After a long wait, I decided to bite the bullet and order a ZIF (Zero Insertion Force) socket adapter suitable for 386 CPUs through Digi-Key. The manufacturer is Aries Electronics and the part number is 196-PRS14001-12, as established some time ago. The main motivation was plug-testing of a pile of 386 CPUs, which is not much fun with standard 386 LIF (Low Insertion Force) sockets.
To be precise, this is an adapter which plugs into a conventional LIF socket and provides a ZIF socket with a classic lever for a PGA chip.
The one big downside of the product is price. The adapter cost me 65 Euro plus VAT, and the minimum order quantity was two units. That adds up quickly. In addition, the item isn’t stocked and it took well over a month to arrive.
The primary upside on the other hand is that the adapter really works, and it saves a lot of time, frustration, and bent pins. Testing twenty 386 CPUs is a matter of minutes and requires no tools, no force, and if anything, straightens the pins of the processors.
The adapter is well made and the lever can be operated very easily. 386 CPUs with straight pins can be inserted and removed effortlessly. Compare with the classic approach:
Such a CPU extraction tool does work, but one has to have one first (the pictured tool came from an Intel 486 OverDrive kit). Even then, it can be problematic if there’s not enough space to use the tool, and care is needed not to bend the pins.
The ZIF adapter isn’t perfect either. The list of disadvantages is substantial, but most of those are relatively minor inconveniences:
- All 196 pins are populated and the adapter won’t fit in a standard 386 socket; the center pins must be bent to allow the adapter to be installed.
- The adapter is sized for 15×15 pins (rather than 14×14 of a 386), hence there will be unused holes; extra care must be applied when aligning the CPU.
- The adapter is relatively bulky and may not fit in the desired orientation; in the worst case, it may not fit at all.
This is what the adapter looks like after it’s been mangled to fit in a 386 socket:
Without that, it’s just not possible to install the adapter.
Pin 1 is not marked on the adapter and the user must pay attention to the orientation of the CPU. This is slightly inconvenient but necessary, since the adapter may only fit in certain orientations depending on the board layout.
The following image shows the adapter installed in a 386 board, including a CPU. The two redundant rows of holes are apparent.
The adapter has been successfully tested with Intel, AMD, Cyrix, and Texas Instruments 386-socket processors. It worked flawlessly with all of them, with bus speeds up to 50 MHz (with an overclocked Am386DX-40).
For Sale: One Aries Electronics 14×14 ZIF PGA socket adapter model 196-PRS14001-12 as described above, unused (no bent pins). The price is 75 Euro plus shipping. Please leave a comment if interested. Buyer within the EU preferred.
I kind of miss having some actual 32bit ISA hardware…. Oh do you have a 386DX-12? I’ve always wondered if the first batch of ultra slow 386’s were ever ΣΣ, or if they were all 16bit-sw only?
No, I don’t have the early 12 MHz 386, I doubt very many were ever made 🙂 And from what I understand, they were so buggy that you were probably lucky to run 16-bit software on them. 32-bit not so much, and forget paging. Oldest I have is a 16 MHz 386 from 1988 or so, not nearly as buggy.
At least it was stable enough for Intel to demo back in late 1985 with 16-bit software.
The early 386s did not have the 32-bit multiplication bug that the lack of the double sigma indicated. IIRC, that math bug occurred with a portion of the B1 stepping but a 12 MHz 386 was probably A1 or B0 stepping. Intel talked about selling 100,000 386 chips when the furor over the math bug was at its hottest so not many failed B1 or earlier steppings existed.
32 bit code should run on A0 but it will be very different than the code that any mainstream 386 compiler would generate. Problems would include difficulties mixing 32-bit code with 16-bit code and an inability to do a long jump at the same protection level. I would have to look at a list of the many other bugs that I don’t remember to see workarounds were possible. I suspect that A0 stepping compatible 32-bit code would be considerably slower than standard 386 code when run on a later stepping 386.
It is next to impossible to find the errata lists for the early 386 revisions; back then Intel didn’t produce buggy chips (ha ha) so it was all super secret. I’ve seen the B0 errata sheet and it is quite long and scary. Especially FPU interfacing and paging were quite buggy. The bugs generally had workarounds, but that was only useful once the bugs became known and understood. Without workarounds, the CPU might hang and misbehave in all kinds of nasty ways. I think very few of the problems affected real mode operation.
The C0 stepping seems to have been the first one without major bugs. I’m not sure when it was introduced, probably late ’86 or early ’87. As you say, the production volume for the early, broken 386s actually wasn’t all that high. Still, several old operating systems went to great lengths to accommodate the B1 stepping errata.
FWIW, the B1 stepping apparently only existed in 16 MHz and 20 MHz variants. It also already lacked the XBTS/IBTS instructions.
Speaking of errata lists, I recently found a pretty good one:
Now, if only I had a buggy 386 to play with…
Excellent, I see that Jeff wrote everything down 🙂 I only have one B1-step CPU, and it’s the double-sigma variant. I have another double-sigma 386 with no S-spec which might be a B1-step as well.
I don’t know of any convenient way (in 2015) to determine exactly which step it is, but Google can find a DOS program called 386ID that can get the CPU signature on most computers. That should at least tell you if it’s not a B1-step.
The problem is that B0 and B1 return the same signature. The PCJS link shows that as does one of the Intel manuals I looked at. It took Intel some time to figure out how to supply customers with the needed information without relying on engineers answering questions on the phone (which comparatively tiny Intel with only a few customers could do in the 70s).
I was looking for the same thing and opened a topic in a forum about CPUs: http://www.cpu-world.com/forum/viewtopic.php?p=238480
There are several options for approx $25 or 45 euros. Still very expensive, but way less than 2 x 65 euros. I still did not get/test any, but hope this can be useful.
Curious, does it work with the C&T Super386?
“12 MHz 386 was probably A1 or B0 stepping. ”
A lot of poking around: The 386 came out at 12.5, 16.
And an ICE indicates that the first speed was 12.5 and not 12:
“The i386 was originally released in 1985 at 12.5 MHz”
Anything about the 386 not written in the 1980s is hearsay unless there’s an actual chip to prove it. But the old datasheet (231630-002) says 12.5 and 16 MHz so that should be reasonably credible. There probably only was a handful of the under-16 MHz chips made anyway.