2500+ unlocking effort, strictly directions

2500+ unlocking effort, post results and possible solutions

please keep this thread clear for unlocking ideas, results, and progress. the "last week 2500+ unlocked" thread is up to 15+ pages, and is getting unwieldy. hats off to all who've gotten their hands dirty so far.

chip=die=silicon square in center of processor
package=printed circuit board with pins the chip is mounted upon
bridges (L1,L2,...L12)=conductors which by their presence or absense (blown) determine a finished processors speed, cache, working voltage, etc.

so.... regarding multi select and cache select blown bridges:

its far more likely that the package was changed, since a change to the chip means new mask sets, different metalization steps, etc, in short, huge process changes.

changing the packaging (essentially a very fine pitch multi layer printed circuit board) is much easier to do.

the chips are tested before packaging for known good die, and perhaps for performance. a fixed multi on die would mean different mask sets for each clock speed, which is just not done.

there is a very remote possibility that the bridges to be blown are on the die, but that would mean they've been there unused since moving to .13um process or the last die circuit update in the .13um process. if this is the case, it would be on one of the die interconnet layers, of which there are 7 or 8 (memory fails me) for this chip.

the reason it is very unlikely that the blow bridges are on the die is that the interconnect layers use copper wiring with an isolation layer to prevent the copper from migrating (dual damascene process). copper will migrate into silicon and pollute it, causing the chip to fail. if the bridges were in this copper layer, blowing them would allow uncontrolled migration of copper. so it really can't be done this way.

the most likely scenario IMHO was suggested earlier, which is using formerly NC pins to blow an internal PCB (package, not die) bridge electrically. this is done with many chips and packages, such as (for chips) programmable read only memory (PROM), asics (application specific ic's), etc.

x-ray might work, but you're resolving feature as small as .2mm (200um) on that pcb, and with all the layers, it will look like a plate of spagetti.

i think hoot made what is probably the only workable suggestion earlier. would need 2 pcb modules with cores removed. each pcb would need to be sanded layer by layer. as each layer is uncovered, take a high resolution close up picture. compare each layer from each pcb by doing a photoedit boolean subtract. the difference will pop right out if the layers being compared are registered very well.

the folks looking at pin electrical characteristics have a problem with numbers of possible combinations. if there are 10 NC pins, they may have to be checked against Vss, Vdd and perhaps active pins. unless i'm wrong, this could be hundreds of resistance checks. not my cup of tea.

but, if and when differences are found on layers, that's the time to follow up with the DMM.

edit:

this circuit describes the concept well, if only in one possible implementation.

The original idea for this comes from darkman101010; this link is to the original picture. Thanks darkman.

http://www.overclockers.at/attachment.php?attachmentid=40885

Link to the original thread of darkman101010's post, in german:

http://www.overclockers.at/showthread.php?s=&threadid=97670&perpage=15&pagenumber=5

UnLoadeD has provided the hosting below, which seems to be faster.



cheers,
phaed

Has anyone tried mucking with the "fake" L3 dots? Seems a semi-obvious place to start

edit: Link added from http://www.emboss.co.nz/amdunlock

hitechjb1 said:

Just to make sure what you (deathstar13) measured:

To VSS (AM28):
L3 pin1 277 ohm (this L3 is cut, Thortron 2000)
L3 pin2 278 ohm (this L3 is cut, Thortron 2000)
L3 pin3 259 ohm
L3 pin4 261 ohm
L3 pin5 258 ohm

To VCC (AM26):
L3 pin1 283 ohm
L3 pin2 283 ohm
L3 pin3 264 ohm
L3 pin4 266 ohm
L3 pin5 264 ohm


L3 pin1 to AN27 0.8 ohm
L3 pin2 to AL27 0 ohm
L3 pin3 to AN25 0.8 ohm
L3 pin4 to AL25 0 ohm
L3 pin5 to AJ27 0.8 ohm

Pls confirm.

Click to expand...

Deathstar13 confirmed the measurements in a post in another thread.

From the other thread, which I started reading last night for the first time:

I've been following this thread with relish. There's nothing like a good challenge and in that spirit, I just went out for lunch and intentionally bought an AQYFA 0342 SPMW for $89.50 from a local dealer. All their OEMs were that stepping and their boxed units were AQXEA 0331. I have it in the lab here, under the microscope. The L1 traces indeed go to the L3 traces and the Pull down resistors up on top, verified with an ohmeter. I'm lucky to have access to top notch equipment and resources here, though no xray machine.

