One of the 570 CPUs I did this to had that happen, but it still worked and overclocked to 600+ MHz. I thoroughly tested it, thinking that it would surely have some sort of issue, so I was pleasantly surprised that it didn't.

What the explanation can be? I mean if there are transistors there it shouldn't work.
The same was happening with socket 462 cpus with small chips on the die they still worked fine.
Taking another look at the picture I also noticed that 2 capacitors are really in bad shape.

Transistors are on the other side, what you see is the back. Thick wafer and luck. K7 had fragile thin weak wafers crumbing at the slightest twist of radiator. I saw multiple tens, maybe over a hundred broken AMD chips with slightly cracked corners. Intel socket 370 Coppermines were bulletproof.

rasz_pl wrote on 2023-05-15, 22:26:

Transistors are on the other side, what you see is the back. Thick wafer and luck. K7 had fragile thin weak wafers crumbing at the slightest twist of radiator. I saw multiple tens, maybe over a hundred broken AMD chips with slightly cracked corners. Intel socket 370 Coppermines were bulletproof.

Thank you for the explanation, isn't the material used all the same? The one of Pentium 3 was different?

Material, monocrystalline silicon, is the same.
'The Amazing, Humble Silicon Wafer' https://www.youtube.com/watch?v=sIRfWyyOFPg
'The 300mm Silicon Wafer Transition' https://www.youtube.com/watch?v=0s5TO9h6fco
Difference is how its treated, processes (Infusion, implantation, thermal deposition, diffusion) in the fab. Intel Coppermines are super hard, AMD K7 are brittle.
Thermaltake Golden Orb is fine on Intel, but will kill AMD cpu http://www.dansdata.com/gz012.htm
"Thermaltake's various Orbs, for instance; barring a rash of nasty core-crushing incidents when the old twist-lock Golden Orb was used on the then-new Socket A CPUs, they've all been pretty much trouble-free, performed well enough for all but the most extreme users, been well priced, and looked terrific."

It isn't about silicon hardness. It's about organic vs. ceramic base packaging. Organic allows for greater non-uniform pressure. It just bends slightly. Ceramic doesn't bend at all and it results in silicon cracks on edges. Coppermines also have near square die size which helps.

ph4nt0m wrote on 2023-05-18, 18:21:

It isn't about silicon hardness. It's about organic vs. ceramic base packaging. Organic allows for greater non-uniform pressure. It just bends slightly. Ceramic doesn't bend at all and it results in silicon cracks on edges. Coppermines also have near square die size which helps.

how will you explain crushed Palomino and later cores then?

rasz_pl wrote on 2023-05-18, 20:54:

ph4nt0m wrote on 2023-05-18, 18:21:

It isn't about silicon hardness. It's about organic vs. ceramic base packaging. Organic allows for greater non-uniform pressure. It just bends slightly. Ceramic doesn't bend at all and it results in silicon cracks on edges. Coppermines also have near square die size which helps.

how will you explain crushed Palomino and later cores then?

Power requirements of Palomino are much higher than of Coppermine, 64W vs. 37W TDP. It explains why people had to use large and heavy coolers with socket latches which they shouldn't really be using at all. Those coolers mounted through the holes in mainboards didn't break anyhting. For example, Thermaltake Big Typhoon or Thermalright SP-97.

I did this mod on 11 CPUs and these are my results:

AMD-K6-2+/570ACZ --> success for 8 of 9

AMD-K6-2+/350AUZ --> success for 0 of 2

It surprised me very much that one of the AUZ has a green die like all the ACZ, but the other AUZ has a maroon die.

I thought the color was just a surface/coating effect, because I have seen multiple colors on K6-2 and Duron, had one at one point that was half and half maroon and green with the transition being from corner to corner.

Yeah, we concluded long ago that if you attempt this, you should not do it on non 570 chips

570 chips aren’t rare, the rest are. And the 570s are working at 3+.

Anyone got a k6-3 non plus they wanna play with the links on to see what happens?

Damn, what are we looking for now?

I could jam up my AMD K6 with like half a pound of strawberry, and see how long it takes until I have an anthill inside.

Though my mind wandered recently to whether Tualatins had any secrets under the spreader and looked up pics, but nothing there apart from die, was wondering because it seems like they are cache halved PIII-S going from 8 way associative to 4 way in the process, but there was nothing on top or on the bottom different so must be laser cut on die or something.

BitWrangler wrote on 2023-08-06, 03:52:

Damn, what are we looking for now?

I could jam up my AMD K6 with like half a pound of strawberry, and see how long it takes until I have an anthill inside.

Though my mind wandered recently to whether Tualatins had any secrets under the spreader and looked up pics, but nothing there apart from die, was wondering because it seems like they are cache halved PIII-S going from 8 way associative to 4 way in the process, but there was nothing on top or on the bottom different so must be laser cut on die or something.

I suspect that k6-3 may have the same options we discovered on the + cpus. Just want to document them.

If we can notate the cpu ids it may be possible for us to discover if AMD ever let any vanilla k6-2 cpus out in the wild as incognito disabled k6-3 /3+ chips.

I suspect no, but it isn’t impossible.

We did discover and document a new cpu id in this thread BTW, which may have been added to the latest version of chkcpu, and hopefully cpu-z? Never got confirmation .

An idea: if these are ever detected it would be nice if user was notified in app and this thread referenced but prob wishful thinking of adding all that to a program.

I don't think they ever got enough yield out of 250nm K6-3 for them to be functional if modded, if there are any around. Seen a lot of decapped K6-2s, have several, all K6-2 shaped, [] , no squarer ones. When AMD wanted slower industrial/embedded processors in 2005, they did a run of 500Mhz morgan durons, and the pluses were in that niche also prior to that. 2001 to 2005 AMD had a lot more profitable product to be fabbing, unless they schlepped the 250nm plant to another site, I don't think there was opportunity in there for later 250nm K6-3 cores and nobbled k6-2s to happen, and it didn't happen in 1999, or examples would have been noted among the first wave of K6-2 decappers. When 250nm was done, it seems to have been done, with a lot of the plant recycled into the 180nm that did the pluses and then was too busy doing durons etc.

https://youtu.be/MFKW55fzDNk?feature=shared

Still hoping someone can test this on a k6-3 non plus and document results with cache, jumpers, and cpu ids

We also need to get Jan and cpu-z to add that missing cpu id to their utilities

Wow, the temperature sensor seems quite useless but the jumpers that enable the l2 cache plus the size it's very cool.

Could probably make a pcb project that did this, a shim that replaces the heat spreader. Just glues onto the chip heatsink goes directly on die, or we could solder a heat spreader to the pcb.

Anyone have the thickness from the purple ceramic to top of the die?

Sphere478 wrote on 2024-04-02, 22:12:

Could probably make a pcb project that did this, a shim that replaces the heat spreader. Just glues onto the chip heatsink goes directly on die, or we could solder a heat spreader to the pcb.

This will be very neat. I have one K6-2+ 500ACZ waiting to be decapitated or (I hope so) successfully upgraded to K6-III+.