WD1772 at 16MHz: actual bench results
Since nobody in forty years has ever published how many chips they tested, I thought I'd better stop complaining about it and go and do it. Here's what I've got so far.
Method
- All tests on the same H5 board, same drive, same write protected reference disk.
- Every chip that failed at 16MHz was then run at 8MHz to prove it is a working part and not just dead silicon.
- "OK 10 mins" means it ran my endurance test for that long with no errors. Longer runs to follow.
Results
Code: Select all
Marking Date Unit 16MHz result 8MHz check
------- ---- ---- -------------------------- ----------
02-02 9427 1 OK, 10 mins -
02-02 9427 2 OK, 10 mins -
02-02 9522 1 OK, 10 mins -
00-02 8614 1 OK (earlier testing) -
00-02 8618 1 would not boot loads STOS OK
00-02 8618 2 would not boot loads STOS OK
00-02 8618 3 would not boot loads STOS OK
00-02 8622 1 OK, 15 mins -
00-02 8622 2 4 bombs, would not boot loads STOS OK
00-02 8622 3 OK, 12 mins -
00-02 8627 1 OK, 10 mins -
00-02 8627 2 OK, 10 mins -
00-02 8627 3 OK, 10 mins -
Code: Select all
Date Chips Worked at 16MHz
---- ----- ---------------
8614 1 1 of 1
8618 3 0 of 3
8622 3 2 of 3
8627 3 3 of 3
9427 2 2 of 2
9522 1 1 of 1
- The unconditional claim that only the 02-02 will run at 16MHz is dead. Seven of my 00-02 chips ran fine.
- Every chip that failed at 16MHz booted and loaded STOS perfectly at 8MHz. So these are good chips that simply won't take the overclock, which is exactly what you'd expect of a part rated at 8MHz.
- The failures are not spread evenly. 8618 failed three times out of three. Every other date code I have produced working chips. That looks like one bad lot rather than a general 00 versus 02 difference.
- It is not a case of later being better either. 8614 works and 8618 is four weeks later and doesn't, so there's no neat progression to point at.
- Three chips per date code is not a lot. Treat all of this as a first pass, not gospel. But it is three more than anyone else has ever counted.
What ten minutes actually proves, and what it doesn't
Worth being clear about the scale here, because ten minutes of my endurance test is not ten minutes of normal use. It is ten minutes of continuous formatting, writing and reading with the clock sitting at 16MHz the entire time, which is a far harder life than any real machine gives a floppy controller.
That matters for my own 1.44 floppy kits, because they only switch the controller to 16MHz while the drive is actually being accessed. The chip spends the vast majority of its powered life back at 8MHz doing nothing. Ten minutes of solid hammering is comfortably more 16MHz operation than most people's kits will see in a year, so on this evidence every chip that passed is fine for that job. The ones that wouldn't boot at all are a different matter and would be no use in any kit, at any duty cycle.
What ten minutes does not tell us is whether these chips will take hours of unbroken 16MHz. That's the case that matters for the factory Mega STE and TT arrangement, and for the older kits that flipped to 16MHz and simply stayed there, and it's where the overheating reports have always come from. So the next round is long runs, several hours each, and that's where I'd expect any real difference between these chips to show up. Passing ten minutes and surviving an afternoon are not the same claim and I'm not going to pretend they are.
On the second line of the marking
My 00-02 8614 carries a second line reading C026028, and my 02-02 8901 carries 013556010101 instead. C026028 is Atari's own part number for the WD1772, which is listed in Thorsten Otto's Atari chip reference:
http://www.tho-otto.de/hypview/hypview.cgi?url=%2Fhyp%2Fchips_x.hyp&charset=UTF-8&index=555
I did wonder whether the two different schemes pointed at different supply or assembly routes, but it doesn't look like it. Two other people's chip inventories list a 00-02 8527 marked c026028 and an 02-02 8909 marked 013850010101, which is exactly the same split I've got here:
https://www.atari-forum.com/viewtopic.php?t=14733
http://atariage.com/forums/topic/128642-wanted-tos-14/
So the second line follows the mask revision rather than telling us anything about where the chip was made.
I did briefly wonder whether Atari had vetted these chips and stamped their own number on the ones that passed, but that doesn't stand up either. In 1986 the ST clocked the FDC at 8MHz exactly as specified, so there was nothing to screen for, and the hand selecting Best Electronics describes was for the TT high density upgrade years later. A C0 number just means Atari bought the part in volume and gave it an inventory number, the same as the Hitachi keyboard controller carries C070122-002.
Checking my own chips backs that up. The only one here carrying C026028 is the 8614, which I tested years ago and which almost certainly came out of a real ST. Everything else in this batch is broker stock that never went through Atari's production, and none of it carries an Atari number, working or failing. So the second line tells you where a chip has been, not what it can do.
Worth stating plainly for the table above: my sample is almost entirely loose broker stock, and my one Atari marked chip is also my oldest, so there was never a comparison to be made there. If anyone wants to test whether parts pulled from real STs behave any differently, that's the way to do it, but I wouldn't expect it to show anything.
Still worth writing down the full marking rather than just the date code though, in case a pattern turns up that I can't see from four date codes.
Testing hundreds of chips isn't realistic, it would take months. But if a few people run the same test on their own chips and post the markings with the result, we'd have a proper picture reasonably quickly, which beats forty years of everyone quoting the same undated packet radio text at each other.
(in theory anyway) 
