Fritzchens Fritzs make some cool picture of the K6-3+ die :
https://www.flickr.com/photos/130561288@N04/a … 157715070825122
There's even the cache mod.

A return appearance of my k6-2+ 570 from earlier in the thread. I turned my Aoyue (I have no idea how to pronounce that) soldering iron up to its max temp, added some solder, then held it against the resistor with some sideways pressure. It surprised me that it came off as easily as it did (you can see a diagonal line where my soldering iron's tip dragged across the surface after it let go).

I had been fooling with my Soyo 5EMA+ and my K6-2+ 550, and the board was already set to 5.5 x 112 @ 2.2 volts, so I left the settings as they were, and popped this in. It booted right up, and was recognized as a K6-III @ 616. I got to the desktop with no problems, then ran 3D 2000. It CTD, so I assumed it needed a little more voltage, and set it to 2.3. It again CTD before completing the run, so I tried reducing the voltage to 2.1 and the bus to 100 MHz, and that did the trick; 3D 2000 completed with no issue. I then tried 5.5 x 112 again, and it again completed it with no issue, delivering 400 more points (4,400) than the K6-2+ had with all else the same.

The cache on this board won't work with the 124 MHz bus, so while 620 MHz gives the best memory performance as measured by Sandra, the 3D 2000 score had always been lower with a K6-2+. I really wanted to try a K6-3+ at this speed, but the 1.6 volt 400 I have will not go this fast. So now I was finally able to try a K6-3+ at 620 and find out if that extra 128k of full speed cache could make up for having to disable the 1024k on the board. And the answer was no. 5 x 124 delivered a score of 4,100, so about a 300 point drop.

It wouldn't POST @ 5.5 x 124 even with the voltage set to 2.4. I got it to POST once at 6 x 112, but it mostly wouldn't. I wonder if this board can't supply the power needed for these overclocks? I'll have to try these settings again with my Epox MVP3G-M and this CPU.

Congratulations to Sphere for pursuing this. I mostly didn't think it was possible because I was sure that if it were, it would have been discovered decades ago.

Today I moved the upper left resistor. The L2 cache was completely disabled and the CPUid got the number 5C8 (it's a Chomper XT now) 😀

Repo Man11 wrote on 2022-03-06, 01:43:

A return appearance of my k6-2+ 570 from earlier in the thread. I turned my Aoyue (I have no idea how to pronounce that) solderi […]
Show full quote

A return appearance of my k6-2+ 570 from earlier in the thread. I turned my Aoyue (I have no idea how to pronounce that) soldering iron up to its max temp, added some solder, then held it against the resistor with some sideways pressure. It surprised me that it came off as easily as it did (you can see a diagonal line where my soldering iron's tip dragged across the surface after it let go).

I had been fooling with my Soyo 5EMA+ and my K6-2+ 550, and the board was already set to 5.5 x 112 @ 2.2 volts, so I left the settings as they were, and popped this in. It booted right up, and was recognized as a K6-III @ 616. I got to the desktop with no problems, then ran 3D 2000. It CTD, so I assumed it needed a little more voltage, and set it to 2.3. It again CTD before completing the run, so I tried reducing the voltage to 2.1 and the bus to 100 MHz, and that did the trick; 3D 2000 completed with no issue. I then tried 5.5 x 112 again, and it again completed it with no issue, delivering 400 more points (4,400) than the K6-2+ had with all else the same.

The cache on this board won't work with the 124 MHz bus, so while 620 MHz gives the best memory performance as measured by Sandra, the 3D 2000 score had always been lower with a K6-2+. I really wanted to try a K6-3+ at this speed, but the 1.6 volt 400 I have will not go this fast. So now I was finally able to try a K6-3+ at 620 and find out if that extra 128k of full speed cache could make up for having to disable the 1024k on the board. And the answer was no. 5 x 124 delivered a score of 4,100, so about a 300 point drop.

It wouldn't POST @ 5.5 x 124 even with the voltage set to 2.4. I got it to POST once at 6 x 112, but it mostly wouldn't. I wonder if this board can't supply the power needed for these overclocks? I'll have to try these settings again with my Epox MVP3G-M and this CPU.

