Monthly Archives: January 2020

2.88 MByte on 3.5 Inches – The floppy that never succeeded

When the floppy first saw the day as IBM 23FD with 8 inch and 80 KBytes and read only capablilites in 1971, it was a big step from punch cards.
In today’s terms 8 inch and 80 KBytes are really clunky for almost nothing to store on it. It only took until 1976 when the Shugart company developed then then-new 5 1/4 inch system with around 100 KBytes initially.
These 100 Kbytes were later raised by nobody else than Steve Wozniak to 140 KBytes with a different recording scheme (GCR, Group Coded Recording), which was adopted by other companies as well.
Tandon brought this to 360 KBytes “Double Density” in 1978.

Back to where we started: Sony introduced their ​3 12 inch format in 1981 with a formatted capacity of 160 KBytes, originally for their own machines (SMC 70). It originally didn’t spread wide due to the dominance of 5 14 inch disks. But it was adopted by the Microfloppy Industry Committee in 1982, albeit with some changes to Sony’s original specs – that was the point when this drive size took off, as it was implemented by both Atari (ST) and the Commodore Amiga as their default format.

Sony invented the 3 12 inch disk originally for their SMC 70 computer

One of its advantages are better durability due to its casing as well as its sliding door, which prevented touches and dirt coming in. Not to forget its much smaller dimensions, which made them fitting for most pockets.

Size wasn’t its advantage in the first place – it started as single-sided disks on the PC with 360 KBytes – the same as the 5 14 inch counterparts. Data was written using MFM encoding (as opposed to Apple’s still used GCR – which made PC/Amiga/ST and Apple formats at the time incompatible), the switch to two sides in 1986 brought 720 KBytes. Only a year later, 1987, IBM introduced the HD format in their PS/2 line – and that’s the floppy we know until today.

When I went over to a friend, it wasn’t unusual for me to take a stack of floppies with me to get the newest things he has…

What if this stack could have been half the size, because these disks could fit double the later-usual 1.44 MByte?

Introducing the ED floppy disk

IBM introduced so-called “Extended Density” (ED) drives with 2.88 MByte formatted capacity as early as 1988 in their high-end PS/2 line, but you might have heard about it – it more or less failed to reach its goals and only left us PS/2 ports and a 180° rotated MCA bus slot for VLB. NeXT used it in their systems beginning with 1991. How did they work, and why didn’t they take over?

ED drives are a special kind of breed and require several changes, to medium, drives and also controllers.
The disk itself spun with 300 rpm, the same speed all other drives of this size had.

Sony MP-F40W-23 with boeder ED disks

The higher data density called for media improvements. The disk used Barrium ferrite, which looks darker and was also thicker (3 um vs. 1.2 um of the HD disk). The thickness had a reason: Do you remember that funny Hitachi video Get perpendicular? That was 2005 – but ED floppies pulled the trick to store bits vertically on the medium more than a decade earlier! This type of recording plus the better media meant that ED floppies are more reliable in the long term.

ED Barrium Ferrite (bottom) is darker than HD cobalt-doped iron oxide

In Intel’s tech notes about the ED format (AP-358) this is documented quite well: Bits are stored vertically instead of horizontally!

Due to its higher data density, the floppy controller must be capable of handling the higher frequencies – the data rate doubles from 0.5 MBit to a whopping 1 MBit of raw data – which you’ll never get, as the unformatted capacity is 4 MB but formatting and IBM typical MFM encoding brings it down to 2.88 MB of usable data. IBM MFM encoding uses a constant angular velocity (CAV), meaning there is as much data in the inner tracks as there is on the outer ones. There could be more on the outside, but in order to keep the bits stable on the inside, the outside is required to use as many bits as the inside in this mode, even though a track is much longer there.

