famicomaster2
Experienced Member
Hello again, I come to show another part of my disk collection off. This time drives from Computer Memories Inc, CMI, best known for being the drive IBM chose to walk away from for the AT. That was specifically the 6426S.
Top left to bottom right:
CM-5619 (306/6/17 for ~15MB formatted)
CM-6426 (640/4/17 for ~20MB formatted)
CM-6426S (640/4/17 for ~20MB formatted) The "S" variant was primarily a board level revision to improve the performance of the drive. It also included a new EPROM with better seek acceleration curves.
CM-6640 (640/6/17 for ~40MB formatted) The little known big brother to the 6426.
CM-3426 (615/4/17 for ~20MB formatted) One of very few half height disks CMI ever produced.
CM-5410C (256/4/17 for ~8MB formatted) Part of the "C" series produced at low cost for OEM usage.
CM-5412 (306/4/17 for ~10MB formatted)
CM-5616C (256/6/17 for ~12MB formatted) The largest of the "C" series drives.
Can you tell that CMI really liked this chassis, introduced with the 5000 series? The same frame, casting, and lid are all used for the 5000, 6000, 7000, and 8000 series.
I suppose it could be that they had no money as well. The 3000 series bears a striking resemblance to several IMI drives, but I'm not sure if this was on purpose or not.
Unfortunately, this part of my collection is rather incomplete, but I'm lucky to have the models I have. CMI drives are very hard to find, especially working as many of mine are. There are actually two other drives not shown, a pair of 6426Ses, one of which works but has horrendous bearings and the other which has lost it's servo bursts. I've had to part them out to fix a couple of these rarer drives a few times already.
If you have or are aware of where to find more CMI drives, PLEASE drop me a line! I am especially interested in the 4000 series half height drives and depending on the working condition I might even place a high bounty upon their head.
Yes, you heard that right by the way, the 6000 series has TWO feedback systems! There is an optical encoder on the end of the servo motor's shaft which reports, I believe it's every 0.5 degrees of rotation. In addition to this, there are "servo bursts" of high frequency, high amplitude noise between the data tracks. Because of the precision of a servo motor, the onboard microprocessor, a 6803, can pretty accurately position the heads over the track it wants. Sadly, however, there is NO absolute positioning on the drive itself, this is relegated to the controller as with most ST-506/ST-412 type drives.
To put it simply, the heads are "inaccurate" or "wide" enough to "hear" the noise pulses. A discriminator on the drive separates them from any data under the head. The pulses are spaced even and odd timing, such that they do not "play" at the same time. In theory, all the drive needs to do to make sure it is centered over a data track is to count the number of servo pulses until it is in the general area, and then balance the amplitude of the servo pulses.
Unfortunately, in practice, this BARELY works. With servo data missing, as with one of my parts drives, the drive will typically go into one of two behaviors: It will seek infinitely (early firmware, especially 16K/32K units) or it will retry it's seek test 3 times before going on to act like everything is normal (every 64K unit I've seen). Unfortunately, this often results in the drive using guesswork and really dumb estimation to "sort of" get where you ask it to be, leading to instant destruction of data. You can determine which revision your drive is by looking at the EPROM in socket Z20 on the logic PCB. Some drives have a 2716, some have a 2732, and later drives (most commonly the 6426S) have a 64K ROM. It is worth noting that "K" here is for Kilobits and not kilobytes.
There were three big issues with this design:
1. The precise head positioning was almost entirely analog in nature, requiring extremely high precision calibration to be reliable. In a hot computer, the values of components can change, especially with age, and so many drives will "lose" their servo tracking despite the actual markers being present on the disk surface just fine.
2. Some drives will pass their seek test without the servo tracking data! This leads to immediate and total data loss in many cases.
3. During manufacturing, the drives themselves actually write their own servo bursts! The "Servowriter" was attached to TP6 and TP17 on the drive, and a special firmware EPROM was intalled into the socket at Z20. This process required the drives to be thermally stable before writing the bursts. When IBM ramped up production of the AT in response to higher than expected demand, CMI's quality suffered greatly. Drives were no longer given the several hours they needed to warm up before writing!
All of these flaws led CMI drives to get a very bad name in the computing industry and IBM not to renew their contract at the end of 1984. This ultimately killed CMI, with customers flocking to competitors and their largest buyer choosing to return all units at the slightest hint of failure to be replaced (at their expense) by a more expensive, high end Seagate ST-4038.
It's a shame what happened to them, but the industry learned from their mistakes... Kind of. Stay tuned for more of my collection and more of my history / engineering lessons, if that's your kind of thing.
If you've read this far, I hope you've enjoyed or that you learned something, or maybe both. If you liked this, maybe you'd enjoy my similar post on Kalok drives, which you can see here.
