Fri 29 Jan 93  8:34
By: Yousuf Khan
To: All
Re: FAQ: 386, 486, SX, and DX?
------------------------------------------------------------------------

The next few messages contain frequently asked questions and their
answers. Feel free to comment.

                               ***

Q1) What is the difference between 386SX, 386DX, 486SX, and 486DX
processors?

                               ***

A1) All 386 & 486 processor models are "32-bit" CPUs. What this means is
that the processors' registers are large enough to load and store data
that is upto 32-bits in length. It does not matter if the processor
model is designated SX or DX, it is still 32 bit.

One misconception about these processors is that their maximum memory
addressibility is only 4 Gigabytes. They can all address considerably
more than that, upto 64 Terabytes. Most of it is virtual memory --
memory that is not real, but simulated from disk or some other storage
media.

Now for the differences. From a software developer's point of view a 486
does not look very different from a 386, and it isn't. A 486 is an
uprated version of the 386, with considerably re-engineered internal
circuitry, which generally allows it to execute equivalent instructions
faster. So given a 386 and 486 rated at the same clock frequency, the
486 will generally be faster. The extra speed of the 486 is due to its
extra wide internal data pathways (128 bit vs. 32 bit for the 386),
which allows blocks of data to flow back and forth concurrently, instead
of consecutively. There is also an internal ram cache, which is a buffer
between the 486 and the slow external ram which could cause the 486 to
wait too long for memory access.

The difference between SX and DX depends entirely on whether you are
talking about 386 or 486. In general, an SX is an economy version of the
full DX processor, usually with some feature removed from it.

In the case of the 386, the SX and SL (an uprated SX for use in low
power notebook computers) versions have a smaller external data path;
16-bit vs. 32-bit for the DX. This smaller data path does not mean that
the 386SX's 32-bit registers are disabled, it simply means that the SX
has to take two passes to fill the same register. This is done
automatically by the processor's own circuitry, and the software
developer does not have to do anything different when writing software
for 386SX that one wouldn't do with a 386DX. Naturally this can mean
that an 386SX will run somewhat slower than a 386DX, given the same
clock frequency. The 386SX also has had some of its address lines
removed. This results in a lower physical memory addressibility, but its
virtual memory addressibility remains at 64 Terabytes. The Intel and
clone 386SXs can address upto 16 Megabytes, the Intel 386SL can address
32 Megabytes, and the Intel and clone 386DX can address 4 Gigabytes
physically.

IBM has a specialized version of the Intel 386SL called the 386SLC,
which has added an internal 8K cache. It is otherwise pin compatible
with the regular Intel 386SL.

AMD and C&T Corp both have clones of the Intel 386SX and 386DX. They are
in general very much the same as their Intel counterparts and are direct
replacements of each other, both in terms of pin compatibility, and
software compatibility.

In the case of the 486, there are different variations within the SX
theme. Intel, Cyrix, and IBM all have made different modifications to
the 486DX design to come up with their economy 486.

The 486DX is distinguishable from all 386 and economy 486 offerings
because it has an internal math coprocessor. The math coprocessor used
to be a separate add-on extension to a processor, because all its extra
functions required more silicon than was available on the CPU. With
increased miniaturization techniques, the 486DX was able to incorporate
the circuitry directly into its own die.

Intel's variation of the economy 486 is called the 486SX.  The 486SX has
had its internal math coprocessor cut, but all data paths and address
lines remain intact. The Intel 486SX and 486DX are not pin compatible
with each other, so one cannot simply remove a 486SX and place a 486DX
in its place or vice-versa.

Cyrix has two levels of economy 486s: the 486SLC and the 486DLC. For the
486SLC, Cyrix has cut the math copro too, but it has also reduced the
size of the internal cache to 1K from 8K, and repackaged the CPU to be
pin-compatible with a 386SX CPU. Since it is pin compatible with an
Intel 386SX, the 486SLC can only address 16Megs of memory, and it has a
16-bit external data path. Cyrix is positioning its offering as a direct
replacement of the 386SX. For the 486DLC, all the same modifications
that Cyrix made for the 486SLC apply, except that it is made
pin-compatible with an Intel 386DX rather than 386SX. Since the 486DLC
is pin compatible with the Intel 386DX, it will be able to address 4
Gigabytes, and it has a 32-bit external data path. The Intel 486SX and
the Cyrix 486xLC are not pin-compatible with each other, but they are
software-compatible.

IBM has come out with its own variation of the 486, confusingly called
the 486SLC. IBM's 486SLC is not pin-compatible with the Cyrix 486SLC, it
is pin-compatible with the Intel 386SL/IBM 386SLC, however. The IBM
486SLC does not have a math copro built-in, but it has a larger internal
ram cache than all other 486s, 16K. Because of its 386SL/386SLC
heritage, the 486SLC incorporates the same power-saving features, and it
can address upto 32Megs, 16-bits at a time.