Congratulations to Sphere for pursuing this. I mostly didn't think it was possible because I was sure that if it were, it would have been discovered decades ago.

Thanks Repo!

I’ll add this reply to the OP list under 570acz

Btw, don’t feel too bad about your results, actually they are quite good results. I can’t really do much other than the same with my 550 3+ (original oem 550 3+)

Fritzchens Fritz wrote on 2022-03-06, 02:43:

Today I moved the upper left resistor. The L2 cache was completely disabled and the CPUid got the number 5C8 (it's a Chomper XT now) 😀

DSC05064_cpuid_5C8.jpg

2022-03-06_031038_479.png

You have got to be kidding me…

Does this mean there are more out there that may be able to be modded?

Edit:

https://en.m.wikipedia.org/wiki/List_of ... 6,_250_nm)

Okay, so if there were more models to unlock I would expect to see non plus cpus that are on the 180nm process. But I don’t see them.

Perhaps you have created a processor that amd never released?

I wonder what the pads that never had anything on them are for....

Sphere’s socket 5/7 cpu collection.

I’m taking a look at my collection here,

Everything since the 266 k6-2 has been hovering around 2.0-2.2v or so with an occasional dip to 1.9 and spike to 2.4v

Nothing is really standing out as a obvious “I’m a 180nm look at me” but all the voltages are within the range that a 2+/3+ could survive.

Unfortunately aside from you guys figuring out a way to identify through software, I think there isn’t much way to tell before pulling the lid.

I fully expect we won’t find more cores that can do this, but if we do that will be cool!

Fritz, can this processor with the link moved still change multiplier from windows using the utilities

cyclone3d wrote on 2022-03-06, 03:14:

I wonder what the pads that never had anything on them are for....

Idk, but someone should put a volt meter on these pins

Re: Tillamook Dual CPU SMP Mod. Is it possible?

I’d be really interested if they show connection to those pads.

(I think the chances of that are almost non existent though) it would make zero sense for amd to design the + cpus with the ability to do mps or at least I can’t understand it

Rather than risk working cpus, fritz may wanna use one of the display cpus.

Well now I gotta find that K6-2P which I bought thinking it was a plus but wasn't...

Bear in mind that AMD was the king of stealth releases in that period, you'd find a CPU for sale and it didn't appear on their site for 3 months.

BitWrangler wrote on 2022-03-06, 03:25:

Well now I gotta find that K6-2P which I bought thinking it was a plus but wasn't...

Bear in mind that AMD was the king of stealth releases in that period, you'd find a CPU for sale and it didn't appear on their site for 3 months.

I’m curious to see a delidded 533 or 550 k6-2 now.

But I’m kinda wondering if amd may have made different models of said cpus, using different cores and didn’t tell anyone though…

Again, I really doubt this. It’s probably nothing more than violating a rare cpu to find this out. I don’t think we are going to find a cache equipped die under there.

At least not a 180nm one….

250nm🤔💁‍♂️ Maybe……..

Fritz, do you have a k6-3 non plus you feel like playing with?

Sphere478 wrote on 2022-03-06, 03:02:

You have got to be kidding me… […]
Show full quote

Fritzchens Fritz wrote on 2022-03-06, 02:43:

Today I moved the upper left resistor. The L2 cache was completely disabled and the CPUid got the number 5C8 (it's a Chomper XT now) 😀

You have got to be kidding me…

...

Okay, so if there were more models to unlock I would expect to see non plus cpus that are on the 180nm process. But I don’t see them.

Perhaps you have created a processor that amd never released?

Yes and no, I saw in a couple of earlier posts the question of what happens when the other caps are moved. So I gave it a try. That the L2 cache can be deactivated in this way is quite interesting but completely useless and does not open up any new usefull mods. Only the top left and the bottem right cap is a 0Ω resistor all other caps are capacitors with round about 15nF or 470nF.

Sphere478 wrote on 2022-03-06, 03:37:

Fritz, do you have a k6-3 non plus you feel like playing with?