The L1 traces do indeed go to the BP_FID pins on the bottom, but then that was covered in the other thread. Just wanted to give a confirming observation.

Interesting to note that if you reverse the polarity of the leads on an ohmeter across the L3 1k ohm pull-down resistors, the resistance is different (930 vs 862 ohms). In case you don't get what that means, the traces must go to a semiconductor device, like the input gates to the Multiplier Control decoder. This top layer is not just a phantom layer.

If you subscribe to the Fab51 theory of design, then something is amiss. Examing the BP_FID pins with an ohmeter as they relate to vcore, there is no sign of the ~200 ohm internal pull-up resistors, assuming they are hard-wired to the vcore bus inside. That was an assumption from Fab51, not asserted as a fact. When I get home, before I put this in my machine, I'll examine my Barton 2600 and see how it compares ohmically <sp>.

Hoot

Putting things into perspective, currently for system with NF7-S rev 2.0 with CPU on air cooling, the impact of locked 2500+ is minimal, at most 3% from best overclocking (2.5 GHz), if the motherboard and memory can do 220 MHz.

If aiming for super FSB to 240-250 MHz for 3Dmark benchmark, then the CPU is limiting the FSB due to the fixed 11x multiplier.

If motherboard or memory is limiting at the low 200-210 MHz, then the FSB is limiting the CPU to 2200-2310 MHz.


Impact of Barton 2500+ w/ locked muliplier on overclocking

If there is no workaround for the locked Barton, these would be the scenarios:

1. It would mostly affect users with KT266A, KT333A motherboads, since those boards can achive max FSB around 150 MHz and 190 MHz respectively, due to 4:1 and 5:1 PCI lock, .... As a result, a Barton 2500+ with 11x multiplier, the max CPU overclocking frequency would be 1650 MHz and 2090 MHz respective. These number are way below the norm of 2.2 - 2.4 GHz for Barton 2500+.

2. It would also affect users using extreme coolings. Assuming they are using nforce2 which can do 220 MHz FSB on the average, and as high as 240+ if using enough Vdd, Vimm and chipset cooling, and "good" RAM, ... So the Barton 2500+ in these systems with 11x multipliers would be limited to overclocked frequency of 2420 MHz (220 FSB) to 2640 MHz (240 FSB). Not 2.7 - 2.9 GHz as hoped for. One would need a Barton with 13x multiplier to achieve 2860 MHz running 220 FSB.

3. For air cooling using a good nforce2 motherboard such as NF7-S rev 2.0, whose FSB averages around 220 MHz +- 10 MHz. The Barton 2500+ should still be able to run at 220 x 11 = 2420 MHz, which is above the norm 2.3 GHz for Barton 2500+. At FSB 230 MHz, which is doable, the Barton would then run at 2530 MHz.

Now the burden becomes putting more demand on the FSB, memory, and how to tune/optimize the FSB reaching 220 - 230 MHz, so that the motherboard, good memory (modules with WinBond CH5/BH5 chips) and FSB are not holding back the Barton 2500+ due to the fixed multiplier 11. If one can only achieve FSB 210 MHz, the Barton would be running at 2320 MHz, which is still above the PR rating of a 3200+ AMD processor.

To achieve 2.3 - 2.5 GHz speed for a 2500+ on air, a copper HSF such as SLK-800/900/947U and a high CFM adjustable fan such as TT SFII would be the choice. It is doable with a TT SFII fan running 3000-3200 rpm at which the noise should be acceptable for 24/7 run. Barton 2500+ at 2.3 - 2.5 GHz can run few degree C cooler and less power than a Tbred B 1700+/1800+ DLT3C delivering the same overall performance.

So I would say at multiplier 11x, currently with high end air cooling, for system w/ nforce2 rev 2.0 board, there is not much an impact for 24/7 usage:
- The impact is at most 12% (2200 MHz at stock FSB 200 MHz) on CPU frequency from the best Barton 2500+ overclocking on air (assuming 2500 MHz). Even in this worst case situation, it is already running as a 3200+ CPU.
- An average system at FSB 220 MHz can run Barton at 2420 MHz (about 3% off the best Barton at 2500 MHz on air).


ADD:

Currently, for those who use air cooling and have NF7-S rev 2.0 and good memory that can run at 220 MHz, there is not much an impact. Because 220 x 11 = 2420 MHz (3% from best air of 2.5 GHz assumed).

For those who run KT266A, KT333A motherboard, it is time for new motherboards if wanting to run a locked 2500+.

For people running extreme cooling, better get CPU with higher PR rating/multiplier.

