• Please review our updated Terms and Rules here

Why wasn't the 386sx CPU developed as a drop-in replacement for the 286?

The cost of the 386 and a bare bones build with just the motherboard and RAM was very high back then.
 
Depends on the year. Checking with a 1991 issue of PC Magazine, the 20 MHz 386 was $50 more expensive than the 386SX-20 and the 386DX motherboards were about another $50 more than a 386SX motherboard. The DX motherboards generally had twice as many SIMMs. One could load up on a faster 386 and cache and install extra memory and make the 386 very expensive but a barebones 386DX system cost 10% more than the 386SX system. The SX would be 10% slower than the DX at the same clock speed. Performance minded went DX; budget minded went 286; those with lavish tastes went 486.

Early in 1992, the prices of the 386SX fell to match the similarly clocked 286 and the 386SX finally offered respectable performance for the price.
 
Depends on the year. Checking with a 1991 issue of PC Magazine, the 20 MHz 386 was $50 more expensive than the 386SX-20 and the 386DX motherboards were about another $50 more than a 386SX motherboard. The DX motherboards generally had twice as many SIMMs. One could load up on a faster 386 and cache and install extra memory and make the 386 very expensive but a barebones 386DX system cost 10% more than the 386SX system. The SX would be 10% slower than the DX at the same clock speed. Performance minded went DX; budget minded went 286; those with lavish tastes went 486.

Early in 1992, the prices of the 386SX fell to match the similarly clocked 286 and the 386SX finally offered respectable performance for the price.

All true. I held out for the 486SX. I discussed this previously just a little while back. I bought a case, motherboard and the 486SX for $700 as a package. It came from some outlet around Chicago O'Hare that I found in Computer Shopper, as they had the best deal at the time. I can remember a 486/166 going for around $450. To me, the 133/166 really made computing take off and the speed was fantastic for the time. Thanks Kreb for responding.
 
Were people still purchasing 80286 systems in any quantity in 1992? My first 80386DX system was a Mylex 20MHz system purchased before that. 80386 systems were being advertised in the likes of Computer Shopper in 1988.
 
Last edited:
Were people still purchasing 80286 systems in any quantity in 1992?

According to PC/Computing magazine, 286 and 386 sales were tied at 40% each in August 1991. By February 1992, 386 had around 55% of the market while the 286 was down to about 15%. By August 1992, it was 60% 386, 20% 486, and 5% 286.
 
According to PC/Computing magazine, 286 and 386 sales were tied at 40% each in August 1991. By February 1992, 386 had around 55% of the market while the 286 was down to about 15%. By August 1992, it was 60% 386, 20% 486, and 5% 286.
Like many things the 286 was at its best, most numerous and cheapest near the very end 1991/1992.

Lots of people screwed by lack of software considering it’s best years were after it was obsolete.

I don’t have the link but several Intel designers reminisced about the 386dx through sx development cycles during an interview.

The original 386dx supported a 16 bit AT bus but was extremely slow.
The sx was actually more difficult to build and Intels yields pushed its mass availability back years.
They had to include a passive 256 byte linear cache (onchip) and more transistor’s in the 386sx for it to perform even at a similar level as a 286.
Intel was loath to communicate that the cheaper 386sx was actually more complex than the dx.
Intel never marketed the sx in higher clocks until competitors forced it to.

It’s packaging was to make it cheap to manufacture with no provision to upgrade and Intel definitely did not want it to be upgradable or retail packaged for consumption.

A more perplexing/unnecessary chip is the 387sx.

Intel pulled a lot of dirty tricks to kill the 286/386
 
On my 1990sZenith 286LP Plus you remove a small daughter board with 286/12 cpu and 287 co-pro on it and slide in a daughter board with 386sx12 and socket for a suitable mats co-pro.

I purchased my first and only brand new x86 system. It was a 286/16 generic clone in 1991. It rompt over my mates 386sx/12 system because it had to double tap when processing 16 bit code iirc.
 
Last edited:
Philips/Magnavox did a similar daughterboard design on some of their desktops-- the manual I had showed it as supporting a 12.5MHz 286 or a 16MHz 386SX. This always seemed like a dramatic level of overengineering for a low-end machine. I don't think the daughterboards were really a sold item, so I guess the primary appeal was that the "rest" of the motherboard could be made in higher economies of scale, and then they didn't have as much risk if they mis-estimated how the demand for new machines broke.
 
