but if the AMD stuff gets cheap enough those will go too. Sorry Intel but you guys have pissed me off with this PIV crap, the locked multipliers, RAMBUS, and the mods you hacked into the Celeron to prevent SMP operation. AMD is doing the enthusiest right as well as the consumer with their pricing. Go AMD!
  • Being as unpopular as they are, first because they're not in the wireless handheld mainstream and second because they're general purpose, there's very little accurate information on exactly how to make an Athlon XP run in dual mode. Some web pages say the bios must be reconfigured. Some say the bridge must be joined. Regardless of what it was, you can be sure it won't work next year and you'll have to pay a lot more for dual cpus, even if there still is such a thing next year.

    • by BLKMGK ( 34057 )
      Umm, Intel is in the wireless mainstream? What's that got to do with AMD XP CPUs?! You lost me there. Yeah, this is a general purpose CPU and not a mobile version - you know foks hacking mobile CPUs? Do tell, I'm al lears as to what's being done.

      A BIOS reconfigure? That actually makes SOME sense. Much like the rewiring some of us had to do at first with the Celeron it's possible a BIOS hack could be done to get the MB to ignore the cut bridge and just assume SMP operation. The ABIT dual Celeron board was sort of like this in that it ignored the fact that the Celeron wasn't supposed to SMP. I wouldn't be surprised if they could do this again if they chose with the new CPUs. Maybe some of the MBs are allowing the BIOS to override the MP\XP switch? Do you have any URLs?

      As for them stopping this - you can still unlock the multipliers right? To prevent that AMD would have to seriously change their internal processes and it would cost them more in cash and goodwill than it would to simply allow it. They're not losing much money if any by allowing hobbyists to do this. Start a business modifying\testing overclocked CPUs and they would probably get pretty upset - and have in the past - but I think this will be here awhile. Actually it's a shame Intel doesn't do their CPUs this way...
  • at weekends I make-up as an MP.
  • How come they assume all their readership is male?

    Read on dudes...

    ??
  • Sorry...someone needs to write up the musical version and call it Paint Your Athlon.

    Seriously...if they work reliably, great. (I just want to know when another chipset maker will come out with a multiprocessor-capable chipset for Athlon; I recall reading that AMD doesn't really want to be in the chipset business...)

  • "Paint Youself An Athalon MP"
    I thought this story was going to give me an idea for a costume party!!
  • Not a smart move.... (Score:2, Interesting)

    by Anonymous Coward
    This move will NOT make AMD more money.

    The real source of MP chip sales is definitely system vendors, OEMs, etc., bigger organizations relying business on their "guarantee of MP operation", and can't take risk on individual chips; of course you might add the occasional enthusiast with more cash to throw into his/her system.

    For normal individuals, however, SMP itself will probably not be enough to make them automatically buy such systems; with a highly probable chance that "normal" XPs could do the same, the scenario is entirely different...

    For every such dual-XP system sold, AMD sells 1 extra XP processor, plus a sale of an AMD chipset motherboard. That's grabbing business from VIA, and together with the extra processor, are all sales that would NOT have gone to AMD on AthlonMPs

    That's just my theory though....
    • There is no way that big businesses, which are the MAIN customers for MP's, would do this. Doing stuff that could easily break your equipment just isn't good business practice.
  • by Animats ( 122034 ) on Monday March 18, 2002 @12:51PM (#3181713) Homepage
    I'd be a lot more impressed with these guys if their article said something like "and then we ran the multiprocessor cache consistency test for 24 hours with no errors". They haven't demonstrated that the two processors play well together.

    IC speed is generally a part selection - all the parts are made the same, then tested and sorted by how fast they'll go. The slowest gate on the chip determines the speed limit. That's why testing matters; you may have 20 million gates that work at a high speed, and one that doesn't.

    AMD's own CPU price list [amd.com] is interesting. The fastest MP is the 2000+, while the fastest XP is the 2100. Does this indicate a speed penalty for enabling multiprocessor mode? It may.

    The faster processors cost far more than the slower ones, and it's unlikely that they really have a severe yield problem that requires this. It's unusual today for a fab to produce large numbers of substandard parts. Today, yields below 80% indicate serious problems in the fab. Some fabs report 99.5% yields. A decade ago, variability in the fab was much higher; a whole range of speeds came out, along with a sizable percentage of rejects. Today, the processes are much more uniform.

    (Think about what this means. Sizable wafers are being produced with almost all the atoms where the design says they're supposed to be. That's an achievement. And then these bozos slap conductive paint on the pins and think they've done something cool.)

    • The fastest MP is the 2000+, while the fastest XP is the 2100. Does this indicate a speed penalty for enabling multiprocessor mode? It may.

      More than likely, they sell hundreds of XP's per one MP that they sell, so it does not matter that the MP is "released" later. Besides, all they are doing is overclocking the current architecture. It is not as if every new step in speed is a newly designed chip. It is the same chip, it is just a matter of how fast you want, and how much cash you've got. This is why it is so easy to overclock chips that are "slower" or earlier in a release cycle, whereas the chips at the end (the fastest PIII's for example) are much more difficult to overclock.
  • Will the #2 pencil trick work for reconnecting the bridge? It worked on Thunderbird-core Athlons for the L1 bridges, doesn't work on Palamino-core Athlons for those bridges, but what about this bridge (L5)?

    Background: L1 is cut to lock the clock multiplier. Reconnecting L1 lets you set whatever multiplier you want. Graphite from a #2 0.5mm mechanical pencil is just conductive enough to work. I did this successfully to two Athlon CPUs. They used a deeper cut on the Athlon XPs, so I decided to leave my XP 1800+ locked.
    • No. In the same manner that you can't use graphite to fix up the L1 on XP to overclock them, it won't work for the L5s either.

      For whatever reason (protecting the amateur overclocker from himself, maybe?) AMD set up the bridges on XP chips to be unlinkable without something that has very low resistance (like silver paint). Graphite won't cut it anymore.
  • I've been running to Athalon XP 1800+ 's in the dual proc board from Asus. My machine boots into WinXP Pro just fine and definately takes advantage of the dual processors (been watching cpu time and usage in the task manager) especially in 3D Studio. I just popped the chips in and the bios saw them as MPs. Does anyone know whats the min speed of the chips (1800, 1900, 2000?) that requires this mod?

    BTW... I haven't had any system instability related to the dual XPs. I had serious crashes caused by the Creative drivers/spyware, and I've had problems with power (the computer drawing more than the UPS can handle and tripping it). The most important thing w/ the dual proc athalon boards is to make sure you have a 450 watt power supply (and suitable UPS) or the system will crash.
  • I run an overclocked Celeron 300 at 450 MHz with great success for well over a year. I've had no problems with stability and it seems to handle all of my software.

    I'm not one who is timid about taking a measured risk in the pursuit of performance. My home server is a dog running at 200MHz. It may be a good candidate for replacement with a dual AMD server. I do want to be sure that it would be stable and run Linux, however.

    I have never run AMD before so I'm interested in hearing any pointers both on the hardware side and on the OS side. ie Does Linux handle AMD straight out of the box or do I need to compile a special kernel?

Dreams are free, but you get soaked on the connect time.

Working...