a 486SX is a no-floating-point 486DX.
A 386SX is on a 16-bit data bus, a 386DX has 32-bit. They're sort of designed to drop into 286 mobo designs with minimal modifications. I have a 386SX PC which was sold as either a 12MHz 286 or 16MHz 386 by using a different modular CPU card.
I built.. Iunno, thirtyish 386 systems give or take? Maybe nearer forty.
The 386SX was always sans-FPU. Period. The whole POINT of the SX was "cost reduction" and part of that was no FPU, no cache. The 386DX was introduced
after the SX to differentiated between which had a cache and standard FPU capability. (The SX uses a non-standard FPU in Intel land. Get to that in a minute.)
Oh, and that only applies to the Intel 80386 vs 386SX vs 386DX - it does not apply to the IBM 386 (under license), the AMD Am386SX, C&T Super386, or the Intel 386SXSA or i386EX (oh, or the 386CXSB which perversely was sometimes called a CX or SX.) It's like 68k-roulette or Guess-The-MIPS but the performance
always sucks.
The ThinkPad L40 is equipped with, to quote
the original sales brochure, "386SX(TM) processing power" but could be expanded to 18MB of DRAM.
You can't do that on an Intel 386SX. So yeah. It
is not an Intel 386SX. It's an IBM 386SLC CMOS 3.3V (vs Intel on CHMOS IV at 5V) which is marketed as a 386SX because it is instruction compatible. But it is NOT pin compatible or voltage compatible with the 80386.
Though I did remember the clocks wrong - 16/20/25 for IBM "386SX(TM)", vs 16/20/25/33 Intel(R) 386SX(TM).
To wit: IBM never once shipped an Intel 386 CPU. Not ever. There is no example of an IBM system prior to late 486DX2 family which shipped with a genuine Intel part. IBM had a license and made their own 100% instruction compatible CPUs on CMOS process - a more advanced process than Intel had by far - exclusively for use in IBM systems. You could not buy an IBM 386 or 486 except installed in a complete system, nor could you purchase an Intel CPU upgrade for an IBM system. (386DX, 132 pins. IBM 386SLC had 160 IIRC.)
Serious folks back then
wanted IBM 386SLCs because they had vastly lower heat dissipation (2.5W @ 25MHz) and gobs of cache (386SLC packed up to 16KB!) which made them faster than equivalent and even greater clock 386DXes. But you couldn't get them. That advantage was somewhat reduced in the 486 era - the 486SLC was still using the 386SLC IP core. (Not to be confused with the IBM Blue Lightning family pre-Cyrix, which could be obtained commercially if you knew the right people, and was fantastic.)
Confused enough yet?
So. As I said: it's a 386 sans FPU. It's
not an 80386.
