andrea wrote on 2025-09-09, 19:28:

Looking at the traces where the cache chips would be, I'd guess that if a real cache + tag was soldered it would work (BIOS nons […]
Show full quote

Babasha wrote on 2025-09-05, 17:09:

While the BIOS has a setting for L2 cache, there is no cache on the board. The setting just shows a message that there's 512KB of cache, even though there isn't. The spots for L2 cache are probably just "dummies" to allow for soldering on fake chips.

Looking at the traces where the cache chips would be, I'd guess that if a real cache + tag was soldered it would work (BIOS nonsense aside).
The routing seems too much effort just for nothing, especially given how cramped the PCB is.

The only board with fake QFP cache and fake traces that i know of is the M919 and the routing they did there (just for show) is much more simpler.

There dead SIS5591 motherboard in my hands with visual compatible cache and tag chips...

Does the mainboard use a x64 Pipelined Burst SRAM?
If there are 128 total pins, then it is a x64 part.
Considering that there is only 1 footprint for the cache, it is highly likely.
For the Pipelined Burst SRAM, you must use at least -5 or -100 part, depending on the manufacturer.
Back in March to April 2025, I managed to purchase 10 pieces of ISSI 64k x64 Pipelined Burst SRAM from a seller on AliExpress.
I believe I paid something like 10 devices for $40 including shipping and sales tax.
It was a -100 part that is compatible with a 100 MHz Socket 7 bus.
For the tag RAM, -8 or better is a must.
-10 or -12 is not adequate for a 100 MHz Socket 7 bus.
I also was able to purchase -8 parts on AliExpress at the same time.
The one I got is manufactured by Winbond.
For these parts, procuring on eBay might be difficult, if not impossible.
Some AliExpress sellers appear to not sell to a person with a US address due to the current trade related tension that exists between the US and China.
The current US administration (now acting like a regime then an administration) Illegally suspending de minimis rule has really hurt me in procuring chipset / clock synthesizer / PMIC / SRAM / connector on AliExpress.
The last comment is a rant obviously, but it has affected me negatively, so I cannot help but to mention it.

andrea wrote on 2025-09-09, 19:28:

Looking at the traces where the cache chips would be, I'd guess that if a real cache + tag was soldered it would work (BIOS nons […]
Show full quote

Babasha wrote on 2025-09-05, 17:09:

While the BIOS has a setting for L2 cache, there is no cache on the board. The setting just shows a message that there's 512KB of cache, even though there isn't. The spots for L2 cache are probably just "dummies" to allow for soldering on fake chips.

Looking at the traces where the cache chips would be, I'd guess that if a real cache + tag was soldered it would work (BIOS nonsense aside).
The routing seems too much effort just for nothing, especially given how cramped the PCB is.

The only board with fake QFP cache and fake traces that i know of is the M919 and the routing they did there (just for show) is much more simpler.

Who manufactured your SiS5591 chipset based mainboard?
SiS5591 had some major design wins at the onset, but quickly fizzled due to the 100 MHz Socket 7 support issue.
Just glancing at the pictures of the SiS5591 chipset based mainboards on The Retro Web, most of them appear to have -6 speed for the Pipelined Burst SRAM (PBSRAM) part(s) and GIGABYTE GA-586SG even uses a -12 Winbond asynchronous SRAM part for a tag RAM.
This mainboard may not even be able to handle a 83 MHz bus clock (only up to 75 MHz).
It appears that only GIGABYTE GA-5SG100 and PC Chips M590 correctly use -5 PBSRAM and -8 tag RAM, and both appear to officially support the 100 MHz Socket 7 bus.
Perhaps, the 100 MHz Socket 7 bus support issue was merely the mainboard manufacturers not using -5 PBSRAM and -8 tag RAM, and there might not have been anything wrong with SiS5591 chipset itself.
Perhaps, people who still own SiS5591 based mainboard can repopulate -5 PBSRAM and -8 tag RAM to see if 100 MHz bus clock issue can be fixed.
Read my reply to andrea where I discuss how I obtained -5 (or for ISSI parts, -100) PBSRAM and -8 tag RAM needed for populating them on the Eagles Power TX mainboard.

