The following picture shows four essentially identical Intel processors in the top row:
The real difference is that some of them are fabricated on an older process and thus sport a larger die size than others. (They’re also not all rated to operate at the same frequency, but that is not a design difference).
The more obvious difference is the labeling. From plain chips to i386 to i386 DX. The chips neatly illustrate an important chapter in Intel’s history. The leftmost chip was manufactured sometime in early 1988 and doesn’t look very different from the original 386s, or any other Intel chip of the era. It’s simply a slab of gray-brown ceramic with etched markings. The only noteworthy feature is the double-sigma (ΣΣ) marking indicating a 386 which reliably performs 32-bit operations.
The CPU looked boring, because it was simply a microchip locked inside a computer and a normal user never got to see it, not unless something went wrong. There was just no reason to make it look like anything special.
The next chip looks extremely similar (and it is!) but the chip designation changed from A80386 to A80386DX. This chip was manufactured around mid-1988 and is just slightly newer. At that time, Intel was in an complex position. The 386 was an undisputed speed king in the world of PCs, and only Intel could manufacture 386s. But 386s were quite expensive and had to compete against much cheaper 286s… which a number of companies manufactured.
To improve its position, Intel decided to release a cheap 386, the 386SX. SX stood for Single-word eXternal, as opposed to Double-word eXternal (DX) of the “real” 386s. Because the 386SX used a 16-bit external bus, existing AT system designs could be relatively easily adapted to work with it. The 386SX was a intended to reduce the 286 market share and thus increase the market share held by Intel, and improve Intel’s position as the only 32-bit x86 chip supplier.
The original 386 was retroactively renamed to 386DX. This is reflected in Intel documentation—pre-1989 manuals talk of 80386, later releases instead mention 386 DX or even Intel386 DX instead.
The next chip can run faster (at 33MHz) but is otherwise 100% compatible. Yet the visual difference is striking. There’s a large Intel logo and an even larger i386 logo. This chip was manufactured in early 1991. In the late 1980s, Intel undertook a major effort to build its own brand. Instead of being just a chip supplier for PCs from IBM, Compaq, AST, or Zenith, there was a concerted effort to brand Intel CPUs as the heart of a PC (and not just a PC, see e.g. the Sun 386i).
This was again directly tied to the fact that 386 was the first Intel processor which was not licensed to second-source manufacturers (AMD, Harris, Siemens, etc.). Intel proactively built up its own brand in order to make it harder for any alternative CPU vendors to compete. The same chip markings can be seen in the following 1988 ad:
The rightmost chip is a variation on the previous one. The Intel logo is the same, but the i386 logo is smaller to make room for the enlarged DX designation. There’s also an extra ® and two more ™ stamps, a sure sign that the legal department had been busy. This design is visually similar to Intel’s 486 processors sold at the time.
The final 386 (in the bottom row), manufactured in late 1992, is shown just for completeness. It’s a different S-spec (SX544 vs. SX366) than the previous one and displays 1985 and 1987 copyrights, rather than just 1985.
Similar treatment was also applied to Intel’s 387 FPUs, two of which are shown in the image at the beginning of the article. The one on the left was manufactured in mid-1990 and sports only a large white Intel logo. The one on the right is from late 1992 and adds large white “i387 DX” as well as the obligatory ® and ™ decorations (it’s also rated at 16-33 MHz, which is unusual for CPU/FPU chips).
With the 486, the DX/SX designations of course took on a completely different meaning, but the basic idea remained—DX was the “real deal” while SX was the cheaper, performance-limited variant. But that’s another chapter in Intel’s history.
I assume you’ve seen this, the Intel 80386 Business Case. The first 386 I ever had was a SX16, and with a bit of luck I traded it and some cash for a 386DX-16… But the XMS memory board wasn’t a 32bit one, instead it was a 16bit ISA board, so it ran just as slow as the SX16.
But I really have to admire the 80386, as it brought 32 bit computing from the VAX/SUN generation to the masses. And we played a lot of DOOM on that thing!
Yes, the 386 Business Case video is excellent.
My first PC was a 386SX-25 and it lasted a long time, in part because it had 4MB RAM (SIMMs, no funny memory cards). I eventually upgraded to a 90MHz Pentium (about the time when 133MHz Pentiums just came out) and everything seemed so, so fast!
But the amazing thing about the 386SX was that it could run just about anything. Windows 3.1 in Enhanced mode, OS/2 Warp, DOOM, Watcom 386 compilers… it wasn’t fast but it could sure do stuff.
Now we live with the legacy of the 386, and wish Intel had just skipped the blasted 286 🙂
Yea, I wished Intel didn’t wait until so late to release the 386SX given the 286’s problems.
No kidding.. IBM skipped the 80186, if only they could have skipped the 80286, but I guess the need for an ‘advanced’ IBM PC in 1984 was simply too great.
I mean, who could resist!?!