The burden is on high FSB and good memory, so the CPU can be max out.

Actually, Barton 2500+ multiplier 11 (even it is locked) is pretty nice, it fits perfectly with NF7-S rev 2.0 with good 3200/3500/3700 memory using CH5/BH5 chips.

Even with locked Barton 2500+, it will get one to 220 x 11 = 2420 MHz (with good air cooling such as SLK-800U/900U/947U), a perfect default overclocking setup.


As long as there are:
- NF7-S rev 2.0
- Winbond BH5/CH5
- SLK-800/900/947 HS
- Barton 2500+

I would not worry about getting a Barton 2500+, whether it is locked or not, for 24/7 usage (for competition to get the last % of everything is a different story).

This top layer is not just a phantom layer.

Click to expand...

hmm ok well im gonna do some more pin modding with my thorton and try the top "so called" fake layer.

at most i want my xtra 25k L2 cache enabled,even if i cant get the multis.

Has anyone heard of someone cutting one of the L3 traces on a locked unit to see if the multiplier changed?

Hoot

Any active devices such as transistors, diodes, CPU VCC-VSS, ... are non-linear devices,
- The resistance between any terminal is NOT a constant.
- The resistance measured depends on the setting of the ohm meter, e.g. x1, x10, x100, ...
- The resistance measured also depends on the polarity of the ohm meter which is itself an ampmeter (current sensor) in series with a battery. That is the ohm meter will apply a voltage/current to the two terminal of the device measured. As a result, depending on the biasing point, the resistance VARIES.

So if attempting to use an ohm meter to measure the pullup and pulldown resistance of the AMD CPU mulitplier resistor newtork, one would not be able to measure exactly the resistance value in the network.


I was able to measure, as shown in a post in another thread (reposted in this thread), that the pulldown resistance of the L3 pin from a Tbred B 1800+ in OLD PCB to be 1000 ohm.

There is a current path from any CPU between VCC and VSS, estimated to be way under 10 ohm and non-linear (I measured and calculated). So any measurement of pullup resistor whose two terminals have a path to other part of the chips would result in much lower resistance. Similarly for pulldown resistors.

I measured the resistance for the other L3 pins to VCC/VSS to be about 70 ohms for the Tbred B with old PBC. (see previous post)

But deathstar13 measured the corresponding L3 pins to VCC/VSS to be about 200+ ohms. (see previous post)

Since the two measurements are done with very different ohm meter, and since the pins are connected to non-linear devices, the two number though is very different, but is NOT conclusive yet (need some more interpretation and thinking), unless one can use the same ohm meter to measure two CPU with old PCB and new PCB (locked ones).

That's exactly what I am about to do. I just got home with the new locked (assumed) Barton and as soon as I finish this post, I'll power down my machine and measure the Barton 2600 I have in it. I use a Fluke 189 DMM. In ohms mode it doesn't provide a lot of gating current like old style analog VOMs do. After I A/B the two chips and put the new locked one in my machine, I'll report back what I found.

Hoot

guys ive redid my L3's and L2 again for like the 4th time now.
unsuccesfull on both accounts.even scraped the top pcb's pins and connected those with no luck.

hoot i dont know if cutting a bridge will make any differance as connecting sure dont.

If possible, measure for both chips w/ old and new package PCB w/ the same DMM.

1. Measure the resistance between VSS and VCC, both polarity. Should be very small (< 10 ohm non-linear resistance from active semiconductor devices).

2. Measure the resistance from each of the five L3 pins to VSS, both polarity
- Same side as L1, each L3 pin should be connected to a L1 pin, and also connected directly (~ 0 ohm) to a BP FID pins
1st ----- AN27
2nd ----- AL27
3rd ----- AN25
4th ----- AL25
5th ----- AJ27
- There are active connections from each L3 pin to VSS/VCC, so resistance should be small (< 100 ohm?).

3. If there are cuts in the L3,
measure the resistance of each of the L3 pins on the other side of cut (away from L1) to VSS, this should be 1000 ohm for each of the cut pins.

Explanation:
- For a cut bridge (open), the 1000 ohm is not pulling down the L3 pin, hence that L3 pin becomes Logic 1.
- For non-cut bridge (close), the 1000 ohm pulls down the node, hence the L3 pin is defaulted to Logic 0.
- The logic 0 or logic 1 is then detected by the multiplier control circuit and FID driver.



The L3 bridge (1st 2nd 3rd 4th 5th)
Tbred B 1800+: open close close close close
Thorton 2000+: open open close close close
Barton 2500+: close close close close close
Barton 2600+: open close close close close
...