Earlier ISA/VLB controllers are usually not capable of handling these drives, integrated controllers of the Pentium era though should be capable of handling them (easy check: If it’s integrated and the BIOS has that option, it’s supported for sure, also known to be compatible: Winbond W83877F, Intel 82077AA, 82077SL, NS PC8477AV, Adaptec AHA-1542). If you have an intel 82077AA or 82077SL controller chip or derivate (produced from 1991 on) on your controller, you should also have a chance.

Then there’s drive compatibility: Drives work differently on signalling the ED data rate to the controller. I have a Sony MP-F40W-23 and had no issues on am Ali Alladin V chipsets, but your mileage may vary.

Why didn’t they take over?

There’s no final answer, but there are indications, each contributing to the bad adoption.

Compatibility: As described, controllers and drives had to be compatible – not to mention BIOS support. A new drive also required a new controller, raising the costs of an upgrade. If there wasn’t BIOS support, you were screwed as well – only a new board (at the time) could possibly fix it. The different signalling of the drives didn’t help, either. You had to get the right combo.

Time: By the time BIOS support had been established (early to mid 90s), the 1.44 MByte High Density drives had alreay been spreaded widely. Once this had been established, it would have been hard for a basically incompatible format (special drives needed) to succeed. It’s similar to the original 3 12 to 5 14 inch situation – only this time there was nobody pushing the format.

Mediocre improvement: While 2.88 MByte may have been much in 1988, the extremely slow and niche adoption meant that it was too few, too late in already the early 90s. CD-ROM was already coming, providing easier access to much more data, and CDs were cheap to produce! Zip media have been introduced in late 1994, offering a lot more space for data at roughly the same physical size with much better speed.

Price: Producing a 2.88 MByte floppy is said to be more complex and hence media have been MUCH more expensive (5 times to 1.44, according to this German article).

So there we were stuck with our stack of HD floppies. If there’s a reason to get them, it would possibly mostly be for nostalgic reasons. However, if you really want to get into them, or just need floppies for whatever weird reason, these are the main advantages over HD: Higher reliability and higher data rate. This is also the reason why this format has seen wider spread in industry purposes. If you can copy an existing game or any other piece of data over to an ED disk, it will not only last longer on that disk, also loading times will decrease.


How quick are they? Here’s a test done on a system with a Sony MP-F40W-23 connected to an Ali Alladin V chipset. Measurements were done by copying 1.3 MByte file to disk and back via Norton Commander using a stop watch. No disk cache was enabled.

1.44 MB: 22.21 KByte/s write, 24.18 KByte/s read
2.88 MB: 27.5 KByte/s write, 28 KByte/s read

That’s not too mindblowing. Faster, but only marginally. A test with a system with Via chipset resulted in similar measurements. My suspicion is that, while the data rate is higher, there are still the same tracks. Moving the head to the next track is not faster between the formats, and when the head has moved, it has to wait for the first sector – and rotation speed hasn’t changed, either. So while the data rate per track is higher, and there are less tracks and hence less head movement, it’s not much faster using ED disks, unfortunately, due to this overhead. For this, rotation speed, head motor speed or logical sector 0 position would need adjustments.

Fun facts

Did you know? ​3 12 inch are actually 90 mm in diameter and hence is 3.54 inches! Their name has been “adjusted” to the existing formats, its dimensions have been metric in their design (like the other designs, as well).

The holes in the disk have the same distance as an A4 paper punch hole. This allowed to store them with documents (albeit it seems that, depending on the folder, you way need to use some force).

The drive

The drive used for testing is a Sony MP-F40W-23. It has some quirks: It requires you to choose the drive ID by using a switch on its side. Moreover, Sony has, for whatever reason, put the connector upside down – meaning that if you put a connector with a directional nose into it, you are putting it into the wrong direction. Only noseless connectors will work – and they have to be put in in the wrong direction – pin 0 to the drive inner.

The drive electronics are a little more than your usual HD disk drive. Assuming that this tech is not so widespread and is probably a little less integrated, this makes sense.