Yeah… people wanted faster PCs, and the 386 was way, way too late. Everyone else had 32-bit CPUs at the time. A 286-based PC was unfortunately inevitable, and if IBM hadn’t done it, someone else would have.
IBM should have just gone with the 68000 from the beginning. The world would be different today 🙂
That’s actually an interesting question. Why did it take Intel so long to release the 386SX? They had the 8086/8088 before, so they must have been well aware of the advantages. Maybe the 386 was too rushed to have the DX/SX parts right away? But they still could have added a SX a year later instead of waiting until 1988.
I think the big delay for the 80386 was the whole iAPX 432 disaster. Shades of Itanium from the early 1980’s. But as always back to an improved 8086, and all is golden.
Intel couldn’t release the 386SX until late in 1988 at the earliest. Late 1987 there were multiple stories of shortages of 386 CPUs and concerns over grey market reimportation of unmarked chips that failed 32-bit validation. Intel also planned on devoting a large portion of early 1988 production to replacing returned 386s. If Intel can’t make enough 386 chips to meet demand at $1000, there was no way they could divert excess production to get packaged with slightly fewer pins at $200. There were some 386SXs in 1989 but what I remember was those were close to a paper launch using chips that failed speed validation. It wasn’t until 1990 that the 386SX arrived in actual volume. No matter the technical merits, the greatest CPU is completely useless if it can’t be bought. The 286 was faster, cheaper and available for purchase.
Entering “InfoWorld 1987 386 shortage” in google books would verify this.
This is a very good point… I’ll remember it the next time someone pulls out the old “OS/2 should have been 32-bit from the beginning” nonsense 🙂
Anyway… Intel did have the option to license the 386(SX) to other manufacturers. They chose not to, and had trouble satisfying demand, but the decision was entirely theirs. And I’m not at all implying the decision was wrong!
Oh yeah, Intel did a pretty good job of erasing the iAPX 432 from corporate history. They must have been extremely embarrassed about the fiasco.
Itanium was more of the same, but even more insane. All those RISC CPUs couldn’t beat x86, but this one would just because the label said “Intel”? Not likely…
On 32-bit OS/2: When OS/2 1.3 was released, IBM staff stated that doing 1.3 delayed 32-bit OS/2 by a year. By that time, most of the users likely to pay about $1000 for OS/2 install and increased memory would be using 32-bit systems. I liked 1.3 a lot but getting 2.0 out faster would have been better than improving the 16-bit version.
One thing that did surprise me when I went looking at the OS/2 1.3 release articles was how negative IBM was on previous versions of OS/2. Challenge for OEMs to sell 1.21 when IBM calls it a poor product. A little later, OS/2 1.3 was disparaged and all forms of outlandish promises were provided for OS/2 2.0 (like a DOS 4 look-a-like replacement shell for corporate use) along with statements suggesting that the 386 was obsolete (in 1991! before OS/2 2.0 shipped!). I just can’t think any of this helped OS/2 gain traction or encouraged OEMs to ship OS/2.
Itanium would have worked out better if Intel had managed to keep to the planned schedule. Itanium was about 2 years late. Original timeline would have resulted in Itanium being twice as fast running x86 code and 5 times faster with specialized Itanium code compared to the fastest Pentium variant. Charging 10 times as much for twice (or more) performance is easy. Delayed Itanium and AMD’s unwillingness to slow development left Itanium only marginal better than x86 at certain tasks but still costing much more which is never a good market position.
On the other hand, doing the Win3.0 compatibility took time too, not to mention drivers, and it would have been a good opportunity to fix the infamous SIQ problem too. I tend to assume that with MS’s support, MS/IBM would have released OS/2 2.0 in late 1991 and 2.1 in late 1992, following a yearly release schedule.
The story I heard from IBMers is that OS/2 2.0 was delayed by about a year because Microsoft quit and IBM basically had to put together a whole new development team, and of course it took them a while to get up to speed. And it can’t have been a year because of that and a year because of 1.3, because that would have put the release at early 1990, while Microsoft barely shipped the first OS/2 2.0 before the end of 1989. Who knows, perhaps both where a factor.
As for outlandish claims, it’s worth reading what Microsoft was saying about the future of DOS in 1983 or so 🙂
Did IBM in 1991 or so actually care about OEMs shipping OS/2? I wouldn’t be so sure. That was always something for Microsoft to take care of…
Yes, Itanium might have fared better if they actually managed to ship it on time, but I see no reason to think that it would have helped much. Time and again, x86 beat the faster and more expensive architectures. The original timeline was probably a pure fantasy. And a big weakness of the Itanium was that the only way it got decent performance was by adding gigantic L2/L3 caches, which was always going to significantly hurt the price/performance ratio. I think Itanium was pure hubris on Intel’s part, but we’ll never know for sure since they did not manage to ship on time 🙂
It seems that the Red X campaign encouraging the move away from the 286 toward the 386SX dates back to 1989.