Simple secrets revealed on K6-2 cpus?

Resources, guides, discussions for modding your system.

Simple secrets revealed on K6-2 cpus?

Postby atang1 » Sun Mar 23, 2003 5:08 pm

I have studied many a k6-2 cpu from date code to current drawn on core or I/O cmos transistors.

I have never took the trouble to cut them open until today. The heat spreader of these k6-2 were shabily put on. And now I believe that the many failure of k6-2 were cooling problems between the aluminum heat spreader and the silicon chip.

The ceramic package was done first class, done in the best design possible. The silicon chip was mounted flip chip style. Then the problem came when they assembled the heat spreade to the silicon chip. AMD used silicone rubber.

Silicon rubber has poor heat conductivity. No heat conductive filler was used. So many k6-2 suffered poor heat transfer to the heatsink and fan. Overheating caused poor reliability.

When you use single edged razor to cut the four corner dots, the heat spreader can be easily pried open. You then remove the thin layer of silicone rubber and you can directly cool the silicon chip.

Wow, the dead 450 AFX came to life at 300 mhz. But we know 300 mhz is due to initial overheating. Soon after it gets low temperature annealing will run at 450 mhz normally.

The package looks and feels like an Athlon. It also has tiny gold dots near the capacotors. What these dots are, maybe some one already knew. But removing the silicone rubber sure resurrected this dead 450 cpu. As a matter of fact, I am posting this secret using this 450 mhz cpu at 300 mhz.

It is not hot, because it has no heat spreader or silicone rubber. Just some thermal grease with zinc oxide. Motherboard was jumped 6x75mhz, but bios shows 300 mhz. Powerleap"cpu control panel" shows 330 mhz calculated, 6x50mhz cpu internal frequency, FSB unknown.

Postby atang1 » Sun Mar 23, 2003 7:41 pm

The dots on the top of the ceramic package appear to be test points.

When I look for the flip chips connections to the pins by cracking a corner of the ceramic package. I separated some pins from the main package. The traces connected to the pins are not visible. Looks like they are so small that I need polishing and staining(standard metallurgical procedure) plus high power microscope to be able to see anything.

Now, I will be opening some 18 K6-2s from 300 to 550 mhz AFR to AFX units. Some AFX failed due to I/O cmos transistors saturated. Some 400 mhz AFR has core that has too high a dark current. They will then go thru a low temperature annealing process, before testing on any motherboards.

Anyone who has defective K6-2 could try some experiments themselves. When the aluminum heatspreaders were removed, the units look like Athlons, just a beautiful sight with different colored silicon dioxides.

Postby atang1 » Tue Mar 25, 2003 6:47 am

So, removing the aluminum heat spreader saved many k6-2 cpus. But they still have problems running at slower FSB internally in the cpu.

One 450 mhz AFX runs at 300(330 estimated by Sandra and Powerleap). stepping 0C.(2.0/6.0x75) ran at 6.0x50mhz internally.

One 300 mhz AFR runs at 150 mhz. stepping 0A.(2.0/6.0x75), except 0A stepping does not have any multipliers over 5.0x. So, 6.0x becomes 2.0x and hence the cpu ran at 150 mhz.

Further studies is obvious needed. Suspicion lies in the gate oxide being too thick as a problem. When they have the aluminum heat spreader, they ran hot and is incapable even to boot. When they ran cool, they have slower speeds. May have to increase core voltage to improve speeds?

What was the failure mechanism? Not operating in the proper range of temperature. Not too hot and not too cool is the sweet spot in the operating range of 30-80 degrees C.

Postby tonycarr » Sun Mar 30, 2003 3:29 am

I've read the k6-2 is very voltage sensitive , many boards can not provide low enough core voltage to run a k6-2 . I would like to try a k6-3 but my board only goes down to 2.2 volts , too much so I have read . Do you think removing the heat cover would allow a k6-3 to run with a 2.2 volts core ?
E Pluribus Unum
Black Belt 2nd Degree
Black Belt 2nd Degree
Posts: 2218
Joined: Fri Dec 28, 2001 12:00 am
Location: SouthEastern , USA

Postby atang1 » Tue Apr 01, 2003 7:45 pm

Amd did things without knowing what they were doing in their Austin Texas semeconductor factory. Now that I removed the aluminum heatspreader, the k6-2s are too cool to operate at factory specified speed. I am using smaller fans and will try even smaller fans or none at all.

Since I am an engineer, the changes are planned and sometimes take a while to implement.

The K6-3 is done in 0.18 micron rule, and the core voltage should be lower. I am thoroughly familiar with the core voltage regulator system design and can modify the core voltages to suit the specific cpu. Read my posts on slot1 core voltage secrets. But k6-2+ or K6-3 varieties, all need AMD fixes(L2 and L3 cache enable) to work with non super 7 socketed motherboards, if you want to experiment.

