What does AMD's new Bridge cutting mean for Overclockers?

[BigJ]

In case it was not known, here is the info, ripped directly from HardOCP.com:



<< Interesting situation was explained to me today. Tristan Smith bought an AthlonXP 2000+ from Newegg and got one that was a bit different when it comes to the way the L1s are cut. Instead of the trace spanning between the connecting points being severed, the inner-most connects had been removed as shown below in this graphic. Tristan did not have the proper equipment to get us a photograph but he is working on it. Hopefully this is a "goof up" rather than a change. AMD has yet to reply to our weekend emails on this issue. >>

Picture Here

Also, here is a picture of what is explained above on an AMD chip:
http://www.hardocp.com/new_img_02/feb/amdcontact1.jpg



<< Here is the most recent information from HardOCP.com
Tristan Smith dropped us a heads up about his AthlonXP that is actually missing the L1 contact points instead of simply being cut as compared in the above graphic. We asked for your feedback on this and have some good news and some bad news.

The bad news is that this is not an isolated incident as we have had a few folks report in that they are seeing the same thing. The good news is that there have been less than 10 of those emails. We are working to get actual photos now and should have them in a couple of hours hopefully.

If AMD is moving over to "hard-locking" their CPUs (And of course we do not know that this contact point removal will cause that.) they could not have picked a worse time with the low end P4s scaling so well. More news on this as it happens...
>>



But what I'd like to know, what does this mean to us? A question to anyone who knows alot about AMD chips: would this definitely "hard lock" the multiplier on the chip, and would there be any way to still properly unlock the multiplier? For anyone here at Anandtech that has these chips, have you been able to successfully unlock them? Like HardOCP said, this could not have come at a worse time, especially when many enthusiasts and hobbyists have seen the amazing overclocks of the low cost P4 Northwoods.

 

[MadRat]

Looks like it is no different than the old way, just slightly offline. Since these are all cut by a machine, a simple misalignment of the laser may have done it.

 

[johncar]

There's an easy way around this problem, simply fill the L1 cavitiies as you do now...but then connect the top of the L1 bridges to one bridge of each of the 5 L3/L4/L10 "bridge pairs". Go to
http://www.beachlink.com/candjac/index.htm Final Decoding Palomino bridges article for circuit diagram/s which will make this all very clear.

The 5 L1 bridges are connected to the 5 L3/L4/L10 bridge pairs when the "normally cut" L1 bridges are closed. And since the 2 bridges of each bridge pair are connected to the lower end of their L1 bridges, then the top ends of both bridges of each bridge pair must be connected to each other. Check it with multimeter if you don't believe it. You simply bypass the "inaccessable" bottom end of the L1 bridges.

Detail....If we label the L1 bridges L to R = A thru E, and label the L3/L4/L10 bridges L to R = 1 thru 10...then simply connect...
A to 1 which also connects to its L3 bridge pair 3.

B to 2 which also connects to its L3 bridge pair 4.

C to 5 which also connects to its L4 bridge pair 7.

D to 6 which also connects to its L4 bridge pair 8.

E to 9 or 10 of L10 bridge pair.

There will be no "crossovers" of the new slightly longer traces. But be careful not to connect to any printing or "dots" on the surface as these are reported to be at "Ground Level" electrically.
John C.

 

[BigJ]

Johncar, your solution is ingenious and I recently sent it over to HardOCP.com. I think you should get in contact with them over there, because they are very interested in the solution you have proposed. Erroneously, I was credited with the solution, but I have sent them an e-mail to make sure, you, Johncar, are given credit for this idea.

Edit: Get in contact with them ASAP and fill in the rest of the community, twas a brilliant idea. You can send them an email here : hardnews@hardocp.com

 

[Scootin159]

<< Johncar, your solution is ingenious and I recently sent it over to HardOCP.com. I think you should get in contact with them over there, because they are very interested in the solution you have proposed. Erroneously, I was credited with the solution, but I have sent them an e-mail to make sure, you, Johncar, are given credit for this idea.

Edit: Get in contact with them ASAP and fill in the rest of the community, twas a brilliant idea. You can send them an email here : hardnews@hardocp.com >>



Looks like they liked your idea, it's posted front page now.

 

[johncar]

BigJ....
Thanks for the kudo, but this was not rocket science...just a bunch of Direct Current Resistor Logic circuits one step removed from a battery flashhlight circuit. We just believe in decoding and tracing these simple circuits so we can "understand" them...and then share that understanding with the OC community. The sequence of articles from Slot A thru Duron/Tbird and Palomino are proof of that.

Anyone could have come up with this simple fix given just the circuit diagram of the L1/L3/L4/L10 bridges. Problem is that not too many people know of our work...they are quite content with "cookbook" solutions that layout bridge settings when you enter a Multiplier.
Cookbook solutions are OK and they have their place, but it's like giving a man a fish...he's fed for just a day. But if you teach him to fish he's fed for the rest of his life, and best of all he can teach others to fish. Point is...we'd like to see others beat us to the solution by "understanding" the circuits. Can you somehow get the word out about our AMD Multiplier etc pages...we're not as stylish as most sites, but haven't seen one yet that beats us on substance.
Thanks again, Johncar

 

[Duvie]

In most instances with 333ddr boards coming out I think will see more fsb ocing anyways...fsb is a better performing oc versus multiplier...untilk thoroughbreds arrive we really wont see were the ceiling is going to be anyways and who knows maybe most can only achieve 200-300mhz anyways and that could be done with very little fsb boost. The multipliers are getting high anyways and so many boards are offering a wide array of ratio and dividers I think multiplier lock is no big deal....

I got a 600mhz oc so far off of a locked p4 northwood...

 

[ScrapSilicon]

kudos johncar !!!! you have an excellent tech site.

 

[johndoe52]

Damn johncar...that's awesome. When first read about this different cut I knew it would only take time before someone figured out a different way. I didn' think it would only take one day though!!! kudos

 

[blackhawk]

The overclockers will be switching back to intel over the next year as it gets harder to get free & stable improvements on the XP's

The northwoods and even many tualitans are going 100 to 133 which is bringing the price down for the end product. The pendulum swings again just hope they both stay healthy for what the competition brings us.

 

[FelixDeCat]

...yet another conspiracy for geeks to ponder...

I dont think this is the future.

 

[Lecho]

AMD is slowly cornering the enthsiast market. Intel hardlocked their CPUs for a reason, and AMD will do it soon, posibly for the Thoroughbreds. It's not as if they are doing this to give overclockers more adventure.

 

[Scootin159]

<< AMD is slowly cornering the enthsiast market. Intel hardlocked their CPUs for a reason, and AMD will do it soon, posibly for the Thoroughbreds. It's not as if they are doing this to give overclockers more adventure. >>



Biggest problem I see is that the unlocked multiplier is what is bringing AMD all their fans.

 

[hoihtah]

more work. but still same deal.