This SiS 85C471 board from Amaquest has a very insidious fake cache: real IC housings with a plausible type name (sounds like Winbond) but the chip doesn't have a bonded die inside:
The chips were soldered without sockets and the cache configuration jumpers are missing.
At the beginning...
...I would like to thank two people very much:
Sergei from Moscow / Russia for the photos of his AV7543 with real cache and the read-out of his BIOS version E23. This version reveals us the manufacturer of this board: Amaquest Computer Corp. in Taiwan.
And many, many thanks to Paul from Brisbane / Australia. He scanned the manual in a first step. Later, he gave me the whole motherboard as a gift for my investigations.
The cache rebuild
I removed all fake cache chips and the tin from the holes of the (empty) cache jumpers / pin headers first.
I used a (relatively) cheap desolder gun from China for this work: ZHONGDI ZD-915. This machine must be carefully maintained but is still quite handy for private use.
I saw on a photo of Sergei: Amaquest used on boards with real cache chips two latches (74F373PC) as driver. So I remounted the passive resistor strings also (to replace them later by 74LS373 chips).
Now I inserted sockets with double spring contact for the cache IC's and added the pin headers for the cache configuration.
Sockets with 32 pins are not longer available, so I used two pieces with 16 pins in series.
For the latches with 20 pins: I only had sockets with precision contacts but this would be not necessary.
I used the passive resistor arrays and 256kB UMC cache for the first test.
Performance
Paul asked me: has the board really fake cache chips? Unfortunately, it has:
The reference is:
- Nine pieces UM61256
- WriteThrough cache strategy
- Drivers U21, U22 are replaced by resistor arrays (passive solution)
Has an active driver (74LS373) an impact on the performance?
The Answer is: no. But I'm convinced it is good for the stability - so I kept the active solution for further tests.
The change of the cache strategy to WriteBack gives a little bit more performance.
The fake cache costs -16% throughput. If I switched off the cache in the BIOS I got exactly the same value. So, in the cache chips can't be anything... :-(
Details can you see here.
The overall performance of the board is quite good (but the Zida 4DVS is a tick faster with the same chipset):
All measurement results are documented here.
Jumper settings
The silkscreen print on the PCB doesn't mention all possible options but Paul scanned the manual for us.
The board hasn't a jumper to set processors with P24D pin out into the 2x/4x clock mode. So, 120MHz (40MHz x 3) is the highest internal clock for an AMD 5x86-P75.
The undocumented jumper P20, pins 3 and 4, sets the internal cache of an AMD 5x86-P75 into the WriteBack mode.
A Cyrix 5x86-100GP processor freezes during the memory test of the bios. But I found this processor type in the hex dump of the bios...?!
Last but not least: the turbo jumper must be set (or the turbo switch closed) to have the full speed.
Different BIOS versions
Version: | Bios image | Bios screen | Remarks |
E20 | AV7543 Rev E20 | 12/23/94-SIS-83C471B/E/G-2C4I9000-00 | From my board, has a year 2000 problem |
E23 | AV7543 Rev E23 | 08/08/95-SIS-83C471B/E/G-2C4I9AD0-00 | From Sergei's board, contains the manufacturer id but is broken (checksum error) |
E24 | AV7543 Rev E24 | 08/14/95-SIS-83C471B/E/G-2C4I9000-00 | From Paul's board, works well until 486DX4-120 |
Facts
Manufacturer / Type: | Amaquest AV7543 Rev1.2 |
Year of manufacture: | August 1995 |
Chipset: | SiS 85C471 (16/1995) |
SiS 85C407 (11/1995) | |
Manual: | VL-BUS True green main board user's manual |
Summery
Strong points:
- Seven full (16bit) ISA slots
- Supports 4x SIMM and 4x DIMM memory (banks are shared, see manual)
- Quiet good performance with real cache modules
Weak points:
- If I play a wav-file with the original fake cache I got so strange micro-stutters in the sound output (but the mean cpu-load is only 25%). This problem went away with real cache chips.
- The BIOS chip blocks two ISA slots:
falk.richter*at*yandex.com
Last update: May 21, 2021