Babasha wrote on 2025-09-09, 19:56:

andrea wrote on 2025-09-09, 19:28:

Looking at the traces where the cache chips would be, I'd guess that if a real cache + tag was soldered it would work (BIOS nons […]
Show full quote

Babasha wrote on 2025-09-05, 17:09:

While the BIOS has a setting for L2 cache, there is no cache on the board. The setting just shows a message that there's 512KB of cache, even though there isn't. The spots for L2 cache are probably just "dummies" to allow for soldering on fake chips.

Looking at the traces where the cache chips would be, I'd guess that if a real cache + tag was soldered it would work (BIOS nonsense aside).
The routing seems too much effort just for nothing, especially given how cramped the PCB is.

The only board with fake QFP cache and fake traces that i know of is the M919 and the routing they did there (just for show) is much more simpler.

There dead SIS5591 motherboard in my hands with visual compatible cache and tag chips...

bracecomputerlab wrote on 2025-09-09, 12:24:

If you try a 64M bit x8 based 128 MB module (total of 16 devices), will it recognize as a 128 MB module?

No - the maximum that the chipset can recognise is 65MB.

Speaking of memory, XMS support is wonky on that board - I do recall with my FreeDOS+HIMEMX setup that HIMEMX needed a specific parameter to be enabled, otherwise it would either hang or not detect XMS at all. I didn't note what parameter it required, I'll keep it in mind the next time I use the mobo.

bracecomputerlab wrote on 2025-09-09, 12:34:

Perhaps, its power circuit not being able to handle the large current draw of faster AMD K6-2s like ones above 500 MHz.
Interesting to see if AMD K6-III / III+ can run on this board.

It natively supports K6-III+ chips - my 400ATZ works fine. Just like Babasha's mobo, mine is incapable of POSTing above 5x multiplier but I've had no stability issues with 100Mhz FSB.

bracecomputerlab wrote on 2025-09-10, 10:10:

Who manufactured your SiS5591 chipset based mainboard? SiS5591 had some major design wins at the onset, but quickly fizzled due […]
Show full quote

Who manufactured your SiS5591 chipset based mainboard?
SiS5591 had some major design wins at the onset, but quickly fizzled due to the 100 MHz Socket 7 support issue.
Just glancing at the pictures of the SiS5591 chipset based mainboards on The Retro Web, most of them appear to have -6 speed for the Pipelined Burst SRAM (PBSRAM) part(s) and GIGABYTE GA-586SG even uses a -12 Winbond asynchronous SRAM part for a tag RAM.
This mainboard may not even be able to handle a 83 MHz bus clock (only up to 75 MHz).
It appears that only GIGABYTE GA-5SG100 and PC Chips M590 correctly use -5 PBSRAM and -8 tag RAM, and both appear to officially support the 100 MHz Socket 7 bus.
Perhaps, the 100 MHz Socket 7 bus support issue was merely the mainboard manufacturers not using -5 PBSRAM and -8 tag RAM, and there might not have been anything wrong with SiS5591 chipset itself.
Perhaps, people who still own SiS5591 based mainboard can repopulate -5 PBSRAM and -8 tag RAM to see if 100 MHz bus clock issue can be fixed.
Read my reply to andrea where I discuss how I obtained -5 (or for ISSI parts, -100) PBSRAM and -8 tag RAM needed for populating them on the Eagles Power TX mainboard.

Babasha wrote on 2025-09-09, 19:56:

andrea wrote on 2025-09-09, 19:28:

Looking at the traces where the cache chips would be, I'd guess that if a real cache + tag was soldered it would work (BIOS nonsense aside).
The routing seems too much effort just for nothing, especially given how cramped the PCB is.

The only board with fake QFP cache and fake traces that i know of is the M919 and the routing they did there (just for show) is much more simpler.

There dead SIS5591 motherboard in my hands with visual compatible cache and tag chips...

