Ready to run!

Deunan wrote on 2022-02-21, 10:57:

Disabled cache might not be down to bad yields, not strictly. By that time AMD had a good idea how well the chips would turn out and perhaps designed a fast cache a bit on the optimistic side. Simply put, they knew not every chip is going to turn out great but some will, and those that will be sold as III+. The rest will be downgraded to 2+ to meet the specs. That's the price for lower latency (which also comes with lower clock ceiling but those are mobile chips after all, perf/W is what they were made for).

We have some results now and already we can see the mod might require better control over temperature, or higher voltage, or both.

BTW if I was desiging that chip I would make both halfs of the cache configurable. I mean unless it is partly configured in the silicon, there should be two jumpers on the CPU, to be set depending on which half of the 256k cache is performing better. So some 2+ chips might require different jumper to be moved, but the final position would be like what III+ parts have.

I had this thought also. Waiting for someone to see what that top link does

0xCats wrote on 2022-02-21, 13:31:

I just want to add this for anyone else seeking to decap their K6-2+

To soften up the silicone adhesive dunk the CPU in acetone upside down for a while, it completely softens it up and you can slice the lid off like it were attached with butter.
Almost no force required.

There is so e sort of epoxy mounting the die to the ceramic. Which may or may not react also

fool wrote on 2022-02-21, 17:28:

Ready to run!

Woohoo!!!

What model?

We need someone adventurous, curious what that top link does, also what happens with no link, link on all three spaces. (Decent chance something may fry but for science!)

I suspect no link may give a third cache option, and all three may be a easier way to mod.

What the top link does I have no idea.

Looks good so far. More testing tomorrow. Now I have to dispose some snow.

Cool, at stock voltage?

One thing bugging me... what if... some 2+ are configured with the other half of the cache disabled... that would point to genuinely faulty part of cache been disabled vs market led binning. There might be another "jumper" that defines top or bottom half that we have not identified yet. Therefore I would suggest that everyone checks that their discrete component layout is identical to the one determined to work before trying it.

For instance, on the Appaloosa core "Applebred" Durons, there were different bridge configurations according to which segment of the 256kB cache was enabled.. there was some activity on cache unlocks of these, but I don't think it got very famous because it was only a short while before it became not possible and was very fiddly to do right anyway... plus many were actual actual cache failed thoroughbreds. AMD seemed to be able to sell Tbreds plenty well enough without having to make any surplus into applebreds to move them.

Nice example of a little bit of knowledge being misleading. I was sure those were just low inductance capcaitors https://www.kyocera-avx.com/products/ceramic- … low-inductance/ . Tricksy/clever of AMD to hide their configuration resistors in with what I still assume are some capacitors.

Things I'm wondering are what are the value of all of the components when removed from the CPU (so, which are 0 ohm links and which capacitors+values). Also, if those tiny pads nearby connect to the larger pads, which might mean using the pencil trick is an option. Might be able to identify signalling pads by whether they connect to CPU voltage pins. Any capacitors probably go between Vcore and Ground, but any signalling pads will probably have a Ground one side, Vcore the other, and the pad itself not connect to Ground or Vcore.

I think it's just that they're not coated with another layer of resin, to provide good surface for part info and humidity protection etc... and caps are done like that because the extra resin might change capacitance.

BitWrangler wrote on 2022-02-21, 18:15:

One thing bugging me... what if... some 2+ are configured with the other half of the cache disabled... that would point to genuinely faulty part of cache been disabled vs market led binning. There might be another "jumper" that defines top or bottom half that we have not identified yet. Therefore I would suggest that everyone checks that their discrete component layout is identical to the one determined to work before trying it.

For instance, on the Appaloosa core "Applebred" Durons, there were different bridge configurations according to which segment of the 256kB cache was enabled.. there was some activity on cache unlocks of these, but I don't think it got very famous because it was only a short while before it became not possible and was very fiddly to do right anyway... plus many were actual actual cache failed thoroughbreds. AMD seemed to be able to sell Tbreds plenty well enough without having to make any surplus into applebreds to move them.

Maybe it was enough that they just made them lower clock versions and treated the cash as a feature to enable or disable?🤷‍♂️ I think we are thinking about this particular part too much and should just see what results are as time goes on. And draw conclusions then.

Someone could start documenting results with different models

The poor results with the 500 I think are revealing. It will be interesting to see what other results come in.

600 mhz stable seems to be the gold standard for a average binned 2/3+ where as 650 stable or so would be a truly exceptional specimen. My 550 can post at 570, but its max stable is around 615-630 which I think is kinda typical for a 3+ 550 3+

Historically, to my knowledge the k6 processors were walling at 600 mhz or a little higher than that it was a limit of the architecture.