Postby atang1 » Tue Apr 29, 2003 6:21 am

Recently, I looked into some of the failure mechanisms of the K6-2 AFX and AHX series. These were 500-533 mhz units. There are some amperage and wattage information published. These units require 12-13 amperes.

Considering the pcboard trace resistance of 0.1 ohm, and the equivalent resistance of regular K6-2(AFR) to be 0.24 ohm, At 2.2 volts, AFR draws about 8-9 amperes.

To draw 13 amperes, the equivalent resistance had to be 0.16 ohm plus the trace resistance added to 0.17 ohm.

This equivalent resistance becomes a major obstacle to the selection of mosfet bypass transistor supplied by 5 volts. Mosfet transistors To220 package have a saturation resistance(RDS on state) of 0.22-0.18 ohm. These transistors can draw 15 amperes. If the motherboard uses mosfet transistor with RDS of 0.40 ohm, then it can only draw 7.69 amperes. Most older motherboards have large heatsinks and use bypass trasnsistors of 5-7 amperes.

Now, we understand why some K6-2 will work on some motherboards and not others. The voltage and ampere requirements are not satisfied. Mystery is resolved by looking up the mosfet transistor number and RDS value. the old designs are for 5-7 ampere supply. You could stretch it to 8-9 amperes running very hot. But the 13 ampere K6-2 will not boot.

A change to a 25-27 ampere TO-220 n channel mosfet transistor is then necesseary. Do you have the motherboards that need changes? Toshiba 2SK2314-ND only cost $1.42 each.

Have fun. Be happy. Life is a bowl of cherries.

Postby atang1 » Fri May 02, 2003 7:10 pm

In the early days, FIC V503+ had problems using K6-2 500-533 mhz cpu because to the amperage of 13 amps. Under intensive usage, the cpu becomes unstabel.

Because they use SC1164 voltage regulator, they have decided to upgrade by the sensing voltage deviders to allow the voltage regulator to maintain the current supply. The voltage devider 1248 and the r149 were reversed. This allows the bypass mosfet transistor to avoid limiting to 12 amps.

Since the bypass mosfet transistor has the gate on the voltage regulator out put, their engineers should have just change the bypass transistor to 25 amps, and leave the sensing voltage the same. In other words, if a 7 amp transistor is marginal, you put two 7 amp transistor to do 14 amps. 3 to do 21 amps and 4 to do 28 amps. Voltage regulator is always set the same.

So much for factory solutions.
Last edited by atang1 on Fri May 09, 2003 9:02 am, edited 1 time in total.

Postby blackhead » Thu May 08, 2003 8:42 pm

Man.. In the whole world isn't word for naming your work!

Thank you for delivering such inforamtion to the lamers like me!

Tell me please:
I have some older MoBo of 02'97 with Socket 7 and K6-233APR of '97 (3.3V core & I/O). But it doesn't work! It worked fine, but one day it dies. I tried it after months again and it comes to the BIOS. But when I set the corect values and try to reboot it does nothing. Black screen.

Where should be the problem? Its rather CPU or voltage regulator TO220 on the board? I heard the FETs "dies in the first line"...

What's your mind?

Thanx a lot for any note!
George Blackhead
Posts: 31
Joined: Mon May 05, 2003 5:33 pm
Location: ZZZ Europe

Postby atang1 » Fri May 09, 2003 8:54 am

If it posts to the bios, Your motherboard and cpu are not dead. If you have an ISA or PCI mini post card, you can see where it stopped working.

The suspicion is overheating. K6-233 probably requires lowering of core voltage. If not available on the motherboard, then use a larger heatsink. Most motherboard which can handle only p54c is not worth it to modify the core voltage. A socket converter is also too expensive.

But it could be as simple as reapplying thermal grease, which might have dried up over the years.

Postby blackhead » Fri May 09, 2003 9:26 pm


1) No POST. (VGA BIOS sing not visible)
2) No POST card. I've seen only one in my whole life.. And it was for about $70 (+/- 2000 Kc - big money in my country).
3) I applied "thermal grease" some times (once per year) but no change when applied after death...
4) I changed core voltage. I tryied every settings in the "legal" range for this CPU (VCore<3.3V). But no change.

Hmm... Maybe is the board (or some of its components) really damaged.

Thank you for your time. :oops:
Posts: 31
Joined: Mon May 05, 2003 5:33 pm
Location: ZZZ Europe


Return to System Modification

Who is online

Users browsing this forum: No registered users and 2 guests