Its Asus SP98AGP-X as possible donnor
Dont have any other mobos or chips for L2 cache soldering(((

Using a couple of free minutes, I prepared a slightly modified version of the intelligent POWERIDE IDE driver! (which has helped me out before with strange and slow IDE controllers).

The driver is in the attachment below. The instructions are extremely simple to follow (plus, read the readme and manual).

What kind of performance does this thing muster?

Is it similar to early UMC chipsets?

I will not recommend using a -6 part for a 100 MHz Socket 7 bus application.
In the same token, -10 tag RAM is also too slow.
All 100 MHz Socket 7 mainboards where 100 MHz clock speed actually works are known to use a -5 PBSRAM part and a -8 tag RAM without exceptions.

Babasha wrote on 2025-09-10, 13:10:

bracecomputerlab wrote on 2025-09-10, 10:10:

Who manufactured your SiS5591 chipset based mainboard? SiS5591 had some major design wins at the onset, but quickly fizzled due […]
Show full quote

Who manufactured your SiS5591 chipset based mainboard?
SiS5591 had some major design wins at the onset, but quickly fizzled due to the 100 MHz Socket 7 support issue.
Just glancing at the pictures of the SiS5591 chipset based mainboards on The Retro Web, most of them appear to have -6 speed for the Pipelined Burst SRAM (PBSRAM) part(s) and GIGABYTE GA-586SG even uses a -12 Winbond asynchronous SRAM part for a tag RAM.
This mainboard may not even be able to handle a 83 MHz bus clock (only up to 75 MHz).
It appears that only GIGABYTE GA-5SG100 and PC Chips M590 correctly use -5 PBSRAM and -8 tag RAM, and both appear to officially support the 100 MHz Socket 7 bus.
Perhaps, the 100 MHz Socket 7 bus support issue was merely the mainboard manufacturers not using -5 PBSRAM and -8 tag RAM, and there might not have been anything wrong with SiS5591 chipset itself.
Perhaps, people who still own SiS5591 based mainboard can repopulate -5 PBSRAM and -8 tag RAM to see if 100 MHz bus clock issue can be fixed.
Read my reply to andrea where I discuss how I obtained -5 (or for ISSI parts, -100) PBSRAM and -8 tag RAM needed for populating them on the Eagles Power TX mainboard.

Babasha wrote on 2025-09-09, 19:56:

There dead SIS5591 motherboard in my hands with visual compatible cache and tag chips...

Its Asus SP98AGP-X as possible donnor
Dont have any other mobos or chips for L2 cache soldering(((

bracecomputerlab wrote on 2025-09-11, 06:52:

I will not recommend using a -6 part for a 100 MHz Socket 7 bus application. In the same token, -10 tag RAM is also too slow. Al […]
Show full quote

I will not recommend using a -6 part for a 100 MHz Socket 7 bus application.
In the same token, -10 tag RAM is also too slow.
All 100 MHz Socket 7 mainboards where 100 MHz clock speed actually works are known to use a -5 PBSRAM part and a -8 tag RAM without exceptions.

Babasha wrote on 2025-09-10, 13:10:

bracecomputerlab wrote on 2025-09-10, 10:10:

Who manufactured your SiS5591 chipset based mainboard? SiS5591 had some major design wins at the onset, but quickly fizzled due […]
Show full quote

