Sphere478 wrote on 2022-01-23, 05:52:

Still haven’t been able to find anyone who owns a k6-3+ that has been de lidded or anyone willing to do it. Anyone? I just can’t bring myself to do it to my 550 🤣

You put that 550 away, dont you dare abuse that poor CPU.

*seriously tho, with the prices that CPU fetches on eBay I would slap myself for having such thoughts 🤣

TrashPanda wrote on 2022-01-23, 05:58:

Sphere478 wrote on 2022-01-23, 05:52:

Still haven’t been able to find anyone who owns a k6-3+ that has been de lidded or anyone willing to do it. Anyone? I just can’t bring myself to do it to my 550 🤣

You put that 550 away, dont you dare abuse that poor CPU.

*seriously tho, with the prices that CPU fetches on eBay I would slap myself for having such thoughts 🤣

ugh, I can't find anyone to help 🤣

If I end up with any dead K6-2+ or dead K6-3+ chips I'll delid them.

I think my fastest K6-3+ is 450Mhz. I do have 550 and 570 K6-2+ chips.

Shame there wasn't a K6-3+ 333.... they'd probably cheap enough to delid 🤣

BitWrangler wrote on 2022-01-28, 04:01:

Shame there wasn't a K6-3+ 333.... they'd probably cheap enough to delid 🤣

I believe there was a 3 non plus 333 but that's no help.

majestyk wrote on 2022-01-20, 07:57:

Chances are that AMD followed a price discrimination policy and sold faultless "castrated" III+ chips as 2+ to meet the demand for cheaper mobile CPUs.

I´m not sure if I remember this correctly but didn´t a larger cache (in the same CPU) mean a lower maximum CPU speed?

AFAIR the III+ chip dies were considerably bigger due to having extra cache on them, that would typically result in more load on the internal signal paths so extra latency cycles were added to help with that. Depending on how it was done, and manufacturing process limitations, you could either get extra latency but higher clocks, rarely same latency with higher clocks, often same latency/clock or even regression but made up for by the cache being larger now.

Point is, there is no reason AMD (or any other chip manufacturer) would cripple their best dies, expensive to make on the best process available, to sell them as parts with lower margins. Not even Intel was that rich and stupid, except when the dies were so easy to make it wasn't much of a loss to rebrand them. AMD at that time had it factories running 24/7 to meet demand and they wanted to sell the latest and greatest Athlons for big bucks. I just don't see someone making an insane decission to forget that, make more III+ cores instead of Athlons and then rebrand them as 2+ to sell at lower price.

As I see it, the only III+ cores that would be rebranded to 2+ (if any) were either defective cache dies or didn't meet other specs (clocks or power draw at rated speed). Now, it might be possible to find a part that is fully functional, and can somehow be unlocked to full cache size, but will require more juice to run and will generate more heat. But that would be like winning a lottery ticket I think. But, good luck, I just hope you people won't kill a ton of 2+ chips trying to delid them to find that one III+ core.

It all depends on the yields. If engineering delivers more than marketing expected, you'll generally see units being sold below their technical limits. Lots of examples of that, both in terms of clocks (P55C, Celeron Mendocino) and in terms of nerfed functionality (486SX, Celeron Coppermine/Tualatin etc). Of couse AMD would rather sell Athlons than under-clocked/featured K6 cores, but re-tooling a fab is a massive, long undertaking, and all decisions aroud that will have been taken long before the CPUs started rolling of the lines. If it turned out they had more fully working CPUs per wafer than expected, they would have to shift these essentially free units somehow and that traditionally involved making lower-price parts out of them.

Given the known good overclockability/undervoltability of late K6-2+/3+, it sounds like yields were far above requirement. The K6 family was discontinued for commercial reasons (Athlon), not an inability to actually produce the last models (as happened with K5).

I share your doubt as to whether the K6-2+ and 3+ actually shared the same die, however if they do, it's not at all farfetched to propose that a lot of the lower-price parts are just re-marked higher ones.

