Found an interesting 386DX system from the 80s, France, the Goupil G50.

It had four slots special RAM modules.

http://msx.fab.free.fr/mpc2/goupil/goupil.htm

And the site below ssays something about 16 MB RAM.

http://silicium.org/site/index.php/28-catalog … -s-m-t-goupil-8

So I assume that each module held a huge capacity of 4 MB at maximum.

If production had continued, larger modules might have been possible a few years later.

The size of the modules was a multiple of standard SIMMs.

Jo22 wrote on 2023-01-11, 22:06:

Found an interesting 386DX system from the 80s, France, the Goupil G50. […]
Show full quote

Found an interesting 386DX system from the 80s, France, the Goupil G50.

It had four slots special RAM modules.

http://msx.fab.free.fr/mpc2/goupil/goupil.htm

And the site below ssays something about 16 MB RAM.

http://silicium.org/site/index.php/28-catalog … -s-m-t-goupil-8

So I assume that each module held a huge capacity of 4 MB at maximum.

If production had continued, larger modules might have been possible a few years later.

The size of the modules was a multiple of standard SIMMs.

1 Holy crap how have I not heard of SMT Goupil before?!?!?! All of their computers are soooo sexy!
2 Sexy slim executive desk model Goupil Golf SX also uses 96 pin DIN connector:

https://computarium.lcd.lu/photos/albums/Goup … C01169_txt.html
https://computarium.lcd.lu/photos/albums/Goup … 01196_text.html

If you look carefully and squint hard enough you can see chips on the adapter suggesting this connector exposes raw CPU bus and you plug combo memory controller + ram cards.

Fascinating company, very stylish and trying their own thing. Reminds me of Olivetti.

Edit: they seem to have had a thing for external ram controllers? http://msx.fab.free.fr/mpc2/goupil/g6.htm is another one

rasz_pl wrote on 2023-01-11, 22:47:

Edit: they seem to have had a thing for external ram controllers? http://msx.fab.free.fr/mpc2/goupil/g6.htm is another one

Interesting cache controller on that motherboard

• On Die 16KB 4 way set associative cache expandable to 64kb if you put 4 of these chips on the board
• 20 bits of tag address
• Full Local Bus/System Bus concurrency so CPU can run from the cache while bust master devices run from main memory

Here's the only thing I see on retroweb for that

https://theretroweb.com/motherboards/s/abc-co … puter-com-a3865

It was limited to 40MB with 5 SIPPs

The talk of virtual memory in this thread has got me thinking... has anyone ever tried running a somewhat modern SSD on a 386 and using it for virtual memory? It would have to be attached via a SATA to IDE adapter, connected to the best ISA or VLB IDE controller you could find. If some combination of those can work (I use a 16GB M.2 SSD ->SATA ->IDE adapter in my 440BX test system), how noticeable would the performance penalty be once physical memory was used up?

Obviously the controller or the interface will be the biggest limitation (less so with VLB), but with the improved latency of a modern SSD I have to wonder what it'd be like if you had like 1MB or 4MB of physical memory and as much solid state virtual memory as you could possibly ever need.

... though I must admit that I have no experience with virtual memory prior to Windows 9x, so I have no idea how it would be utilized.

Anything attached to the ISA bus is going to be (at least) several times slower than even a 386SX main memory access speeds - most will struggle to get more than single-digit IDE transfer speeds.... that's even with modern IDE drives with their large caches. It would crawl.

A VLB IDE controller may be different - you *might* (might) get a few tens of MB/sec from one of those - and that is approaching the memory speed of a 286 or 386SX (in the region of 10-25MB/sec) , but a 386DX would have some multiple of that.

In terms of actual use - the easiest method I can think of is for a DPMI (protected mode) application where additional virtual memory (can be) used via pagefile on disk. Windows 3.1x can also utilise an on-disk swap file. Dos (at least the normal dos we all know and love) doesn't have the concept of virtual memory - though some memory managers (like the DPMI providers) do.

Ozzuneoj wrote on 2023-01-12, 07:00:

... though I must admit that I have no experience with virtual memory prior to Windows 9x, so I have no idea how it would be utilized.

You have to be running in protected mode to use virtual memory, so that pretty much rules it out for plain DOS. Closest you could come to a protected mode DOS would be something like Quarterdesk Desqview 386.

