• Please review our updated Terms and Rules here

Question on 386DX

moetzmoet

Banned
Joined
Jun 6, 2022
Messages
5
Location
United States

Would a 386DX running on a motherboard with no cache on it act like a 386SX-33?

If so, what speed DX would do this? Curious as I want to setup a 386SX Machine and 33 is the speed I'd like to run, but I have a 386SX-25 board with no cache and also a 386DX-25 board with no cache and trying to get as close as I can....
 
The 386sx is a total performance dog. Its got half the bus width of the dx and from what I remember was painfully slow on Windows/3.
So even with no cache a 386dx is much faster than 386sx so according to this article


33% faster, so it says a 25mhz 386dx no cache is about as fast as a 386sx-33 but from memory i would say this is under-rating the 386dx.
 
FWIW, if you look at old benchmarks you can find instances where, say, a 25mhz machine with fast memory scores almost as high as a 33mhz machine with more wait states, so ultimately I think either one of these machines are going to be roughly in the same benchmark class as random "33mhz 386sx". (IE, I'm going to guess that if you *had* a cacheless 33mhz 386SX to compare to these two the spread between them all would probably only be around the 33% ballpark. IE, they're all going to be able to run about the same range of software "acceptably".)

Or I guess to put it another way, I'm not quite sure why a 33mhz 386SX is the specific target you'd want to hit. SXes this fast came out pretty close to the end of the chip's life; they're 1993's poster child for a bottom-feeder office PC.
 
A 386SX will run Windows 3.1 just fine. It's basically a 32-bit 286.
 
I have a 386 SLC50 thats pretty close in speed to my DX33... So around 25-33% seems right.
 

Would a 386DX running on a motherboard with no cache on it act like a 386SX-33?

If so, what speed DX would do this? Curious as I want to setup a 386SX Machine and 33 is the speed I'd like to run, but I have a 386SX-25 board with no cache and also a 386DX-25 board with no cache and trying to get as close as I can....
I'm curious, why are you targeting this specific processor / speed combination?
 
As noted, they'd be in the same broad performance class - but do note that even a cacheless 386DX has a 32-bit bus, while the 386SX is 16-bit externally. Assuming they're using memory/memory controllers with the same performance characteristics for the sake of argument, the DX would still have the edge.
 
If all things were equal, then the DX would beat the SX (by around 35%)?
But, in real life things would not be equal as the design of the SX motherboard would certainly be much newer with tighter memory timings.
I would guess it's possible for a mature SX to be an old DX.
It would definitely be an interesting test. Somebody should test it.
 
Would also depend on the application - if it's just running real-mode 8086/286 DOS software that doesn't know or care about the 32-bit capabilities, the gap would be much narrower, probably close to nonexistent.
 
And there were 386sx systems with CHIPSET cache and the IBM 386slc that seemed to outrun a DX of the same speed
 
And there were 386sx systems with CHIPSET cache and the IBM 386slc that seemed to outrun a DX of the same speed
The IBM 386SLC was derived from the Intel 386SX codemask - despite being limited to the 24-bit address bus and a 16-bit data bus, it supported all Intel 486SX instructions, and had an 8Kb L1 cache. Later derivatives of the 486SLC2 and 486SLC3 had 16Kb of L1 cache, and were clock-multiplied internally. The IBM 486DLC2 was a slightly modified PQFP 386DX pinout, and definitely the fastest x86 CPU of that 386DX form-factor.
 
At the same clock rate, a 386sx performs just like a 286 - in rare cases even slightly worse. The whole point of the 386sx was that old 286 chipsets and hence a more simple and way cheaper design could be used.
 
At the same clock rate, a 386sx performs just like a 286 - in rare cases even slightly worse. The whole point of the 386sx was that old 286 chipsets and hence a more simple and way cheaper design could be used.

But on the flip side, the 386 is pretty much at the same “tech level” as the 286 when it comes to instructions per clock, IE, it’s not a lot faster at the microcode level. So in many cases the only significant advantage the 386DX has over either the 286 or 386SX is the benefits of the wider bus when it comes to instruction/data fetches. Because Intel allows non-word-aligned code on these CPUs it’s possible to do stupid pet tricks like compiling the same code so it’s properly 16 bit word aligned in one case and intentionally misaligned in the other, and in such a circumstance the 386 running the misaligned code may be no faster than the 286, possibly even slower.

This issue does crop up in real life with DOS software because alignment didn’t matter on the 8088, and thus software compiled for maximum code density can kneecap the 386DX (And the 286 too, sometimes) in ways that drags their performance down significantly compared to theoretical.
 
Last edited:
I have a 386DX-40 which runs great. One of the things about the DX over the SX is memory management. I believe that the SX tops out around 16MB and the DX around 4GB, also has the advantage of the L2 cache.
 
also has the advantage of the L2 cache

Neither the SX or the DX has cache unless you add it. Granted cacheless SXes were probably more common.

But yes, the 16MB limit of the SX is a bummer. Didn’t matter much in 1988 but near the end of its life… it really sucked for otherwise kind of awesome spin-offs like the 486SLC2 that had internal cache and clock doubling but retained the 24 bit address bus. They were still selling motherboards with those as late as 1995-ish, when a 16MB limit *should* have looked like a deal-breaker for a new computer.
 
My reference to the L2 cache was a misnomer as in the the case of the AM386-40, the cache is commonly referred to as a "L1" motherboard cache. Sorry for the confusion.
 
Neither the SX or the DX has cache unless you add it. Granted cacheless SXes were probably more common.

But yes, the 16MB limit of the SX is a bummer. Didn’t matter much in 1988 but near the end of its life… it really sucked for otherwise kind of awesome spin-offs like the 486SLC2 that had internal cache and clock doubling but retained the 24 bit address bus. They were still selling motherboards with those as late as 1995-ish, when a 16MB limit *should* have looked like a deal-breaker for a new computer.

I have a motherboard with a 486SLC2 @ 66 MHz and it's not at all bothered by a 16 MB memory limit, because it's too slow to use any software that would require anything near that much memory. Despite the high clock speed, it really only performs somewhere between a real 486SX 25-33 MHz. Windows 3.x is fine, but Windows 95 chugs and doing anything more than running single simple applications is painful.

What made it worse is the 16 bit data bus, so even on this board that has 32 bit VLB slots, it has to break up 32 bit transactions into two 16 bit transactions, significantly slowing down disk and video I/O when using VLB cards.
 
I have a motherboard with a 486SLC2 @ 66 MHz and it's not at all bothered by a 16 MB memory limit, because it's too slow to use any software that would require anything near that much memory. Despite the high clock speed, it really only performs somewhere between a real 486SX 25-33 MHz. Windows 3.x is fine, but Windows 95 chugs and doing anything more than running single simple applications is painful.

What made it worse is the 16 bit data bus, so even on this board that has 32 bit VLB slots, it has to break up 32 bit transactions into two 16 bit transactions, significantly slowing down disk and video I/O when using VLB cards.
The SLC does run 16 bit code at parity with a normal 486 clock for clock. So yeah it’s specialized heavily to of the day 16 bi t applications but…


Board model and specs including L2 or it didn’t happen.

My experience is that a 386sx40 is far slower than even a 486sx16 and a 486slc with L2 cache is far faster than a 386dx clock for clock.

Unless you lack L2 and have very poor component selections you should be doing better than you state.
Admittedly a 486sx-40 class system isn’t very fast but…
 
Back
Top