Sourcing Legit MC68882 FPUs - Data

[zigzagjoe]

Compiled this data earlier today to assist in determining which 68882s are fake (sanded off and remarked) and which are legit. It's usually quite obvious by poor quality of markings, or old masks with new date.

Date codes are the last 4 digits of the last line on the CPU center. Given a date code of QEJQ9019, this is year 1990, week 19

Some guidelines:

1. Markings should be well defined, the text fine and the logo legible
   1. If the text is thick, it's probably fake.
   2. no logo = fake!
2. Font should be closely spaced, narrow and taller characters
3. Freescale logo only on chips 2005 or newer
4. Older masks will not significantly overlap years with a newer mask
   1. ie. OC35H made in 2005 is fake
5. Markings will not be removed by acetone
6. PGA: some missing pieces of letters is acceptable
7. PGA: will have engraved text on the ceramic that corresponds to actual mask
8. PGA: country of origin usually on ceramic (uncommon to be on the gold)
9. PGA: gold Y on copper should be bright, not dark

Gallery of fake 68882s:

https://imgur.com/a/J3WZkDs

Here is the compiled data so far, in order of age of mask. This was sourced by looking at vintage equipment with 68882, and other verifiable sources - ie. not something purchased by hobbyist off ebay. Not at all complete, and it's obviously harder to get verifiable sources past 1994 or so given it was an obsolete part by then.



Full data: https://docs.google.com/spreadsheets/d/1z4L56dWWPeJ-x0bQX8XpUEkuwVoOnD27/edit#gid=261616021

Regarding clock frequencies, anything 1C12R or newer should do 50mhz without problem regardless of markings. OC35H should be able to hit 50 usually as well, but not guaranteed. The takeaway here is with most masks being good for 50mhz, even fake ones have a pretty decent chance.

 

[ymk]

Thanks for the info. Looks like I have a fake based on the date:

MC68882FN40A
1C12R
QEGU0509

I'm running it at 31MHz and it seems to work fine.

 

[zigzagjoe]

Thanks for the info. Looks like I have a fake based on the date:

MC68882FN40A
1C12R
QEGU0509

I'm running it at 31MHz and it seems to work fine.

Yes, I'd say it's probably fake. While I don't have an authoritative idea how many were produced, it seems like many were OC35H mask or newer, outside of early uses like the SE/30, which gives rather good odds of 50mhz operation and anything less is a shoe-in.

For the PLCC chips, I think there may be a chance that stampings on the bottom may give the true date of packaging.... unfortunately, I've just accidentally sold on my last loose PLCC so I can't verify this.

 

[eharmon]

It might also be helpful to cross-reference with the sources I have listed on this page for plausible mask revision years. Many of the old Moto product data catalogs explain when masks were introduced and cancelled.

eharmon | Motorola 68k Processors

Index of mask revisions for 68k processors used in Macintosh

eharmon.net

At some point I want to update the tables there to add PLCC packaging and FPUs...haven't gotten around to it though.

 

[obsolete]

This inspired me to check the one I bought for my Reloaded build. I bought the cheapest one I could find on eBay, because I only need it to do 16MHz, so who really cares, right? Here are the pictures:



Week 17 of '90 with a 0C35H mask checks out. The markings are not affected by acetone at all. Seems legit! Makes sense to me that nobody is trying to fake an early mask and low speed grade. The markings on the bottom don't seem very helpful, just Malaysia and 7.

 

[zigzagjoe]

This inspired me to check the one I bought for my Reloaded build. I bought the cheapest one I could find on eBay, because I only need it to do 16MHz, so who really cares, right? Here are the pictures:
View attachment 69841
View attachment 69842

Week 17 of '90 with a 0C35H mask checks out. The markings are not affected by acetone at all. Seems legit! Makes sense to me that nobody is trying to fake an early mask and low speed grade. The markings on the bottom don't seem very helpful, just Malaysia and 7.

Yup, I agree on all points. If it shows a date in the 90s, odds are it's real.

Dang. Occasionally you'll find molded pointers corresponding to date in the bottom from the plastic molds, but I guess Motorola wasn't doing that.

 

[zigzagjoe]

Some more interesting info.

Per this doc, MC68882CFN denotes an extended temperature range. https://datasheet.octopart.com/MC68882RC16A-Freescale-Semiconductor-datasheet-7549409.pdf

1J23S is the same as 2C12R, just made in Japan: https://web.archive.org/web/20030201235213/http://e-www.motorola.com/collateral/PCN5210.html

1J23S/2C12R use the same process as 0F91C mask MC68030 per that same doc, so this would make it a 0.8um. Looking at dates, I'd assume that likely means 1C12R is one process larger.

 

[obsolete]

View attachment 69841

Week 17 of '90 with a 0C35H mask checks out. The markings are not affected by acetone at all. Seems legit! Makes sense to me that nobody is trying to fake an early mask and low speed grade. The markings on the bottom don't seem very helpful, just Malaysia and 7.

I am happy to report that this humble 25MHz 0C35H part seems to run just fine at 47MHz in my SockeBooster. The system booted and completed a Norton benchmark without crashing, at least. Is there a good way to stress test the FPU to ensure reliability at higher speeds? Also, would it still be worth trying to source a later mask part for lower heat output?