It would be like taking an ISA memory card (terrible to begin with) and wrapping it inside the PATA protocols ( which would take terrible to completely awful ).

Not to mention bus saturation and a host of other issues. Back in the PC XT days, the ISA bus ran at twice the CPU frequency, making memory on the ISA bus reasonable. In the 386 days, the ISA bus ran at 1/2 to 1/5 of CPU speed.

So interesting idea from a scientific perspective. Total waste of time if you want to do any sort of real world process.

douglar wrote on 2023-01-12, 02:01:

Interesting cache controller on that motherboard […]
Show full quote

rasz_pl wrote on 2023-01-11, 22:47:

Edit: they seem to have had a thing for external ram controllers? http://msx.fab.free.fr/mpc2/goupil/g6.htm is another one

Interesting cache controller on that motherboard

Untitled.png

Untitled1.png

• On Die 16KB 4 way set associative cache expandable to 64kb if you put 4 of these chips on the board
• 20 bits of tag address
• Full Local Bus/System Bus concurrency so CPU can run from the cache while bust master devices run from main memory

Here's the only thing I see on retroweb for that

https://theretroweb.com/motherboards/s/abc-co … puter-com-a3865

It was limited to 40MB with 5 SIPPs

This is rarely used. Only found on compaq 4/33i and 3/33i and 4/25i and 3/25i (these four models that used convertible 386 or 486 motherboard using jumpers and switch, and deskpro/M with 386 processor board. Far as I know.

I'm sure there was few commerical or specialized boards using this.

zero wait state on the bus for 386, 395 and 387 group.

Cheers,

The Sequent Symmetry S81 supported up to 384MB and up to 30 20 MHz Intel 80386 processors! I wonder what it cost...

douglar wrote on 2023-01-12, 12:39:

You have to be running in protected mode to use virtual memory, so that pretty much rules it out for plain DOS. Closest you cou […]
Show full quote

Ozzuneoj wrote on 2023-01-12, 07:00:

... though I must admit that I have no experience with virtual memory prior to Windows 9x, so I have no idea how it would be utilized.

You have to be running in protected mode to use virtual memory, so that pretty much rules it out for plain DOS. Closest you could come to a protected mode DOS would be something like Quarterdesk Desqview 386.

It would be like taking an ISA memory card (terrible to begin with) and wrapping it inside the PATA protocols ( which would take terrible to completely awful ).

Not to mention bus saturation and a host of other issues. Back in the PC XT days, the ISA bus ran at twice the CPU frequency, making memory on the ISA bus reasonable. In the 386 days, the ISA bus ran at 1/2 to 1/5 of CPU speed.

So interesting idea from a scientific perspective. Total waste of time if you want to do any sort of real world process.

Thanks for the info. How much of a difference do you think a good VLB EIDE card would make for this task? If it actually works it seems like that would eliminate a lot of the BUS limitations.

Ozzuneoj wrote on 2023-01-12, 17:43:

Thanks for the info. How much of a difference do you think a good VLB EIDE card would make for this task? If it actually works it seems like that would eliminate a lot of the BUS limitations.

Let me get this straight. You are essentially asking if it helps to use a Sata drive on a VLB controller for your paging file on a 386, yes?

I had fun running Win98 on a 386sx with fast CF & 8mb ram during Covid quarantine. Took about 8 hours to install. It was definitely digging into the paging file during its 15 minute boot up, but I'm not sure that adding another 8MB of ram or a faster page file was going to make a big difference in usability there. The same CF device would boot Win98 in 0:16 seconds on an Athlon 2500+ w/ UDMA-5. I just don't feel like switching to a controller that had a faster bus-master IO for that 386 was going to change much of anything. Maybe theoretically it could boot in 14:00 instead of 15:00 with a SATA device at DMA-2 instead of a PATA device at PIO-0 if such a setup were possible. Maybe it boots in 5 minutes with another 8mb RAM. The extra ram is likely the way to go, except 8mb was max for that board. It didn't like 4MB simms. Tried adding 8MB via RAMPAT+ for giggles, but the BIOS didn't see it.