Who manufactured your SiS5591 chipset based mainboard?
SiS5591 had some major design wins at the onset, but quickly fizzled due to the 100 MHz Socket 7 support issue.
Just glancing at the pictures of the SiS5591 chipset based mainboards on The Retro Web, most of them appear to have -6 speed for the Pipelined Burst SRAM (PBSRAM) part(s) and GIGABYTE GA-586SG even uses a -12 Winbond asynchronous SRAM part for a tag RAM.
This mainboard may not even be able to handle a 83 MHz bus clock (only up to 75 MHz).
It appears that only GIGABYTE GA-5SG100 and PC Chips M590 correctly use -5 PBSRAM and -8 tag RAM, and both appear to officially support the 100 MHz Socket 7 bus.
Perhaps, the 100 MHz Socket 7 bus support issue was merely the mainboard manufacturers not using -5 PBSRAM and -8 tag RAM, and there might not have been anything wrong with SiS5591 chipset itself.
Perhaps, people who still own SiS5591 based mainboard can repopulate -5 PBSRAM and -8 tag RAM to see if 100 MHz bus clock issue can be fixed.
Read my reply to andrea where I discuss how I obtained -5 (or for ISSI parts, -100) PBSRAM and -8 tag RAM needed for populating them on the Eagles Power TX mainboard.

Its Asus SP98AGP-X as possible donnor
Dont have any other mobos or chips for L2 cache soldering(((

I think its possible to solder my cache and tag to check is it work at 66-75MHz or is it places for just fake chips?

Babasha wrote on 2025-09-11, 08:34:

I think its possible to solder my cache and tag to check is it work at 66-75MHz or is it places for just fake chips?

Assuming it works I'd say a cached 83MHz bus system is still a much better experience than 100MHz without.

Also while on the throne I was thinking, it is most certainly odd for NEC to make their own chipset for a PC-compatible. A repurposed/adapted PC98 chip also seems strange as they use completely different io-ranges and everything and also aren't single chip.
Could it be a single-chip design from another vendor (so SiS pretty much) made in their own fabs?

Well, the -6 part should work up to 83 MHz, but you actually got Winbond W83194R-17A populated, not
W83194R-17.
This is actually a small problem because -17A appears to be meant for Slot 1 / Socket 370, but it is used on this mainboard.
I say this because -17A does not support 30 MHz PCI bus clock for 75 MHz FSB clock, but -17 does.
With 75 MHz, -6 should be fine with some extra margin to spare.
-10 tag RAM might be okay for 83 MHz, but ALi Aladdin IV or V demanded -9 tag RAM for 83 MHz in their databook.
It is an unrelated chip, but just wanted to mention it as a reference or FYI.

I forgot the name of the benchmark name, but necroware videos on YouTube often tests memory bandwidth.
If you use this, by the memory bandwidth numbers, you should be able to figure out if the external cache is really working or not.

Babasha wrote on 2025-09-11, 08:34:

bracecomputerlab wrote on 2025-09-11, 06:52:

I will not recommend using a -6 part for a 100 MHz Socket 7 bus application. In the same token, -10 tag RAM is also too slow. Al […]
Show full quote

I will not recommend using a -6 part for a 100 MHz Socket 7 bus application.
In the same token, -10 tag RAM is also too slow.
All 100 MHz Socket 7 mainboards where 100 MHz clock speed actually works are known to use a -5 PBSRAM part and a -8 tag RAM without exceptions.

Babasha wrote on 2025-09-10, 13:10:

Its Asus SP98AGP-X as possible donnor
Dont have any other mobos or chips for L2 cache soldering(((

I think its possible to solder my cache and tag to check is it work at 66-75MHz or is it places for just fake chips?

My guess as to who designed this mysterious Socket 7 chipset, I think it is RDC Semiconductor.

https://www.rdc.com.tw/index.php

It appears that they designed a 486SX compatible ASIC core 20+ years ago, but in the public documentation published by them that I happened to read, they will insist that it is a RISC processor and not an x86 processor.
It comes with one 8254, two 8259, and two 8237 wired in a way to work like a PCI based IBM-PC/AT compatible, so I guess if it walks like a duck and quacks like a duck, I guess it is an x86 SoC.

http://www.sima.com.tw/download/R8610_D06_20051003.pdf

DM&P Vortex86 DX and EX are likely the renamed version of their EmKore DE and DS.

https://www.rdc.com.tw/index.php?route=home/emk

That being said, early Vortex86 was a rebadged SiS550 with Rise Technology mP6 Pentium MMX class processor integrated.
I think RDC kept a low profile due to Intel's history of suing anyone trying to develop an x86 compatible processor.
Of course, not everyone got sued by Intel, but you get the point.