 

[zigzagjoe]

I am happy to report that this humble 25MHz 0C35H part seems to run just fine at 47MHz in my SockeBooster. The system booted and completed a Norton benchmark without crashing, at least. Is there a good way to stress test the FPU to ensure reliability at higher speeds? Also, would it still be worth trying to source a later mask part for lower heat output?

These are the tests I perform on each booster card to make sure they test fine with a known good FPU.

Basic tests: Can it boot and open speedometer without crashing, does Mac Test Pro test the logic board OK (includes FPU tests), and do the speedometer scores look consistent with expected.

To stress it I run 500 reps of one of the Speedometer FPU tests. The values returned should be constant, I've noticed an OC35H running too warm will start getting random values or return 0 scores entirely. The average score may flip negative at some point, this is an overflow and nothing to be worried about.

I had an 16mhz OC35H that would pass basic tests but fail during stress when it was fully hot; a small heatsink was enough to resolve that. Heat output isn't really a concern as the poor FPU is being roasted by the toasty video GALs underneath it. I've actually moved the FPU slightly in the newer revision of card I'm working on now. A later FPU will only very slightly reduce temperature and be guaranteed to work at 50mhz, so it's not worthwhile it if the OC35H tests OK.

 

[trag]

I think I may have bought fakes for my SE/30 reloaded boards. The date code on all of them is QEED0823, which puts the year before the week, and 08 is later than any of the dates in the table. Unless they were made in 1923. Also the emblem is Motorola's not Freescale's.

Mask label is 1C12R. PLCC chips.

According to my Ebay history the seller was ketech_electronic, but when I click on that name in my purchase history, it leads to gc_supermarket.

I bought them about 10 months ago. I haven't tested them at all yet.

Mine do have Malaysia and 12 on the back. The front lacks the 5 letter code at the bottom edge of the chip.

 

[zigzagjoe]

I think I may have bought fakes for my SE/30 reloaded boards. The date code on all of them is QEED0823, which puts the year before the week, and 08 is later than any of the dates in the table. Unless they were made in 1923. Also the emblem is Motorola's not Freescale's.

Mask label is 1C12R. PLCC chips.

According to my Ebay history the seller was ketech_electronic, but when I click on that name in my purchase history, it leads to gc_supermarket.

I bought them about 10 months ago. I haven't tested them at all yet.

Mine do have Malaysia and 12 on the back. The front lacks the 5 letter code at the bottom edge of the chip.

I just ran into the same with a chinese seller I had previously established as having a supply of 1J23S FPUs. New name, same listing number but contents all different...

I'd agree that it seems likely those are remarked, but for a reloaded board any mask should work (so long as it's actually a functional 68882).

 

[Phipli]

Just to chip in, as stupid as it is that things are relabeled, make sure you do test them. "Fake" is a bad YouTube clickbate name for them, what most of them are is relabelled, and some are relabelled / dishonesly up-spec'd. Most chips bought like this work, and most of them work at the labelled speed.

I'm not in anyway defending the practice, it drives me up the wall. I'd often rather an honest 16MHz part than a repainted 25MHz part, but... try not to over react. Just test and validate your budget parts.

I once got a "6MHz" Z80 that happily ran at 20MHz. The re-grading can work both ways (when you're not buying too good to be true, 75MHz 68060s for £2).

 

[trag]

Thank you, gentlemen, @Phipli , @zigzagjoe . I appreciate the reassurance. I wasn't too worried for the reasons you mention and I'll find something I can test them in before applying any solder. I mostly posted to add another report to what's going on.

At the price they sell these for, it's amazing they can afford to relabel them. Of course, if it's a non-working part, that's even cheaper, but large PLCC packages aren't cheap, even without a working die installed...

 

[obsolete]

Rolled the dice on eBay on one that looked like it might be legit, and this is what showed up. Total fake. Bummer.

 

[Iesca]

I don't know if the one I bought awhile back for my LC III is fake or not, but it seems to be recognized correctly by the system!

EDIT: I guess mine is probably fake, unless it was genuinely manufactured in 2008!

 

[theirongiant]

I bought one off eBay and it came in that black plastic packaging, but it *appears* to be legitimate. I paid around $30.

It passes the Speedometer tests with flying colors. I ran 100 iterations of the "Benchmark Mix," and 100 iterations of the FPU benchmark.

The numbers are as follows:

MC68882FN33A
1C12R
QEWT9320

This is running in a Macintosh LC III (68030@25MHz). I haven't done the overclocking trick yet, which would bump this up to 33MHz, but given your data indicating that the FPU can survive running up to 50MHz, I guess I'll be alright.

I bought some Raspberry Pi-sized heatsinks a few weeks ago. They're 20mm x 20mm x 10mm — just a bit over 0.875" square and around 3/8" tall. They fit perfectly atop the CPU and FPU, and there is still room between the heatsink and the Farallon Ethernet card in the PDS slot.

 

[theirongiant]

Here is a picture of the FPU, sourced from eBay, in my LC III with a tiny heatsink on the 68030.