This is a quote:

By 1990 286s were “commodity” CPUs, with an average sale price of $16; 386s were still expensive,
with an average sale price of $121 in 1990 and $104 in 1991. (486s even more so: $600 in 1990, $384 in 1991.)
 
March 1991 was when AMD came out with their 386 cpus so prices dropped for Intel's 386 also. I jumped from a 286/12 Packard bell to a home built AMD 386DX/40 sometime in 1992 I believe.
 
This is a quote:

By 1990 286s were “commodity” CPUs, with an average sale price of $16; 386s were still expensive,
with an average sale price of $121 in 1990 and $104 in 1991. (486s even more so: $600 in 1990, $384 in 1991.)
Link please.

A quote without the source is worth dilly squat......
 
Philips/Magnavox did a similar daughterboard design on some of their desktops-- the manual I had showed it as supporting a 12.5MHz 286 or a 16MHz 386SX. This always seemed like a dramatic level of overengineering for a low-end machine. I don't think the daughterboards were really a sold item, so I guess the primary appeal was that the "rest" of the motherboard could be made in higher economies of scale, and then they didn't have as much risk if they mis-estimated how the demand for new machines broke.
With respect the the Zenith systems a 386 with the 386sx daughter board wear sold as separate systems. With the fact they could run MS Windows 3.x in enhance mode. The fact is Standard mode is actually faster than Enhanced mode........

I read that in the early '90s and just upgraded my 286/16 system to 4megs using 30 pin SIPPs. When I replaced the 286 mobo with pre used a 486 mobo with 486DX33 cpu in it I just desoldered the SIPP pins making them 30 pin SIMMs and added them to the 4megs already on the 486 mobo for a total of a whopping 8megs of ram.
 
This is a quote:

(486s even more so: $600 in 1990, $384 in 1991.)
I like how you leave out 1992 when “486’s” went down to $82 to kill 286/386 demand and leave out the $1000 spread that existed high to low clocks.


Link please.

A quote without the source is worth dilly squat......
The above is why 286 chips moving away from pga already didn’t make sense when they did. It was only cost effective a year and did damage to manufacturing volumes .

Any cost advantage at an system integration level was lost by segmentation of the market reducing volumes of each chip package, it also meant you couldn’t deliver chips almost POS to minimize costs. (PGA is much easier to install a chip before sale without breakage)

The 387sx was always curious as to why it even existed as a separate and unreliable insta break package.
Either it and the 287 should have shared this package/ pinout if it was so much better or it should not have existed
 
Last edited:
I like how you leave out 1992 when “486’s” went down to $82 to kill 286/386 demand and leave out the $1000 spread that existed high to low clocks.



The above is why 286 chips moving away from pga already didn’t make sense when they did. It was only cost effective a year and did damage to manufacturing volumes .

Any cost advantage at an system integration level was lost by segmentation of the market reducing volumes of each chip package, it also meant you couldn’t deliver chips almost POS to minimize costs. (PGA is much easier to install a chip before sale without breakage)

The 387sx was always curious as to why it even existed as a separate and unreliable insta break package.
Either it and the 287 should have shared this package/ pinout if it was so much better or it should not have existed
But they did, sort of. When the fast 286s arrived, Intel converted the 387SX into the 287XL using a 40 pin DIP. Intel might not want to sell a fast 286 but they weren't willing to concede the coprocessor market. The chip had to be in a DIP form since all the 287 motherboard sockets at the time were DIP. Sometime later, the 287XLT was introduced placing the chip in a PLCC mounting. I had a 286 in PLCC packaging but I have never seen a motherboard with the 287 socket in PLCC. The packaging for PLCC was about the half the price of the packaging for DIP. Unfortunately, PLCC can only support so many pins around the circumference.

Note that the 386SX and 486 prices plummeted when AMD launched their competing chips and Intel introduced the DX2 line.

One major manufacturer that supported upgrade CPU boards was AST which introduced a 286 model that used 32-bit memory boards to be ready for the 386 entering production. Not related to the comment I was replying to but not worth a second post.
 
I had a 286 in PLCC packaging but I have never seen a motherboard with the 287 socket in PLCC. The packaging for PLCC was about the half the price of the packaging for DIP. Unfortunately, PLCC can only support so many pins around the circumference.
The 386 support both 287 and 387 coprocessors; the later 287XL are just 387-cores in 287-compatible packaging.
 
Back
Top