All I see is decoupling, no jumpers/bridges on K6s

dionb wrote on 2022-01-28, 12:07:

It all depends on the yields. If engineering delivers more than marketing expected, you'll generally see units being sold below their technical limits.

Look at it from money/finances point of view. Let's not forget these companies exist to make money, and selling fast CPUs is the means and not the goal. I have first-hand experience of i486SX-25 part happily working at 40MHz but these are 1993+ models, by that time Intel had no issues making them and was already focused on Pentium. Now this is someting I've read but apparently in the early i486 days it was easier to find and buy the more expensive DX-33 than SX-20. Why? The question should be why not - why would Intel even bother making the cheaper part if they were pretty much selling every expensive one?

Frankly the K6-III+ and 2+ look like pipe-cleaner products. These were made using 180nm tech when first Athlons were on 250nm still. Why would AMD even do that? Well, perhaps becuase Athlon was so great they wanted it out the door ASAP and 180nm was still somewhat unproven at that point. So AMD made a shrink of their by then well-understood core to work out any process issues. However once that was done why keep making those slower K6 if you have K7 that's the fastest x86 out there period and you can command premium prices for it?
Same with Duron overclocking, people often forget that by the time AMD released Durons they'v already had the Athlon core well polished and kept improving it, and the 180nm process was already clocking to 1GHz and even beyond. And more importantly, there were no longer any problems with meeting the demand so it made sense to expand the product line beyond the most expensive Athlons.

dionb wrote on 2022-01-28, 12:07:

I share your doubt as to whether the K6-2+ and 3+ actually shared the same die, however if they do, it's not at all farfetched to propose that a lot of the lower-price parts are just re-marked higher ones.

Well actually I might not have made myself clear, sorry. It does make prefect sense for 2+ to be a cut-down version of III+ due to defects (and it works well with the pipe-cleaner theory). My point was that it would be done to rescue as many expensive 180nm dies as possible, not to artificially segment the product. I have no proof of that, obviously, but the fact that this CPU was a short-lived run of older cores on a newer process, during a time of general CPU shortages, really make me question the logic of creating two different products form the same die just for marketing reasons.

In short, it's very well possible that every 2+ is a cut-down III+ (though we have no proof of that, wikipedia is not a reliable source and this topic would not exist if we knew) - but I seriously doubt there would be that many 2+ CPUs out there that could work without issues as III+ (again assuming there is a way to restore them to full size cache). Such CPU might exist but would be in fact so rare that it'd be easier and cheaper to just buy a III+ than look for a suitable 2+ specimen.

Either way, I just wanted to offer my two cents, this is an interesting subject but as I've stated I hope it will not lead to many 2+ CPUs getting destroyed in the process. K6+ CPUs are already stupidly rare and expensive as it is.

Deunan wrote on 2022-01-28, 14:41:

Frankly the K6-III+ and 2+ look like pipe-cleaner products. These were made using 180nm tech when first Athlons were on 250nm still. Why would AMD even do that?

This is just a small nitpick but the K7 Pluto was already 180nm and released in 1999 while the K6-III+ and K6-2+ were released in 2000. The regular K6-III was 250nm still, just like K7 Argon.
The Athlon was AMD's first CPU on 180nm.

Doesn't affect your overall argument of market segmentation though.

Edit: Maybe the '+' CPUs could rather be 'pipe cleaners' for on-die L2 cache?

Doornkaat wrote on 2022-01-28, 14:51:

Doesn't affect your overall argument of market segmentation though.

Edit: Maybe the '+' CPUs could rather be 'pipe cleaners' for on-die L2 cache?

Ha, I got the dates wrong. Well then, it actually does affect the argument because the mobile K6 core can't be a pipe-cleaner if AMD already had K7 shrink working at that point. In fact you just invalidated my entire point with that one sentence, good job 😀

Well yes, it could still be a test bed for on-die L2, possibly though AMD needed a mobile chip with even lower power draw than what K7 could manage. Or perhaps it was some contractual obligation to make mobile chips (or it could even be just a smoke screen for stock owners, Intel sure did similar things to claim they got a product out). So, unless it was some kind of process/tech test, then the reason for K6+ to even exist can be non-technical. And in that case it's entirely possible two different dies were made and my conclusions are wrong.

