• Please review our updated Terms and Rules here

Is there a case to be made for the 386?

hunterjwizzard

Veteran Member
Joined
Mar 21, 2020
Messages
791
My earliest family computer I can remember was a 386 running windows 3.1, so I've long had some fascination with the CPU. It must have been versatile, since they were still being manufactured for niche roles as late as 2009. That is some very long legs for a CPU design.

So, is there a case to be made for the 386, or is the 486 just objectively better?
 
Which 486 versus which 386? The fast 386s were fully competitive with the non-clock doubling 486SX models. Finding a good motherboard is more important than the difference between the 386-40 and the 486-33.

I prefer the clock doubled 486 since there are many options to tweak the speed by turning off cache or the clock doubling or changing the bus speed.
 
It was sort of a first for Intel, (not counting the ill-starred iAPX 432 from 1981) being a 32-bit microprocessor--and it signaled that Intel had finally got the virtual memory thing halfway right. I recall that the i960 came out around the same time as the 80386, but timewise, I think the 386 had a slight edge on introduction date, but the same month.

So it's historical. It's a matter of conjecture as to whether we'd have been better off with the 960 instead of the 386.

Anent that, I have a 960 or 860 book that has BillG saying that Microsoft was fully committed to the new processor. Was any version of Windows ever developed for either the 860 or 960?
 
Last edited:
At Siemens (where I used to work years ago) we still were using 386sx's quite heavily until I left (2007). While the 486 may have an edge in some aspects, 386's were workhorses for a long time. I've also had some late 386's blow away early 486's on performance as well.
 
At Siemens (where I used to work years ago) we still were using 386sx's quite heavily until I left (2007). While the 486 may have an edge in some aspects, 386's were workhorses for a long time. I've also had some late 386's blow away early 486's on performance as well.
They were still being manufactured in 2009. Apparently some handhelds used them, PDAs and early predecessors to smartphones.
 
The fast 386s were fully competitive with the non-clock doubling 486SX models

This was true insofar that on integer code the fastest 386s (IE, the 33mhz Intel and 40mhz AMD specs) were roughly tied with the slowest 16/20mhz 486SX models, but just about any benchmark shows the 486 running about twice as fast per clock (IE, a 386 would have to be pushed up to 66mhz, which ain’t happening, to beat the dirt common 486/33mhz), and if you throw floating point into the mix the 486 further ground-and-pounds the 386+387 combo because of its faster coprocessor interface.

Example 386 vs. 486 benchmark results here:


There might be edge cases where the 486 comes off worse, like if you were comparing a really high-end 386 with cache to a 486 stuck in a badly engineered motherboard that doesn’t support burst mode (a thing that did happen sometimes) and no secondary cache, but generally speaking the 486 smokes the 386 on IPC. The 386 is barely any more cycle efficient than the 286, which is why those various ”486 in a 386 socket” chips from Cyrix and IBM are able to squeeze significant performance improvements out of the platform *without* changing chipsets.
 
it was only internal. ever wonder why WNT started out with vers 3.x ?

Wasn‘t the story behind it debuting as 3.x because it was originally intended to be OS/2 v3 before IBM and Microsoft has their little falling out? (IE, the plan was 1.x was 16 bit Intel-only, 2.x 32-bit Intel, single user, and 3.x multi-platform, multi-user?)
 
? 3.1 was because it was an upgraded version of 3.0. I recall many moons ago working for a school that had a copy of 3.0 on the shelf. Really wish I'd looted all the vintage software from that place before it got thrown out.
 
I was under the delusion that NT 3 was so-called to give the impression of compatibility with Windows 3.1 which ruled the MS roost.

I wonder what i860 hardware was used for development. Perhaps just a coprocessor card in a PC AT?
 
I suspect the "better" consisted of "we have a bazillion-dollar decades-long contract to manufacture a specific system with a 386 in it. If we were to swap in something else, it would trigger a design recertification that cost years and millions."

From a technical perspective, the lower integration of a 386 might have benefits in some very narrow use cases:

* It was the last main-line x86 CPU without a L1 cache; I'd expect a system without caches would provide more deterministic performance for hard-real-time needs. You can theoretically tie down some cache-enable pins for the same effect though.
* You didn't have to pair it with an x87-style FPU-- you could use a Weitek 3167 with appropriate boards. There was also the 4167, but I suspect that it was difficult to convince mainboard designers to support an aftermarket FPU once the 486DX existed.
 
The 386 was a truly revolutionary CPU. Not only did it feature full 32-bit addressing, an even more impactful feature was its built-in MMU. The MMU is what allowed you to quickly switch between apps on Windows 3.11. So the real question is not whether the 386 is relevant, the question is if did the 486 do enough to unseat the 386? Sure the 486 is objectively better, but it was not nearly as ground-breaking as the original 386.
 
The 486 did have an instruction that made switching between apps faster.

Where does an improved 386 like the IBM 386SLC fit into this comparison?
 
Considering that most installations still ran 16-bit applications years after the introduction of the 386, the 32-bit aspect is questionable.

In the day the only thing a 386 could do better under dos or Windows was more easily multitask real mode dos programs.

The 32 bit aspect was only used as memory address space to speed extended ram access and improve multitasking.
 
Back
Top