Another Way .....

You work by AMD ... and you must make a Multi Lock ... what make you???




To the Moment is no Problem to Overclock ... the FSB is not locked ... and for the Prozzesor is over 200 Mhz no Problem (normaly 166 Mhz)

When you must make a 100% Lock you must Lock the FSB ... for this Lock must change the DIE ... is this right??

AMD can change FSB for the new Bartons to 200 Mhz and set the Multiplier to 9* ... The overclocking is at end!! ... AMD cant this not make 50 - 70% have a old Mainboard with 166 Mhz FSB .... new CPU with 200 Mhz FSB not work ... 1 Year later ????

Question: Have locked and unlocked CPU´s the same not existet Pin´s ??? ... or have eventually Resistors of the Socket Side???

The two resistor Array´s have 1000 Ohm(102 printed on Resistor Array) from one Side to any pins of another Side.... Is the Resistor works or is it short connected of both sides (~250 Ohm from every Pin to Vss or Vcc) and 0 - 0,8 Ohm (Tolerance!!!) from L3 Bridges 1 -> 2 , 1 -> 3 ....????

... under the black Safebutton (4 pices safe the DIE from Cooler ... ) left from Resistor Array ... any place for Hidden Bridges???


Darkman

I've tried measuring the resistance between quite a few of the "fake" pins on my locked Barton and all are infinite (ie: not connected). However, under a microscope the fake pins look very much like where the normal traces drop down into another layer of the PCB, so quite possibly they are connected to something that we don't know about yet.

darkman101010 said:

...
Question: Have locked and unlocked CPU´s the same not existet Pin´s ??? ... or have eventually Resistors of the Socket Side???

The two resistor Array´s have 1000 Ohm(102 printed on Resistor Array) from 1 to another Pin´s ... or 4 seperate Resistor in one Chip?... can you control this?.... Is the Resistor works or is it short connected of both sides (~250 Ohm from every Pin to Vss or Vcc) or is a dummy?
Darkman

Click to expand...

I measured resistance from each L3 pin to VCC or VSS, each is around 70 ohm. I used an unlocked Tbred B 1800+ w/ OLD PCB. (see previous post)

Deathstar13 measured them, each L3 pin is around 270 ohm to VCC or VSS. (see previous post)

So if yours measured around 250 ohm, is consistent with what Deathstar13's result. Deathstart13 used a locked Thorton 2000+.

Resistance is same to VSS or VCC because the resistance between VSS and VCC is very small (< 10 ohm non-linear resistance from semiconductor devices).

For an unlocked Tbred B,

for those L3 pin whose bridge is open, the resistance after the cut, which is on the same side as the pulldown resistor, is 1000 ohm measured to VSS.

Sorry hitechjb1 i have change my last post ... i have no measured ... i calculate ... 4 Resistors with 1000 Ohm to paralell conected is 250 Ohm ... right?? ... ok when Deathstar13 measured have eventually not a 100% Contact(or Tolerance from Multimeter) ....

By the unlocket Thorton 2000+ have a Contact over the internal Pullup Circuit and Internal Multiplier Controler and 70 Ohm can right when is unlocked



@hitechjb1 by the last post this is right and normally ... 1000 Ohm from Vss to any open Bridges ... Look at the Picture on this Link: http://fab51.fc2web.com/pc/barton/athlon-e20.html#index

... from Vcc to open Bridges is the same!!! ... ok!!!

Thanx ... i anderstand the low Resistace fom VCC to VSS ... the Prozzessor have high Power high Ampere with low Volt (R=U/I) ....



Darkman

darkman101010 said:

...

@hitechjb1 by the last post this is right and normally ... 1000 Ohm from Vss to any open Bridges ... Look at the Picture on this Link: http://fab51.fc2web.com/pc/barton/athlon-e20.html#index

... from Vcc to open Bridges ??? Ohm

Darkman

Click to expand...

The resistance from open bridges to VCC is almost the same as that to VSS. This is because the resistance between VCC and VSS is very small (< 10 ohm through non-linear semiconductor devices).

That is
pulldown_resistance_of_open_bridge (after the cut) to VCC
= pulldown_resistance_of_open_bridge (after the cut) to VSS
= 1000 ohm

For L3 pins same side of L1,
resistance_of_L3_pin to VCC = resistance_of_L3_pin to VSS ~ 70 ohm
(from an unlocked Tbred B 1800+)

has anyone compared the unlocked bartons to the locked ones?