Deunan wrote on 2022-01-28, 15:08:

Doornkaat wrote on 2022-01-28, 14:51:

Doesn't affect your overall argument of market segmentation though.

Edit: Maybe the '+' CPUs could rather be 'pipe cleaners' for on-die L2 cache?

Ha, I got the dates wrong. Well then, it actually does affect the argument because the mobile K6 core can't be a pipe-cleaner if AMD already had K7 shrink working at that point. In fact you just invalidated my entire point with that one sentence, good job 😀

Well yes, it could still be a test bed for on-die L2, possibly though AMD needed a mobile chip with even lower power draw than what K7 could manage. Or perhaps it was some contractual obligation to make mobile chips (or it could even be just a smoke screen for stock owners, Intel sure did similar things to claim they got a product out). So, unless it was some kind of process/tech test, then the reason for K6+ to even exist can be non-technical. And in that case it's entirely possible two different dies were made and my conclusions are wrong.

Reading your post again I have to admit I must have really misread it the first time!
But then isn't misreading posts what the web is all about?😅😉

As far as I understood it, K6 2/3 plus were fabbed at the dresden plant as soon as it was upgraded to 180nm and it was the Duron that pushed them out of production there. Though they might have been making both concurrently for a bit but duron demand made them go 100% duron.

Any of the new posters have a k6 3+ they want to de lid to run some tests? Or one that’s already de lidded?

I haven't seen this mentioned, but most of the Slot-A Athlon CPUs we're sold as lower speed CPUs than what the dies were actually tested at and marked.

500 was often 650, 650 was often 800 or 850.

700 was often 850 or 900.

Sphere478 wrote on 2022-01-28, 04:48:

BitWrangler wrote on 2022-01-28, 04:01:

Shame there wasn't a K6-3+ 333.... they'd probably cheap enough to delid 🤣

I believe there was a 3 non plus 333 but that's no help.

I have a non + 333 🙁 it runs at 350 and 400mhz no problem, any higher and it wont post on any board no matter the voltage or bus speed set.

400 was the slowest K6-3+

ODwilly wrote on 2022-01-29, 03:23:

Sphere478 wrote on 2022-01-28, 04:48:

BitWrangler wrote on 2022-01-28, 04:01:

Shame there wasn't a K6-3+ 333.... they'd probably cheap enough to delid 🤣

I believe there was a 3 non plus 333 but that's no help.

I have a non + 333 🙁 it runs at 350 and 400mhz no problem, any higher and it wont post on any board no matter the voltage or bus speed set.

There must have been a reason they chose to include a random 333 in the k6 3 lineup. Maybe they had a bunch of em not passing muster.

Sphere478 wrote on 2022-01-29, 08:14:

There must have been a reason they chose to include a random 333 in the k6 3 lineup. Maybe they had a bunch of em not passing muster.

Actually, there was a K6-III/333AFR, K6-III/333AFK, K6-III/350AFK, K6-III/366AFK.....

All close enough together that it must have been down to more than just binning them.

debs3759 wrote on 2022-01-29, 08:22:

Sphere478 wrote on 2022-01-29, 08:14:

There must have been a reason they chose to include a random 333 in the k6 3 lineup. Maybe they had a bunch of em not passing muster.

Actually, there was a K6-III/333AFR, K6-III/333AFK, K6-III/350AFK, K6-III/366AFK.....

All close enough together that it must have been down to more than just binning them.

Okay, dang, I didn’t even realize that there were 350s and 366s

For those just tuning in we are still waiting on someone to show up who has a k6-3+ or a k6-2+ and has de lidded it or is willing to. We need high res pics and resistance measurements.

At repoman11

If you wanna measure all possible combos on those tiny pads or something and document it that wouldn’t hurt.