CPU Reminiscences

One thing I remember about the early PC is the 8088 CPUs had a bug that prevented them from operating correctly with an 8087 FPU. IIRC, IBM sold CPU/FPU upgrade kits as "matched pairs" to ensure accuracy.

There is (or was) a shareware program that allows a PC/XT/PC jr or other clone with the NEC V20/V30 to boot CP/M from DOS. It will also emulate a Z-80 at a much slower speed. My PC jr with a V20 would blow away my 12 MHz 286 and even my Xerox 820-II running CP/M programs.

The 80286/80287 pair is unique in the PC world in that as originally designed the FPU ran at 2/3 the speed of the CPU. Thus an 8 or 10 MHz 287 worked fine with a 12 MHz 286. When 287s became available in speeds matching the 286, some companies sold 287 accelerators to allow installation of equal or faster speed FPUs. Intel (or was it AMD?) finally caught on and released the 287XL, which had a built in 1/3 clock multiplier to automatically match the CPU speed. It's supposedly not a good idea to use a 287XL in a 287 accelerator interposer. Most 287XLs I've heard of are rated 6 to 16 MHz. (I always wanted a 287XL for the various 286 clones I've had over the years, but back when they would've been useful they were still insanely expensive.)

Remember the NEAT chipset? That was the hot thing because it was the first chipset for 286 systems that allowed extensive "tweaking" in the CMOS setup. I eventually got one, but that was after I had a 486 so by then it was an "eh" sort of deal.

A PLCC 386SX or 486SLC can usually be directly swapped into a 286 motherboard. IBM actually sold their own version of the 486SLC2 "Blue Lightning" with more level 1 cache than the Intel version as an upgrade for some of their 286 based PS/2 systems. IMHO, these CPUs were created to leverage existing 286 chipset designs to maximize the return on the investment. Sorta like a precursor to the Celeron.

One "hot box" that could be built for less when a 486DX2/66 was out of reach was to take a 386DX40 and swap out the CPU and FPU for a 486DLC2/80 and "487"DLC2/80, which was actually just a clock doubled 387 with a name change. Coupled with 128 KB of 15ns level 2 cache it would beat the pants off any 486DX2/66 and give the 486DX4/75-100 a run for the money. That advantage faded some when the 486 got the VESA Local Bus and accelerated video cards.

Like the 486SLC series, the 487DLC was used in tons of "cheap" boxes that were at heart a 386DX with a BIOS updated to report the presence of the 486 CPU.

What you may want to make clearer on the LEPC 486 page is that if it's just a 486, it runs at 1x clock speed. The various companies made 16, 25, 33, 40, and 50 MHz 1x chips. DX2 is of course 2x clock speed and came in 50, 66, 80, and (really rare in my experience) 100 MHz. (I never could find a board that would run 100% perfect at 50 MHz bus.) Then Intel decided that DXn was a "trademark" instead of a clock speed indicator and applied DX4 to its 3x clocked 486. I've seen them from the various companies in 75, 100, 120, and might have heard of a 150 but I'm not sure.

I won't touch on the Cyrix 5x86, because I never dealt with them, and if you think an Athlon runs hot just try touching one of these POS CPUs. (I actually blistered a finger with but a quick brush against the heatsink on one once.) IMHO this chip was the start of the "crap period" at Cyrix that they didn't begin to come out of until their first Pentium clone with MMX and without that silly "P Rating." I haven't seen much press on their Cyrix/VIA Socket 370 "Samuel" series of chips. VIA sucked up Cyrix and the WinChip division of IDT Centaur to acquire patents and people to create "Samuel."

The AMD 5x86 133 was the absolute height of the 486 pin-and-instruction-set compatible CPU line. ISTR that it appeared soon after AMD's purchase of NexGen, and their pin-incompatable Nx586 CPU and FPU pair that required a special motherboard and chipset. (NexGen tried to work around paying Intel a license fee on the socket design.) Several sources say the AMD 5x86 was originally designed as a 160 MHz chip, and most of the series is quite happy tripping along at 4x40 MHz. On a stable 50 MHz board they'll even make it to 200 MHz most of the time.

There are two major revisions in the PGA package: ADZ and ADW. The ADZ chip has a nominal operating temp of 85 degrees C while the ADW runs at a relatively chilly 55 degrees. While both are rated at some odd amount over 3 volts, they're happy with the Intel DX4 spec 3.3v, and the ADW version will suck up a full 5v for months on end without complaint. This is apparently because the core is "5V tolerant." Apparently AMD made the I/O extra robust. I'd never attempt running the ADZ version on 5v.

The third major variant was a small PQFP type intended for mobile and embedded uses but the lion's share apparently made their way to upgrade makers like Kingston, Evergreen, Gainberry, and PowerLeap who soldered them onto little PCBs with voltage regulators and jumpers for use in everything from the first IBM AT all the way up to the fastest 486.

If a particular 486 board doesn't deliver the sort of clock and multiplier signals the 5x86 expects to kick it into 4x mode, the CPU can be "hot-wired" with an easy-to-make jumper wire wound around two pins at one edge. It is also possible to force it into write-back cache mode, but those pins aren't near an edge and, of course, compatibility/stability testing needs to be done before leaving it that way.

I used to have an ADW version running at a hot-wired 160 MHz on an ancient board that only supported "DX1" CPUs at 5v. It ran cooler than the PQFP on a PCB weirdo Intel SX33 it replaced.