I was examining a DataExpert ExpertBoard EXP4044 Ver 1.0
It is a 5 Volt only 486 BAT motherboard with 3 VL slots plus 3 plain ISA16 slots.
The chipset is an OPTI 82C895 plus an OPTI 82C602 supplemental chip.
Behind the ZIF socket 3 there is a space for a QFP196.
What I was wondered about is the imprint on the QFP196 space. It says "For best performance. We recommend the Am486 microprocessor".
The question was which Am486 they could mean.
As the board is 5 Volt only, there are not so many Am486 that could be concerned. As far as I know the fastest 5 Volt Am486 is an Am486DX2-80. All 5 Volt Am486DX2-80 have no write back L1.
But an option in the BIOS gives a hint for an write back mode in L1.
This will work fine with a Pentium Overdrive.
But they have implemented write back for Cyrix processors too.
For the older ones without the label "STANDARD PINOUT".
There seems to be some space for a not included VRM (Q1A, Q1, Q1B), but it is unpopulated.
So I considered adding a voltage adaption socket. I asked a friend of mine to examine my thoughts:
Last edited by Disruptor on 2020-07-05, 11:38. Edited 1 time in total.
Not a lie as such, just pure marketing. Probably they got some kind of kickback from AMD, the way system builders did if they stuck an "Intel Inside" sticker on the case.
Also, consider that 3.3V CPUs arrived very late in the desktop arena - only about mid 1994. Before that 3.3V CPUs were the domain of SQFP laptop CPUs. It's entirely possible that 3.3V CPUs didn't exist when this board was released, and the Pentium Overdrive certainly didn't (that was a 1995 part). So whatever claims the board makes are related to 5V non-overdrive, non-WB CPUs, i.e. it's Am486DX2 vs i486DX2. And we already know that that's a dead tie 😉
Here is a picture of the board. Contrary to the manufacturer suggestion, the top performance 486-class processor is installed. The PODP-83 is likely faster in most benchmarks, but it misses the 486-class target. The heatsink has been removed for showcasing the processor type.
The datasheet of the Opti 82c895 has these bullet points about caching modes:
Direct mapped cache
Two banks interleaved or single bank non-interleaved
64, 128, 256 and 512K cache sizes
Programmable wait states for L2 cache reads and writes
2-1-1-1 read burst and zero wait states write @ 33MHz
No Valid bit required
Supports CPUs with L1 write-back support
The jumper description seems to indicate that the board supports three pinout variants: the "classic 486DX pinout", the "PODP pinout" (with write-back control pins on the extra outer rings of pins) and the "Cx486DX" pinout. But wait! Doesn't the Cx486DX include write-back in L1, just like the PODP, just with a different pinout? Does the board support it? Cyrix started to advertise L1 WB on their 486DX series processors with the "Cyrix Writeback" logo on the later Cx486DX2 models, as can be seen on the cpu-world http://www.cpu-world.com/CPUs/80486/Cyrix-Cx486DX2-66.html. Also Cyrix advertised their supposedly enhanced Cache ("FasCache") on Cx486S processors (that's the Intel 80486SX compatible product line) like on this picture http://www.cpu-world.com/CPUs/80486/Cyrix-Cx486S-33-GP.html. No indications of special caching on the Cx486DX processors like in the following picture, though:
If you run CTCM on this system with the Cx486DX, a cache diagnostics tool by the german IT publisher heise, it will tell you "L1 cache mode: write-through". Performance given by that tool, using the classic instructions like "REP MOVSD" or "REP STOSD" also doesn't indicate any benefits of "L1 write-back" compared to the CMOS setting "L1 write-through". So maybe the imprint is right, and for best (non-PODP-) performance, a Am486 processor will do?