Back in the day, I had thought people were saying that 550Mhz was the limit of the k6 architecture, and beyond was just not feasible. Never mind the overclockers were already going past that; this is just what was said in causal talk. I had no idea that with the k6-2+ there actually was 570Mhz rated speed out there until years later, because by the time these mobile parts really were seen, people had already decided to move on. I had wanted one, but you could never really find them, and thought they never actually did 570Mhz until I saw some pictures. It's funny that there are so many brand new ones available, because it really shows how much these chips were skipped over back in the day in the market.

Nemo1985 wrote on 2022-02-21, 18:01:

Cool, at stock voltage?

Asus P5A min 2.1V. It feels stable @630MHz (a very quick test) but 660 no post.

fool wrote on 2022-02-21, 19:22:

Nemo1985 wrote on 2022-02-21, 18:01:

Cool, at stock voltage?

Asus P5A min 2.1V. It feels stable @630MHz but 660 no post.

Thank you... some years ago I tested my collection of k6-3+ 400 and they got the same results, more or less:

k6-3+ 450\acz v2.0
1a: 550? v2.0?
2a: 550 v2.0

k6-3+ 400/atz v1.6
3: 550 v1.80
2a: 550 v1.65
2b: 550 v1.65
2c: 550 v1.65 - 600 v2.2
2d: 550 v1.6 - 600 v2.0
2e: 550 v1.6 - 600 v2.1

Where the 600 mhz is not present it was because it wasn't stable even at 2.3

This is epic! I must confess I was among the ones 100% convinced this would be impossible, and wow I'm so glad i was wrong!

Hats off to all those involved, especially to Sphere478 for pursuing this and Fritzchens Fritz for providing the crucial evidence and being the first to test it!

[edit: typo]

https://en.m.wikipedia.org/wiki/List_of ... 2,_250_nm)

Someone was asking if it was possible to mod any of the non plus cpus,

I don’t see any that look like they may be + cpus in disguise.. so I’m pretty sure this mod is for cpus labeled 2+ only.

quicknick wrote on 2022-02-21, 19:32:

This is epic! I must confess I was among the ones 100% convinced this would be impossible, and wow I'm so glad i was wrong!

Hats off to all those involved, especially to Spehere478 for pursuing this and Fritzchens Fritz for providing the crucial evidence and being the first to test it!

indeeed!!! Thanks Fritz!!

By the way, guys if you do this mod please post. We could use a large data set on the results of these mods.

Primarily interested in max stable clock speed at 2.0-2.1v (confirm with prime95 or super pi for a hour or more) and a successful series of runs (say 200-300 which doesn’t take too long) using the prior mentioned memtest settings

If I was forced to guess next best slim chance K6-III in disguise... it would be 250nm mobiles... but only late production dates... curiosity might eventually entice me to look at the one I've got, but only after I get some acetone since that seems like the "low violence" way to do it. But I'd call it something like a one in ten shot... annnd I'd not expect to get a particularly fast K6-III out of it given that the same process made approximately 100Mhz slower K6-III than K6-2 in general.

BitWrangler wrote on 2022-02-21, 23:33:

If I was forced to guess next best slim chance K6-III in disguise... it would be 250nm mobiles... but only late production dates... curiosity might eventually entice me to look at the one I've got, but only after I get some acetone since that seems like the "low violence" way to do it. But I'd call it something like a one in ten shot... annnd I'd not expect to get a particularly fast K6-III out of it given that the same process made approximately 100Mhz slower K6-III than K6-2 in general.

Yeah I was kinda wondering about those, but I don’t think anything that isn’t 180nm is going to have this mod. Perhaps there is a similar version where they were disabling the cache on k63 non plus chips but I really doubt it, that makes no sense to me since they could just use regular k62 cores.

Anyone got one of those IR cams for a phone or similar device though? I figure if you power them on briefly at 2x50 or something low, you can tell by whether it makes a square or rectangular warm spot what core is under heatspreader.

BitWrangler wrote on 2022-02-21, 23:47:

Anyone got one of those IR cams for a phone or similar device though? I figure if you power them on briefly at 2x50 or something low, you can tell by whether it makes a square or rectangular warm spot what core is under heatspreader.

due to the properties of aluminum this won’t work. It reflects IR. Like a mirror.

But if you put tape over it, it may work.
But then we get into the thermal properties of aluminum. I don’t see us being able to identify cores this way.

I do have one of these camera’s though.

We need a 3d printed alignment tool for reinstalling the ihs squarely anyone able to make one?