Sign in with
Sign up | Sign in

Phenom II: Unlocking Cores, Cache, And A Free Lunch

Phenom II: Unlocking Cores, Cache, And A Free Lunch
By

Every so often, a story breaks that goes a little something like this: manufacturer x, known for its flagship product y, recently started offering a mid-range product z based on silicon that wouldn’t bin up to y. Rather than throw the entire die out, x disables one part and sells it off as z—something less expensive.

Much of the time—it’d be disingenuous of me to guess how often, exactly—this happens because part of the die is really defective. But sometimes a vendor simply needs to fill in a price gap in its lineup where a competitor is eating its lunch. ATI’s Radeon HD 4830. Nvidia’s GeForce GTX 260. AMD’s Phenom II X3 700- and 800-series. These are all parts based on more expensive components.

Ah, AMD’s Phenom II.

A retail X3 we purchased to test the unlock techniqueA retail X3 we purchased to test the unlock technique

A couple of months back, a report came from Korea claiming that you could unlock the fourth core on a Phenom II X3 processor by simply enabling Advanced Clock Calibration on certain motherboards, in essence, turning a $135 Phenom II X3 720 Black Edition into a $190 Phenom II X4 920 (Black Edition) with a simple BIOS switch. We’ll skip over the obvious question for now: “Is saving $55 worth a potentially-unstable system?”

If you remember back to our Phenom II launch piece, AMD claimed that ACC technically did nothing for Phenom II, as the technology’s benefits—which were supposed to help Phenom achieve higher overclocks—were already baked into Phenom II. Now it seemed that the SB750-based feature was having some other effect on Phenom IIs as well.

Why this worked remains a mystery. Two motherboard vendors—ASRock and Biostar—claim to support the ability to unlock X3 CPUs, but neither company is able to divulge exactly how they’re doing it. To be fair to both companies, this is less likely to be trickery on their part, and the closest we've come to an explaination goes a little something like "ACC is able to adjust the error checking process of the CPU, enhancing its error tolerance and avoiding the potential for collapse under certain circumstances." As the theory goes, increasing the error tolerance with ACC makes it possible to "revive" the disabled silicon.

A retail X4 we purchased to test the unlock techniqueA retail X4 we purchased to test the unlock technique

A product manager from Biostar says that any board with an SB750 southbridge can technically achieve the unlock. ASRock confirms the same, and gave us its test data with a number of CPUs indicating a fairly even split between chips that will unlock and run stably, chips that will unlock but aren’t stable, and chips that won’t unlock at all.

Stop That

Here’s the interesting part. AMD claims that the noise made over this ACC/Phenom II X3 phenomenon is doing great for the sales of a product family that we’ve always thought was just a little weird. After all, when the first X3s launched, there were applications out there that didn’t quite know what to do with three cores. To that end, the company says it has no plans to try stopping enthusiasts from unlocking the X3’s fourth core by striking ACC from SB750.

But one motherboard vendor claims that AMD is going to discontinue ACC in its SB750 starting this month. Another board vendor says that it is getting pressure from AMD to update the ACC embedded controller firmware immediately and post it to an “updated” BIOS. Though the first guess might be conjecture, the second is most definitely true. There’s no such thing as a free lunch, so they say. Want proof? Keep reading.

Display 65 Comments.
This thread is closed for comments
Top Comments
  • 13 Hide
    cangelini , April 16, 2009 7:15 AM
    Well, but the point isn't the benchmarks. We already know that most games are going to be limited more by graphics horsepower versus whether a CPU has three or four cores/4MB shared L3 or 6MB shared L3.

    In fact, when it comes to gaming, you're going to be better off looking for the fastest overclock possible with your three good cores or 4MB of known-good cache, really.
Other Comments
  • 7 Hide
    dirtmountain , April 16, 2009 6:49 AM
    Good article. Now if you can just get your ad clowns from sticking us with those annoying ads.....!
  • 2 Hide
    Anonymous , April 16, 2009 7:09 AM
    test them with games... some people care about that :p 
  • 13 Hide
    cangelini , April 16, 2009 7:15 AM
    Well, but the point isn't the benchmarks. We already know that most games are going to be limited more by graphics horsepower versus whether a CPU has three or four cores/4MB shared L3 or 6MB shared L3.

    In fact, when it comes to gaming, you're going to be better off looking for the fastest overclock possible with your three good cores or 4MB of known-good cache, really.
  • 1 Hide
    Anonymous , April 16, 2009 7:28 AM
    omg you replied to me... i'm so honoured :p 

    but yes, i agree... but if you had crossfire gpus, this would make a difference. but then again, i think you'd have the money to buy the real thing (phenom II 920)
  • 5 Hide
    tacoslave , April 16, 2009 8:13 AM
    i like this article toms should do more stuff like this.
  • 2 Hide
    cangelini , April 16, 2009 8:26 AM
    Hey Apache! No error, sans = without.
  • 2 Hide
    Summer Leigh Castle , April 16, 2009 9:05 AM
    Good article but can someone explain how ACC % works? Also, where do we start in terms of ACC % if we're tweaking for stability?
  • -1 Hide
    ravenware , April 16, 2009 9:22 AM
    cangeliniHey Apache! No error, sans = without.


    The only reason I know that is because of Wayne's World2.
  • 0 Hide
    cangelini , April 16, 2009 9:23 AM
    I do wish I could give you more detail on ACC, but AMD has played that card close to its chest. In terms of where to start, I'd say "Auto" is your best bet, and then move up and down in 2% increments in each direction.
  • 0 Hide
    raden_muaz , April 16, 2009 11:09 AM
    can you compare it with PII 910? PII 810 have higher clock speed.
  • 9 Hide
    Slobogob , April 16, 2009 11:10 AM
    Great article. Reminds me of how Tomshardware used to be.
  • 3 Hide
    empstar , April 16, 2009 11:59 AM
    this reminds me of the 1st Athlon with gold finger and the athlon with the "tape" on top to link connection for Socket A. 1998 I start reading this web..... :-) where is Dr. Toms P go ? I wonder..
  • 0 Hide
    apache_lives , April 16, 2009 12:08 PM
    tom sold it ;) 
  • 0 Hide
    empstar , April 16, 2009 12:19 PM
    sold it to Bestofmedia? omg since when?
  • 0 Hide
    Pei-chen , April 16, 2009 12:24 PM
    AMD is not overclocker friendly by locking the multiplier, cores and cache. What happened to the AMD in Athlon and Athlon XP era?
  • 2 Hide
    apache_lives , April 16, 2009 12:55 PM
    Pei-chenAMD is not overclocker friendly by locking the multiplier, cores and cache. What happened to the AMD in Athlon and Athlon XP era?


    its called sales - if your $100 processor was the same as your $500 processor why would you buy the $500 processor?
  • 1 Hide
    neiroatopelcc , April 16, 2009 1:31 PM
    I supposed it wouldn't be impossible to custom tailor a bios for a non asrock board. I haven't dared try (nor needed to), but I expect it to be quite likely you could just extract the microcode from an older, working, asrock bios, and replace the code in your gigabyte, asus or whatever bios with it. All I think that is required, apart from knowledge on bios tinkering I don't have, is a 750 chip on the board.
  • -1 Hide
    apache_lives , April 16, 2009 1:38 PM
    i believe it depends partly on the IO chip too (usually one of those ITE 87xx chips), then flash part type, rom size, bios brand (award, AMI etc) - beyond most of us and more for the motherboard engineers etc.
Display more comments