Yes

Fritzchens Fritz wrote on 2022-03-06, 04:01:

Yes and no, I saw in a couple of earlier posts the question of what happens when the other caps are moved. So I gave it a try. T […]
Show full quote

Sphere478 wrote on 2022-03-06, 03:02:

You have got to be kidding me… […]
Show full quote

Fritzchens Fritz wrote on 2022-03-06, 02:43:

Today I moved the upper left resistor. The L2 cache was completely disabled and the CPUid got the number 5C8 (it's a Chomper XT now) 😀

You have got to be kidding me…

...

Okay, so if there were more models to unlock I would expect to see non plus cpus that are on the 180nm process. But I don’t see them.

Perhaps you have created a processor that amd never released?

Yes and no, I saw in a couple of earlier posts the question of what happens when the other caps are moved. So I gave it a try. That the L2 cache can be deactivated in this way is quite interesting but completely useless and does not open up any new usefull mods. Only the top left and the bottem right cap is a 0Ω resistor all other caps are capacitors with round about 15nF or 470nF.

Sphere478 wrote on 2022-03-06, 03:37:

Fritz, do you have a k6-3 non plus you feel like playing with?

Yes

Btw fritz, what equipment/method/settings temps/wattages are you using to move those resistors, your work is very clean.

Well, this is useful, it suggests the possibility that late in the game AMD may have filled their k6-2 offerings with cache equipped cores but simply disabled them, or at least they gave themselves the option.

We of course have the ability to disable to cache already in bios, so yes, there isn’t much use for this as why would we want to do this.. but it gives us more info to further understand what’s goin on here.

I’m suspecting that the k6-3 non plus will have these options also, a 0kb, 128kb, and 256kb option.

As for the missing surface mount components,

Do see if you can read continuity between them and some pins, that would help us find out what they are for.

Edit: the new 0-ohm link was a false reading.

I went ahead and did some looking,

These pads are for pulling something to vss

And I found another 0 ohm link

Going around to the various SMD components I can seem to find a connection to vss on most all of them,

I guess if you are bored, you could try installing a capacitor there, if that did nothing, a 0 ohm link. But that kind of experiment will probably release the magic smoke.

Sphere478 wrote on 2022-03-06, 04:11:

Btw fritz, what equipment/method/settings temps/wattages are you using to move those resistors, your work is very clean.

-pre tinning gold pad with solder iron
-CPU with the pins on a camping hotplate at 180°C - 200°C
-Hotair station at 240°C (low air speed)
-some flux
-waiting 30 - 60 seconds and the resistor can be moved very easy with a tweezer

Sphere478 wrote on 2022-03-06, 04:11:

I’m suspecting that the k6-3 non plus will have these options also, a 0kb, 128kb, and 256kb option.

As for the missing surface mount components,

Do see if you can read continuity between them and some pins, that would help us find out what they are for.

With the K6-III (non plus), I would also assume that something similar can be done. But since I only have only one working cpu, I will not test this.

The pins and pads I have briefly tested for connection but see no chance there, because I have already connected all three gold pads for the bottom right resistor and it was still a 128kb K6-2+

Sphere478 wrote on 2022-03-06, 04:22:

5339C029-8F30-41D9-B185-26FE01978195.jpegI went ahead and did some looking, […]
Show full quote

5339C029-8F30-41D9-B185-26FE01978195.jpegI went ahead and did some looking,

These pads are for pulling something to vss

And I found another 0 ohm link

Going around to the various SMD components I can seem to find a connection to vss on most all of them,

I guess if you are bored, you could try installing a capacitor there, if that did nothing, a 0 ohm link. But that kind of experiment will probably release the magic smoke.

Hm... my bottom middle left is not a 0 ohm cap

Fritzchens Fritz wrote on 2022-03-06, 04:25:

-pre tinning gold pad with solder iron -CPU with the pins on a camping hotplate at 180°C - 200°C -Hotair station at 240°C (low a […]
Show full quote

Sphere478 wrote on 2022-03-06, 04:11:

Btw fritz, what equipment/method/settings temps/wattages are you using to move those resistors, your work is very clean.

