A 'once and for all' declaration regarding L88M mask MC68040 CPUs

[jessenator]

I'm interested to see what you find!

I let my MC68040LRC40 (running at 33 MHz because it's in my unmodified Quadra 650 board), with heatsink, heat soak without a fan, but with the cover off for a while. I topped out at ~49 C (with my Harbor Freight IR thermometer so take as much salt as needed). I might let it run on my full-40 as well.

Edit: I'm getting about the same shelf on my 40 MHz running one as well.

 

[herd]

Are you guys changing voltages? I know on later chips, like a 7445 vs a 7447, the newer chip will use around the same power as the old... at the same voltage. The advantage is that the new chip can usually reach the same speed as the old with a lower voltage, and that reduces the power use.

 

[Fizzbinn]

Test results using in an unmodified 25MHz Quadra 605 with no drives or RAM SIMM installed, all temps measured with a FLIR One Pro after system was running for 15 minutes (with blinking boot disk icon).

Baseline, original Quadra 605 CPU:
.65 μm - 49.0C - Q605 XC68LC040RC25 E23G

.80 μm - 65.8C - Q700 XC68040RC25B D50D
.65 μm - 54.7C - Q630 XC68040RC33M E31F
.57 μm - 57.1C - eBay ”MC68040RC33A L88M”

…so I would have expected a real L88M to run cooler than the E31F, not sure there is another explanation. Perhaps the measurements under load would be better.

Images:

 

[Fizzbinn]

Are you guys changing voltages? I know on later chips, like a 7445 vs a 7447, the newer chip will use around the same power as the old... at the same voltage. The advantage is that the new chip can usually reach the same speed as the old with a lower voltage, and that reduces the power use.

I don’t think so… I think all 040 Mac CPUs operate at 5 volts? With process size shrinks my general understanding if nothing else changes you realize power savings. Of course there could be other tweaks (bug fixes?) done at the same time that could complicate things…

 

[trag]

.65 μm - 54.7C - Q630 XC68040RC33M E31F
.57 μm - 57.1C - eBay ”MC68040RC33A L88M”

…so I would have expected a real L88M to run cooler than the E31F, not sure their is another explanation. Perhaps the measurements under load would be better.

There can be huge variation amongst chips -- even chips whose dies came from the same wafer. Slight differences in the level of doping and layer registrations can cause changes in power consumption and heat production.

At least, that was all true later with 45 nm and 300mm wafers. I'm not certain those "facts" apply to the chips made 15/20 years earlier.

Anyway, a 3 degree temperature difference is less than I typically found when characterizing chips from the same 45nm wafer. So I wouldn't be horribly surprised that some .65um chips run cooler than some .57um chips. Again, assuming the same level of variation applied 15/20 years earlier.

When they bin the parts, they just have to meet the specs at the extremes of temperature and voltage. If they do that they can get the 40 MHz (or whatever) rating. That leaves a lot of room for them to draw different levels of power while operating.

 

[Nathan_A]

FWIW, my Interware Booster CL40-40 has an XC68040RC40 on it, and it's without a heatsink. Because it ends up positioned upsidedown over the original mainboard and a heatsink would create physical interference (not to mention get immediately thermally saturated anyway given the tight space and lack of airflow).

 

[Fizzbinn]

FWIW, my Interware Booster CL40-40 has an XC68040RC40 on it, and it's without a heatsink. Because it ends up positioned upsidedown over the original mainboard and a heatsink would create physical interference (not to mention get immediately thermally saturated anyway given the tight space and lack of airflow).

Interesting, what mask version is it? (the last 4 characters in the upper right hand corner)

 

[Fizzbinn]

There can be huge variation amongst chips -- even chips whose dies came from the same wafer. Slight differences in the level of doping and layer registrations can cause changes in power consumption and heat production.

At least, that was all true later with 45 nm and 300mm wafers. I'm not certain those "facts" apply to the chips made 15/20 years earlier.

Anyway, a 3 degree temperature difference is less than I typically found when characterizing chips from the same 45nm wafer. So I wouldn't be horribly surprised that some .65um chips run cooler than some .57um chips. Again, assuming the same level of variation applied 15/20 years earlier.

When they bin the parts, they just have to meet the specs at the extremes of temperature and voltage. If they do that they can get the 40 MHz (or whatever) rating. That leaves a lot of room for them to draw different levels of power while operating.

That all makes good sense, thanks. What one would then expect is that a ”MC68040RC40A L88M” (if I had one) would run cooler than the 33 one in my test setup. i.e. 2 chips out of a L88M production wafer/batch are binned as 33 and 40 respectively based on power consumption/heat production test results.

Perhaps a hypothetical explanation (using my temps as stand ins for factory power consumption/heat production test results):

.65 μm - 54.7C - Q630 XC68040RC33M E31F
- Binned 33 as temp was between 54 and 62 (a cooler side E31F 33)

.57 μm - 57.1C - eBay ”MC68040RC33A L88M”
- Binned 33 as temp was between 50 and 58 (a hotter side L88M 33)

The above binned temperature ratings being lower for L88M as the newer, smaller process should generally produce cooler chips.

 

[jessenator]

Perhaps a hypothetical explanation (using my temps as stand ins for factory power consumption/heat production test results):

I'd say that's a very reasonable hypothesis

As much as I'd like to say my HF thermometer is comparable to your FLIR results, I may have to take my brother-in-law up on letting me borrow his so i can put my chips through their paces with a singular test board.