That being said, most x86 related patents except for SSE4/AVX/AVX2/AVX-512 have already expired, so in theory, anyone can design and sell an x86 compatible processor if that is want one wants to do.
This includes x86-64 (AMD64) related patents needed for the 64-bit extension portion (AMD patented these).
Probably no one tries to design a new x86 design anymore since RISC-V provides (no pun intended) a risk free platform that is open and ISA (Instruction Set Architecture) itself is safe from litigation unlike x86 ISA in the '80s and '90s

andrea wrote on 2025-09-11, 20:31:

Assuming it works I'd say a cached 83MHz bus system is still a much better experience than 100MHz without. […]
Show full quote

Babasha wrote on 2025-09-11, 08:34:

I think its possible to solder my cache and tag to check is it work at 66-75MHz or is it places for just fake chips?

Assuming it works I'd say a cached 83MHz bus system is still a much better experience than 100MHz without.

Also while on the throne I was thinking, it is most certainly odd for NEC to make their own chipset for a PC-compatible. A repurposed/adapted PC98 chip also seems strange as they use completely different io-ranges and everything and also aren't single chip.
Could it be a single-chip design from another vendor (so SiS pretty much) made in their own fabs?

bracecomputerlab wrote on 2025-09-12, 07:50:

My guess as to who designed this mysterious Socket 7 chipset, I think it is RDC Semiconductor. […]
Show full quote

My guess as to who designed this mysterious Socket 7 chipset, I think it is RDC Semiconductor.

https://www.rdc.com.tw/index.php

It appears that they designed a 486SX compatible ASIC core 20+ years ago, but in the public documentation published by them that I happened to read, they will insist that it is a RISC processor and not an x86 processor.
It comes with one 8254, two 8259, and two 8237 wired in a way to work like a PCI based IBM-PC/AT compatible, so I guess if it walks like a duck and quacks like a duck, I guess it is an x86 SoC.

http://www.sima.com.tw/download/R8610_D06_20051003.pdf

DM&P Vortex86 DX and EX are likely the renamed version of their EmKore DE and DS.

https://www.rdc.com.tw/index.php?route=home/emk

That being said, early Vortex86 was a rebadged SiS550 with Rise Technology mP6 Pentium MMX class processor integrated.
I think RDC kept a low profile due to Intel's history of suing anyone trying to develop an x86 compatible processor.
Of course, not everyone got sued by Intel, but you get the point.

That being said, most x86 related patents except for SSE4/AVX/AVX2/AVX-512 have already expired, so in theory, anyone can design and sell an x86 compatible processor if that is want one wants to do.
This includes x86-64 (AMD64) related patents needed for the 64-bit extension portion (AMD patented these).
Probably no one tries to design a new x86 design anymore since RISC-V provides (no pun intended) a risk free platform that is open and ISA (Instruction Set Architecture) itself is safe from litigation unlike x86 ISA in the '80s and '90s

andrea wrote on 2025-09-11, 20:31:

Assuming it works I'd say a cached 83MHz bus system is still a much better experience than 100MHz without. […]
Show full quote

Babasha wrote on 2025-09-11, 08:34:

I think its possible to solder my cache and tag to check is it work at 66-75MHz or is it places for just fake chips?

Assuming it works I'd say a cached 83MHz bus system is still a much better experience than 100MHz without.

Also while on the throne I was thinking, it is most certainly odd for NEC to make their own chipset for a PC-compatible. A repurposed/adapted PC98 chip also seems strange as they use completely different io-ranges and everything and also aren't single chip.
Could it be a single-chip design from another vendor (so SiS pretty much) made in their own fabs?

Tnx for your appenition.
I read RDC history and its one of suspets for develop "NEC" chipset
But! If we agree with official RDC history time-line - https://www.rdc.com.tw/index.php?route=about/milestones
They are develop 8/16 bit MCU in 1997-99 and no 32bit system until late 2020. Who knows maybe they got secret chipset without x86 CPU core or maybe not.