-pre tinning gold pad with solder iron
-CPU with the pins on a camping hotplate at 180°C - 200°C
-Hotair station at 240°C (low air speed)
-some flux
-waiting 30 - 60 seconds and the resistor can be moved very easy with a tweezer

Sphere478 wrote on 2022-03-06, 04:11:

I’m suspecting that the k6-3 non plus will have these options also, a 0kb, 128kb, and 256kb option.

As for the missing surface mount components,

Do see if you can read continuity between them and some pins, that would help us find out what they are for.

With the K6-III (non plus), I would also assume that something similar can be done. But since I only have only one working cpu, I will not test this.

The pins and pads I have briefly tested for connection but see no chance there, because I have already connected all three gold pads for the bottom right resistor and it was still a 128kb K6-2+

Sphere478 wrote on 2022-03-06, 04:22:

5339C029-8F30-41D9-B185-26FE01978195.jpegI went ahead and did some looking, […]
Show full quote

5339C029-8F30-41D9-B185-26FE01978195.jpegI went ahead and did some looking,

These pads are for pulling something to vss

And I found another 0 ohm link

Going around to the various SMD components I can seem to find a connection to vss on most all of them,

I guess if you are bored, you could try installing a capacitor there, if that did nothing, a 0 ohm link. But that kind of experiment will probably release the magic smoke.

Hm... my bottom middle left is not a 0 ohm cap

I re did the test, you are correct, I think my probe touched another pad and made a connection through vss

Grabbed these in dollar store earlier, gonna try them for levelling pads (like Athlon XP) ... regular stick-on foam feet are too thick usually.

BitWrangler wrote on 2022-03-06, 04:45:

Grabbed these in dollar store earlier, gonna try them for levelling pads (like Athlon XP) ... regular stick-on foam feet are too thick usually.

Interested to see how it works out, but I really think the effort should be to put the ihs back on.

These dies are too brittle.

Edit:
Trying to figure out how well the k6-2+ to k6-3+ mod is holding up.
I got the ihs reinstalled, and it’s running super pi 32m at 600mhz no issues, I suppose that means that the adhesive I used is conducting well..

I ordered adhesive from ebay but they only sent 1 tube of the two tube epoxy… so I was forced to use the adhesive as the thermal medium as amd did.

The adhesive I used isn’t as hard as what amd used. So I’m still trying to figure that out.

Repo Man11 wrote on 2022-03-06, 01:43:

A return appearance of my k6-2+ 570 from earlier in the thread. I turned my Aoyue (I have no idea how to pronounce that) solderi […]
Show full quote

A return appearance of my k6-2+ 570 from earlier in the thread. I turned my Aoyue (I have no idea how to pronounce that) soldering iron up to its max temp, added some solder, then held it against the resistor with some sideways pressure. It surprised me that it came off as easily as it did (you can see a diagonal line where my soldering iron's tip dragged across the surface after it let go).

I had been fooling with my Soyo 5EMA+ and my K6-2+ 550, and the board was already set to 5.5 x 112 @ 2.2 volts, so I left the settings as they were, and popped this in. It booted right up, and was recognized as a K6-III @ 616. I got to the desktop with no problems, then ran 3D 2000. It CTD, so I assumed it needed a little more voltage, and set it to 2.3. It again CTD before completing the run, so I tried reducing the voltage to 2.1 and the bus to 100 MHz, and that did the trick; 3D 2000 completed with no issue. I then tried 5.5 x 112 again, and it again completed it with no issue, delivering 400 more points (4,400) than the K6-2+ had with all else the same.

The cache on this board won't work with the 124 MHz bus, so while 620 MHz gives the best memory performance as measured by Sandra, the 3D 2000 score had always been lower with a K6-2+. I really wanted to try a K6-3+ at this speed, but the 1.6 volt 400 I have will not go this fast. So now I was finally able to try a K6-3+ at 620 and find out if that extra 128k of full speed cache could make up for having to disable the 1024k on the board. And the answer was no. 5 x 124 delivered a score of 4,100, so about a 300 point drop.