It turns out: No! The board does support L1WB on the Cx486DX (and its Cx486DX2 sibling), whereas there is no support for L1WB on any 5V AMD processor. You can get the correct display of "L1 cache mode: write-back" in CTCM if you disable L2, set maximum RAM wait-states and you install a Cx486DX2 processor. To prove that this is not a mis-diagnosis by CTCM, a counter-check with "L1 cache mode: write-through" in the BIOS setup indeed make CTCM indicate L1 WT again. So the board definitely support L1WB on Cyrix processors. The remaining question is: Does the Cx486DX also operate in write-back mode?
The problem with detecting L1WB on the Cx486DX (without clock-doubling) is caused by the FSB being "fast enough" for most write operations, so that there is no penalty if the L1 cache operates in write-through mode. That's a "Challenge Accepted!" moment for me! I went out to write a program that can detect the write-back mode on the Cx486DX processor reliably, even with zero-waitstate L2 cache writes. And indeed, I found out a performance advantage of the write-back mode in the (quite esoteric) case that program tested. You can find that micro-benchmark tool at https://github.com/karcherm/cx486wb/tree/r1.0, including performance numbers.
End result: This board supports the improved cache of all Cyrix 486 processors (with the original Cyrix 486 Write-Back pinout), it supports the write-back mode of the Intel Pentium Over-Drive processor, but does not support L1WB on the "Enhanced Intel 486DX4 pinout", which is also used on 3.3V Cyrix processor (labelled "standard pinout") and most Am486DX4/5x86 processors. The recommended AMD processors offer no performance advantage over Intel processors. The from a technical view, the AMD recommendation seems flat-out wrong.
Thank you mkarcher for your detailled examination.
The thing I'm struggling with is, why does the board recommend Am486 processors but does only support L1 WB cache on Cx486 processors?
Yes, it must be marketing indeed. Or fake news.
You're really interpreting too much into this statement. It really is all about marketing, just as when you buy a dishwasher there's usually a sticker "we recommend X" when independent tests clearly show that other alternatives might be cheaper and/or better. Just as dionb said: Intel did the same and so did many others. That's why independent research and tests are soooo important.
I run this board with the Cx486DX2-80 now, and with 3 VL cards:
CL542X/SOJ/SMT VESA LOCAL BUS REV 1.0 (FCC ID # KDECL5248SMTHFVGA) with a Cirrus Logic CL-GD5249-86QC-B
PT-627B (FCC ID: HBQUMC8631) as secondary IDE port, floppy port and multi-IO
Tekram DC-680T as primary IDE port with cache (with 1.5 firmware; I'm still looking for a complete set of 2.x BIOS)
VL waitstate recommendation on board is set.
I've no problem running 3 VL cards at 40 MHz, when they sit properly.
Some AMD CPUs back then used 40 MHz FSB, ie. very nice for VLB.
Intel had either 33 MHz, or 50 MHz - the latter being pretty much incompatible with VLB.
So, maybe that was the reason to recommend AMD?
Żywotwór planetarny, jego gnijące błoto, jest świtem egzystencji, fazą wstępną, i wyłoni się z krwawych ciastomózgowych miedź miłująca...
P24T was the code name for Socket3 Pentium Overdrive. It finally came out in 1995, but originally it had been expected to appear in 1993. You can find lots of 1993 boards with P24T settings, as well as support for writeback on chips like the Cyrix 486S. I always found it interesting that these boards could not support the writeback cache on AMD and Intel 486 chips. A lot of 1993/94 boards don't support writeback mode with the P24T (POD5V) chips that were actually released, and it is said Intel changed some of the product specifications that broke compatibility with old designs. Others have speculated that the way the chipsets handled writeback cache was different from what the newer CPUs expect. Does anyone have links to supporting documents?
We've given this board another chance and added a Tekram DC-680T cache controller with a connected SD to IDE adapter. It seems that this Tekram can handle 4 GB SD cards with a good CHS remapped transalation.
After adding a sound card (OPTI931 based) we found a crash during Windows 95' start, right after startup sound. It seems the L1 write back mode does not really cooperate with DMA transfers.
Sorry!