I think you are reading it wrong.
If I switch the language to English and look at their corporate milestones, it mentions completing the development of a 32-bit microprocessor between 2000 to 2005.
They are cryptically referring to RDC R8610's 486SX compatible core mentioned, and furthermore, R8610 came out in 2005 if you take a glance at their 250+ page long databook referenced in my previous posting.
Again, they use this cryptic language for legal reasons (i.e., avoiding the wrath of Intel legal department).
That being said, some key patents of 386 architecture probably expired by 2005 or so, and I assume many aspects of 486 architecture were likely not be patentable due to prior art (i.e., someone else patented a similar design already).
Intel asserted certain 386 architectural patents like '338 patent or Crawford patent (named after John H. Crawford, the one of the principal designers of the 386 architecture).
Crawford patent was asserted against Cyrix back in the early '90s, and Microprocessor Report covered it extensively back then.

https://websrv.cecs.uci.edu/~papers/mpr/MPR/A … CLES/080202.PDF
https://websrv.cecs.uci.edu/~papers/mpr/MPR/A … CLES/060404.PDF

Babasha wrote on 2025-09-12, 15:43:

Tnx for your appenition. I read RDC history and its one of suspets for develop "NEC" chipset But! If we agree with official RDC […]
Show full quote

bracecomputerlab wrote on 2025-09-12, 07:50:

My guess as to who designed this mysterious Socket 7 chipset, I think it is RDC Semiconductor. […]
Show full quote

My guess as to who designed this mysterious Socket 7 chipset, I think it is RDC Semiconductor.

https://www.rdc.com.tw/index.php

It appears that they designed a 486SX compatible ASIC core 20+ years ago, but in the public documentation published by them that I happened to read, they will insist that it is a RISC processor and not an x86 processor.
It comes with one 8254, two 8259, and two 8237 wired in a way to work like a PCI based IBM-PC/AT compatible, so I guess if it walks like a duck and quacks like a duck, I guess it is an x86 SoC.

http://www.sima.com.tw/download/R8610_D06_20051003.pdf

DM&P Vortex86 DX and EX are likely the renamed version of their EmKore DE and DS.

https://www.rdc.com.tw/index.php?route=home/emk

That being said, early Vortex86 was a rebadged SiS550 with Rise Technology mP6 Pentium MMX class processor integrated.
I think RDC kept a low profile due to Intel's history of suing anyone trying to develop an x86 compatible processor.
Of course, not everyone got sued by Intel, but you get the point.

That being said, most x86 related patents except for SSE4/AVX/AVX2/AVX-512 have already expired, so in theory, anyone can design and sell an x86 compatible processor if that is want one wants to do.
This includes x86-64 (AMD64) related patents needed for the 64-bit extension portion (AMD patented these).
Probably no one tries to design a new x86 design anymore since RISC-V provides (no pun intended) a risk free platform that is open and ISA (Instruction Set Architecture) itself is safe from litigation unlike x86 ISA in the '80s and '90s

andrea wrote on 2025-09-11, 20:31:

Assuming it works I'd say a cached 83MHz bus system is still a much better experience than 100MHz without.

Also while on the throne I was thinking, it is most certainly odd for NEC to make their own chipset for a PC-compatible. A repurposed/adapted PC98 chip also seems strange as they use completely different io-ranges and everything and also aren't single chip.
Could it be a single-chip design from another vendor (so SiS pretty much) made in their own fabs?

Tnx for your appenition.
I read RDC history and its one of suspets for develop "NEC" chipset
But! If we agree with official RDC history time-line - https://www.rdc.com.tw/index.php?route=about/milestones
They are develop 8/16 bit MCU in 1997-99 and no 32bit system until late 2020. Who knows maybe they got secret chipset without x86 CPU core or maybe not.

For those following this thread, someone out of Australia does have one up for sale.
I wish I bought one from someone in Kazakhstan for about $75 a year or so ago.

https://www.ebay.com/itm/285715637505