Top left to bottom right:
CM-5619 (306/6/17 for ~15MB formatted)
CM-6426 (640/4/17 for ~20MB formatted)
CM-6426S (640/4/17 for ~20MB formatted) The "S" variant was primarily a board level revision to improve the performance of the drive. It also included a new EPROM with better seek acceleration curves.
CM-6640 (640/6/17 for ~40MB formatted) The little known big brother to the 6426.
CM-3426 (615/4/17 for ~20MB formatted) One of very few half height disks CMI ever produced.
CM-5410C (256/4/17 for ~8MB formatted) Part of the "C" series produced at low cost for OEM usage.
CM-5412 (306/4/17 for ~10MB formatted)
CM-5616C (256/6/17 for ~12MB formatted) The largest of the "C" series drives.
Can you tell that CMI really liked this chassis, introduced with the 5000 series? The same frame, casting, and lid are all used for the 5000, 6000, 7000, and 8000 series.
I suppose it could be that they had no money as well. The 3000 series bears a striking resemblance to several IMI drives, but I'm not sure if this was on purpose or not.
Unfortunately, this part of my collection is rather incomplete, but I'm lucky to have the models I have. CMI drives are very hard to find, especially working as many of mine are. There are actually two other drives not shown, a pair of 6426Ses, one of which works but has horrendous bearings and the other which has lost it's servo bursts. I've had to part them out to fix a couple of these rarer drives a few times already.
If you have or are aware of where to find more CMI drives, PLEASE drop me a line! I am especially interested in the 4000 series half height drives and depending on the working condition I might even place a high bounty upon their head.
Yes, you heard that right by the way, the 6000 series has TWO feedback systems! There is an optical encoder on the end of the servo motor's shaft which reports, I believe it's every 0.5 degrees of rotation. In addition to this, there are "servo bursts" of high frequency, high amplitude noise between the data tracks. Because of the precision of a servo motor, the onboard microprocessor, a 6803, can pretty accurately position the heads over the track it wants. Sadly, however, there is NO absolute positioning on the drive itself, this is relegated to the controller as with most ST-506/ST-412 type drives.
To put it simply, the heads are "inaccurate" or "wide" enough to "hear" the noise pulses. A discriminator on the drive separates them from any data under the head. The pulses are spaced even and odd timing, such that they do not "play" at the same time. In theory, all the drive needs to do to make sure it is centered over a data track is to count the number of servo pulses until it is in the general area, and then balance the amplitude of the servo pulses.
Unfortunately, in practice, this BARELY works. With servo data missing, as with one of my parts drives, the drive will typically go into one of two behaviors: It will seek infinitely (early firmware, especially 16K/32K units) or it will retry it's seek test 3 times before going on to act like everything is normal (every 64K unit I've seen). Unfortunately, this often results in the drive using guesswork and really dumb estimation to "sort of" get where you ask it to be, leading to instant destruction of data. You can determine which revision your drive is by looking at the EPROM in socket Z20 on the logic PCB. Some drives have a 2716, some have a 2732, and later drives (most commonly the 6426S) have a 64K ROM. It is worth noting that "K" here is for Kilobits and not kilobytes.
There were three big issues with this design:
1. The precise head positioning was almost entirely analog in nature, requiring extremely high precision calibration to be reliable. In a hot computer, the values of components can change, especially with age, and so many drives will "lose" their servo tracking despite the actual markers being present on the disk surface just fine.
2. Some drives will pass their seek test without the servo tracking data! This leads to immediate and total data loss in many cases.
3. During manufacturing, the drives themselves actually write their own servo bursts! The "Servowriter" was attached to TP6 and TP17 on the drive, and a special firmware EPROM was intalled into the socket at Z20. This process required the drives to be thermally stable before writing the bursts. When IBM ramped up production of the AT in response to higher than expected demand, CMI's quality suffered greatly. Drives were no longer given the several hours they needed to warm up before writing!
All of these flaws led CMI drives to get a very bad name in the computing industry and IBM not to renew their contract at the end of 1984. This ultimately killed CMI, with customers flocking to competitors and their largest buyer choosing to return all units at the slightest hint of failure to be replaced (at their expense) by a more expensive, high end Seagate ST-4038.
It's a shame what happened to them, but the industry learned from their mistakes... Kind of. Stay tuned for more of my collection and more of my history / engineering lessons, if that's your kind of thing.
If you've read this far, I hope you've enjoyed or that you learned something, or maybe both. If you liked this, maybe you'd enjoy my similar post on Kalok drives, which you can see here.
This also means that my index pulse is likely fine. The difference of about 0.40Hz, though likely not significant enough, is probably due to measurement error on my oscilloscope.