486SX vs 486DX: A closer look

When Intel released the 486 CPU in 1988, there was no SX or DX designation on it. But similarly to what they had done with the 386, they followed up with an economy version. And they borrowed the same name designations they’d used on the 386. But the difference between the economy version and the premium version wasn’t the same in the two chip generations. In this blog post, we will explore the 486SX vs 486DX.

486 improvements over the 386

The 486 improved on the 386 in several ways. It was more efficient than the 386, so it got more work done on a clock cycle by clock cycle basis than the 386 did. It was a better improvement on its predecessor in this regard than the 386 had been. The 486 also included some additional instructions, although not a lot of software took advantage of that. An even bigger improvement was including 8 KB of level 1 cache right on the CPU. The 386 could use external cache, but including cache on the processor die gave a speed improvement.

But there was a fourth change in the 486 design, and this is the one it tends to be most remembered for. The math coprocessor, which sped up mathematical operations involving decimals, also moved on to the processor die rather than being an optional separate chip.

The coprocessor, or FPU, ended up being critical to the future direction of the 486.

The economy 486SX vs the performance 486DX

The actual difference between the 486 DX and SX has long been a topic of debate.

There is a very persistent rumor that Intel had yield problems with the 486, where the processor worked fine except for the math co-processor. So, the rumor went, they disabled the co-processor and sold those units as a less expensive chip called the 486SX, borrowing the naming convention from the 386 series. Just like the 386, the higher end 486 gained the DX designation.

This may be a case of people misremembering the story behind the 60 and 75 MHz Pentiums, which were the result of early yield problems. But I do remember hearing these rumors even in the early 90s, although I don’t remember where. I can’t find any mention of yields driving the release in the early 1991 coverage of the chip.

Others argue that Intel intended to be able to disable the FPU all along, and when they released the 486SX in April 1991, they just disabled the FPU in enough chips to meet demand and labeled them SX chips.

This practice was wasteful, so once the 486SX proved it would sell well, Intel just omitted the math co-processor portion entirely in later revisions of the chip. Building chips with the FPU and then disabling it was a temporary measure that let Intel react quickly. And in April 1991, they needed to react quickly with a new budget CPU. More on that in a minute.

Clock rate differences

The other difference between the 486SX and 486DX was the clock rate. The 486SX was available in 16, 20, 25, and 33 MHz models. The DX was available in 25, 33, and 50 MHz models before giving way to the clock multiplied DX2 at 50 and 66 MHz and DX4 at 75 and 100 MHz. Intel only released the 50 MHz 486SX2 under duress.

The 486SX chips running at 16 and 20 MHz didn’t last as long on the market as the 25 and 33 MHz versions.

But other than the presence of an FPU and possibly the clock rate, the SX and DX were the same chip. They both were full 32-bit processors. That contrasts with the 386SX and DX, where the SX was a 16-bit chip externally, like a 286.

The reason for the 486SX

The 486SX solved a problem for Intel. Intel had not licensed the 386 to any second sources like they had with previous CPUs. And they pushed the 386 hard, launching the Intel Inside marketing campaign to take back as much market share as it could from those second sources.

But in March 1991, AMD released its own version of the 386. It was completely compatible with Intel, and they even created a version that ran at 40 MHz, while Intel topped out at 33 MHz.

The 486SX, released about a month later, allowed Intel to compete with the new AMD chip without lowering the price of its higher end 486 CPUs. Intel charged about $250 for a 486SX, vs $750 for some of its 486DX models.

Even though the 486 was more efficient, AMD could still compete. A 40 MHz 386 with external cache could outpace a 486SX at 16 or even 20 MHz, and hold its own even against a 25 MHz 486SX, thanks to the combination of its higher clock speed and higher bus speed. The price difference was enough that you could buy a 40 MHz 386 with external cache for less than a 25 MHz 486SX. And furthermore, if you decided later that you needed the math coprocessor, you could get a 387 coprocessor for much less than the cost of replacing your 486 with a version that had a math co-processor.

What about the Intel 487?

Intel released a math co-processor for the 486SX, would was called the 487 or 487SX. It plugged into a separate socket on the motherboard next to the original CPU just like the 387. But the 487 was really a full CPU that, when present, disabled the original 486SX CPU. So it wasn’t really a co-processor at all, it just looked like one. And it cost just as much as the 486 DX, if not slightly higher.

The later Intel Overdrive upgrades put an end to the shenanigans, claiming to be a clock multiplied upgrade, which is exactly what they were. They came in versions that could directly replace a socketed 486 processor, or in a version that fit the 487 socket, making them usable on systems with a soldered 486 SX processor.

How the Intel 486 competed against AMD

Intel won this marketing war by pitting the 486SX vs 486DX vs AMD’s 386DX. It segmented the market, protecting the profit margins on its higher-end chips, while cutting AMD’s margins.

AMD didn’t get into most of the top-tier brands, and when they did, they didn’t necessarily stay there very long. Regardless of whatever incentives Intel may have given directly, the makers of inexpensive PCs benefited a lot from the Intel Inside campaign. That campaign gave the perception that the computer you bought didn’t matter as long as it had an Intel processor in it.

And from an economy of scale standpoint, splitting a product line between the 386 and 486 wasn’t necessarily ideal. If you could produce one motherboard that you could clock at 16, 20, 25, or 33 MHz, and then put the appropriate 486 CPU for the price target, and then adjust the amount of external cache and the amount of video memory, you could use the same motherboard in dozens of different combinations to match whatever price point you and your retail partner wanted to meet.

AMD became a favorite with smaller system builders, and the 40 MHz 386 retained a dedicated following for several years because the AMD-based systems really were a good value. AMD couldn’t afford to play to win at this point, they were playing to stay in the game. That’s called an infinite game.

The AMD 486

In April 1993, AMD released its own version of the 486, including both SX and DX variations. A year later, AMD also did something Intel didn’t want to do. They created clock doubled versions of the 486 SX. Intel responded by creating a 50 MHz 486SX2, but they never made a 66 MHz version. AMD’s 66 MHz SX2 isn’t a common chip, but I can confirm it existed. I bought one in the summer of 1994, and I still have it today.

The introduction of the Pentium CPUs eliminated the SX and DX designations for good. By that time, games that made use of the math coprocessor were common enough that the math co processor was a selling point. Instead, Intel segregated its market based on clock speed, clock multiplier, and bus speed.

How much difference 486SX vs 486DX really made

How much difference the DX made over the SX was always a point of contention. You’ll find people who insist it didn’t make much difference, and you’ll also find people said the SX wasn’t worth having at all.

But there wasn’t as much software in the 486 era that used the FPU as we remember. Falcon 3.0 was the most popular game that did. Even Doom didn’t use the FPU. There were a number of professional applications, like AutoCAD, where the FPU made a big difference. A couple of other examples included Fractint, which used the FPU to calculate and draw fractals, and Povray, which used the FPU to draw ray-traced images.

I don’t mean to downplay those titles, because for the people running those titles, the FPU was indispensable. But the people I remember making the biggest deal about the FPU at the time weren’t running those titles.

When you saw a difference when you were running Microsoft Word or playing Doom, it was often due to other factors. The SX was all about hitting low price points, so the SX systems frequently didn’t have any level 2 cache. 486DX and DX2 systems often did. Also, the people making the comparisons didn’t always compare like for like anyway. They may have been comparing a 25 MHz SX against a 66 MHz DX2. And you would notice that difference.

For a large part of the market, especially the home market, as long as you were running the same speed chip on the same configuration of motherboard, you didn’t really notice a difference. The floating point unit was for bragging points, until the Pentium era and Quake.