Tangentially related, I need to modify my 3D-printed 040 puller to be more like this PGA tool from DayStar to make removal a bit less of a PiTA

Just for giggles until I get a more scientific-method-compatible setup, I put my LRC40 under load and overclocked to 44 MHz with the same heatsink setup, just to see, and it still sticks right at 49ºC even under load for a considerable duration. Guess that means the heatsink is doing its job.

 

[Fizzbinn]

Tangentially related, I need to modify my 3D-printed 040 puller to be more like this PGA tool from DayStar to make removal a bit less of a PiTA

I'd be interested! I'm still using a DIP/IC puller, trying to be a careful as I can.

 

[Nathan_A]

Interesting, what mask version is it? (the last 4 characters in the upper right hand corner)

Looks like the number there is 02E31F.

 

[jessenator]

Looks like the number there is 02E31F.

At least it was the first die shrink! Looks to be the same mask as the Quadra 840av (at the one shared on Wikipedia).

Here're a few 040s and the Macs they came from (where applicable)
Quadra 700 | Centris 650
Quadra 650 | (bought pulled)

I didn't attempt to overclock the Quadra 700's CPU to 33, but now I'd be interested to see where it gets to thermally. I bought the E42K 040 from Herb Johnson (I bought another on a whim from a SoCal-based seller). Incidentally, did any Quadra/Centris machines ever ship with the D43B mask, I wonder? My Q700 is the D50D, so pre-die-shrink anyway.

Here's an 040 Quest Completionist's hunt: get all the mask revs

Also, I've got my printer up and working again, so I'll hopefully have my new puller design assembled tonight and able to test it.

I made a narrower version of the "pulling" piece, since some of the 040 Performa/LC machines have tight quarters around the socket. The "tower" piece (I call it) should fit in the gap where the mounting feet on the socket are, so it should fit either way. Once I've at least tested the thing to where it works, I'll share it on the 3D Print thread.

 

[jessenator]

3D-Printed Objects

I've made a better back shell. Not tried printing any of this yet, though - the side vents/speaker grilles might not come out that well... But if anyone is interested, I have attached STLs

68kmla.org

it works!

 

[pax]

I bought this from utsource.net in January after a member wrote they were successful in sourcing authentic ‘040s there. Comparing the mask with the tables above I’m guessing this is a fake?

 

[pax]

By the way, thank you @jessenator, the puller came out great.

 

[jessenator]

I bought this from utsource.net in January after a member wrote they were successful in sourcing authentic ‘040s there. Comparing the mask with the tables above I’m guessing this is a fake?

There's every possibility that there are other masks, perhaps eng samples. I felt clever with my whole "G23E is E23G backwards!" thing earlier in the thread, but I can't and won't say that the list of masks is 100% exhaustive. There's lot that was never publicly released, and probably never will be about the line as a whole.

I would test it. Tattletech or MacBench should root it out—it'll at least tell you if there's an FPU. Maybe Norton system benchmark from the era will do it?

the puller came out great.

Nice! I'm printing my newest version, just to see if I can refine it a bit more, but glad it's working for you!

 

[nickpunt]

To add my experience, I've got 2 'L88M' chips:
1. Freescale LRC040A (acq 2021, mfg 0603, prob from same buyer as you @jessenator ) - runs hot bot not scalding without heatsink at 40mhz
2. Motorola RC040A (acq 2014, mfg 0430) - runs warm without heatsink at 33mhz

I haven't compared temps side by side though (#2 is in storage), but I remember #2 being quite cool to the touch at least relative to XC's. Both verified in tattletech as full 040s. I *think* they're both genuine L88Ms, the date codes lined up and they weren't freshly printed looking.

 

[jessenator]

1. Freescale LRC040A (acq 2021, mfg 0603, prob from same buyer as you @jessenator ) - runs hot bot not scalding without heatsink at 40mhz

I'm going to say with 99.98% germ-killing certainty it was the same seller, who is now sold out of them, incidentally.

I 'accidentally' did a temp test today of one of the LRCs operating at 33 MHz and it was hovering at around 48-50 C. I then did an actual test at 40 MHz and found it above 50 C at times, and feeling much hotter to the touch. Hopefully one of these days I'll actually go to my BiL's and borrow his FLIR and get to add to @Fizzbinn 's collection of numbers and images

In light of the impromptu tests, I decided to also take a closer look at them to see if seeing them all together would be able to weed out any glaring inconsistencies, but I think I've been staring at them for too long...

Only the eBay chips appear to be the outliers with the printing. The right-angle mark in the upper left looks almost blurred compared to the others. But having not seen a high-res scan of any 040 from the Freescale era, I can't say it means much there, but it is mildly suspicious when you compare the two E42K chips... IDK, more derailing my own thread in the interim, I guess.

 

[cheesestraws]

The right-angle mark in the upper left looks almost blurred compared to the others

Compared to the C650 one they don't look much blurrier to me?

 

[Bolle]

Check the other numbers in the lower left corner. As far as my data collection goes those do give away the mask as well.
As far as I can tell it's like this:

D98D ends with -010
E31F ends with -001
E42K ends with -101
K63H have seen them with both -104 and -103
L88M ends with -104