It wouldn't POST @ 5.5 x 124 even with the voltage set to 2.4. I got it to POST once at 6 x 112, but it mostly wouldn't. I wonder if this board can't supply the power needed for these overclocks? I'll have to try these settings again with my Epox MVP3G-M and this CPU.

Congratulations to Sphere for pursuing this. I mostly didn't think it was possible because I was sure that if it were, it would have been discovered decades ago.

You just left your resistor off I take it?

Looks like we have another 570 success story.

This is my second 570 that is stable at (at least) 600mhz unlocked.

Edit: 616mhz 112fsb lookin good, edit2 inconclusive, thinkin 112 is giving agp issues.

Repo Man11 wrote on 2022-03-06, 01:43:

Congratulations to Sphere for pursuing this. I mostly didn't think it was possible because I was sure that if it were, it would have been discovered decades ago.

There was never an incentive to try, as for the longest time there were plentiful K6-2+ and K6-3+ to be found. K6-3+ only ran out some years ago and prices are not low anymore, so now there's incentive to try out mods like these.

Iirc a somewhat similar cache mod is possible with some of the sA Athlon XPs, but I never ended up trying it (it involved something with those bridges).

I have wondered for a longer time if a similar mod could be possible on s370 CPUs btw, that could make things very interesting if this were the case.

Fritzchens Fritz wrote on 2022-03-06, 04:01:

Yes and no, I saw in a couple of earlier posts the question of what happens when the other caps are moved. So I gave it a try. T […]
Show full quote

Sphere478 wrote on 2022-03-06, 03:02:

You have got to be kidding me… […]
Show full quote

Fritzchens Fritz wrote on 2022-03-06, 02:43:

Today I moved the upper left resistor. The L2 cache was completely disabled and the CPUid got the number 5C8 (it's a Chomper XT now) 😀

You have got to be kidding me…

...

Okay, so if there were more models to unlock I would expect to see non plus cpus that are on the 180nm process. But I don’t see them.

Perhaps you have created a processor that amd never released?

Yes and no, I saw in a couple of earlier posts the question of what happens when the other caps are moved. So I gave it a try. That the L2 cache can be deactivated in this way is quite interesting but completely useless and does not open up any new usefull mods. Only the top left and the bottem right cap is a 0Ω resistor all other caps are capacitors with round about 15nF or 470nF.

Sphere478 wrote on 2022-03-06, 03:37:

Fritz, do you have a k6-3 non plus you feel like playing with?

Yes

The only potentially (at least somewhat) useful mod reason that I could think of, is to revive a dead + chip after it has had its cache region damaged in some way (which would usually be user error after having removed the aluminium lid and crushed part of the core with the heatsink). These cores seem quite fragile, so someone might mess this up sooner or later 😜
And it may help achieve higher clockfrequencies in raw megaherz contests or something 😜

Trying to figure out dates to see what HS are worth looking under https://techmonitor.ai/techonology/amd_dresde … chips_next_year that says K6-III were first produced in July 1999 (edit: on 0.18 micron) , so I highly doubt we'd see anything earlier than 9925

BUT, another source from late 1998 implied AMD would be testing on 0.22 micron in early 99 before going full on 0.18 micron... so are there some "tweeners" ???

Looking at the 1.8V K6-2 Mobile I dug out, it's date code is 9905, so I think that's probably too early to bother delidding.

Sphere478 wrote on 2022-03-06, 05:22:

BitWrangler wrote on 2022-03-06, 04:45:

Grabbed these in dollar store earlier, gonna try them for levelling pads (like Athlon XP) ... regular stick-on foam feet are too thick usually.

Interested to see how it works out, but I really think the effort should be to put the ihs back on.

These dies are too brittle.

Did I mention it's not my first rodeo? Here's a K6-2 I was playing with 20 years ago, probably my second rodeo. Had to shave those pads down after finding they were too thick. Note it's wearing a scotch tape "bib" to keep thermal paste off the discretes. Think this is a non-CXT 350 I was trying to get 400 out of. Still has all it's corners, just needs a cleanup, but wanted to leave it looking like I didn't just do it this morning....