So even if you find a 386 system that 1) works with VLB bus mastering and 2) has a controller that talks "DMA 2" because "UDMA-2" didn't exist until after VLB died, and 3) you attach a contemporary solid state device that is somehow willing to negotiate a dead end branch of the ATA protocol that was abandoned 10+ years before your device was built, and 4) you get your OS to support the same dead end branch of the ATA family tree, you are still going to be so CPU bound on everything else that when your OS starts swapping memory pages to and from your page file , you won't see any difference in usability between a legacy drive or a SATA drive. You will still be benchmarking your time to complete tasks in minutes , not seconds. 386 systems were pretty much CPU bound on everything, even when PIO wasn't part of the picture. The added CPU load needed to complete swap tasks just makes most things unusable even if the CPU didn't have to push bytes to and from the disk. In most cases, a 386 CPU would have been idled during busmaster activity anyway.

edit - Cleaned up some of the phrasing

douglar wrote on 2023-01-12, 18:53:

Let me get this straight. You are essentially asking if it helps to use a Sata drive on a VLB controller for your paging file on a 386, yes?
...

:]
TLDR: there is nothing that would run well on a 386 and require more than 8MB of ram. Same deal with 486 and ~12MB of ram.

rasz_pl wrote on 2023-01-13, 01:39:

douglar wrote on 2023-01-12, 18:53:

Let me get this straight. You are essentially asking if it helps to use a Sata drive on a VLB controller for your paging file on a 386, yes?
...

:]
TLDR: there is nothing that would run well on a 386 and require more than 8MB of ram. Same deal with 486 and ~12MB of ram.

There we’re definitely windows programs that ran a lot better on a 486 with 20mb of ram over 12mb

rasz_pl wrote on 2023-01-13, 01:39:

TLDR: there is nothing that would run well on a 386 and require more than 8MB of ram. Same deal with 486 and ~12MB of ram.

The TLDR is that it is adding more ram has a much bigger impact on performance than building a faster page file. Usually cheaper too.

Thanks for explaining things regarding virtual memory on a 386. Obviously this thought experiment serves no practical use... arguably neither does trying to make a 386 system "better" in 2023... but here we are on VOGONS, building these things anyway. 😁

The question "Are modern drives fast enough to actually be indistinguishable from RAM on something as slow as a 386?" came to my mind and it sounds like the answer is simply that between the extremely low speed of a 386 and the platform\software it has to work with, it's just not fast enough to actually allow a modern drive to provide enough benefit.

Of course more RAM is always going to be better than virtual memory in absolutely any situation (as long as you have the RAM). But.... that probably won't stop me from experimenting with it on some kind of solid state storage anyway if I can ever get my Wintech\Edom MV0008 386\486 VLB board working. But that's a topic for another thread...

Thanks again!

This is about as practical as maxing out the 48-bit memory bus on an Athlon 64

rmay635703 wrote on 2023-01-13, 02:51:

rasz_pl wrote on 2023-01-13, 01:39:

douglar wrote on 2023-01-12, 18:53:

Let me get this straight. You are essentially asking if it helps to use a Sata drive on a VLB controller for your paging file on a 386, yes?
...

:]
TLDR: there is nothing that would run well on a 386 and require more than 8MB of ram. Same deal with 486 and ~12MB of ram.

There we’re definitely windows programs that ran a lot better on a 486 with 20mb of ram over 12mb

Windows 95 RTM ran better with 16 MB than with 8 MB or 4 MB (yikes!).
OS/2 needed 8 MB as a real minimum requirement, not 4 MB.

Edit: Windows for Workgroups or other network OSes (Novell Netware etc)
always needed a lot of RAM:
These systems were maintaining the internal e-mail traffic in offices.
I'm talking about the server(s) here, not the desktop PCs.

However, hardware-wise, they didn't differ that much.
In the late 80s/early 90s, a "server" was just a powerful PC in a closet designated
to the task of serving resources to other PCs in the network.

These servers also did printer spooling etc,
so that multiple people had access to the laser printer, for example.

To these systems, the CPU isn't/wasn't so much the bottleneck, but the RAM.

PS: That's the main cause of the misunderstanding here, I think:
Many people see the whole IT with the eyes of the consumer, the office user.
The professional side is totally ignored, which often makes me smile.

It's as if C64 users with their 64 KB machines saw an IBM PC/XT with 640 KB first time
and
start debating if its insane memory capacity is just a curiosity or if there are two more machines world wide with such a RAM specification. 🥲

Anyway, some might be surprised to find out how advanced PCs and workstation in the professional, medical or research fields already were in the 80s/early 90s! 😃
A far cry from beige boxes with 2MB RAM, 40 MB HDDs and Standard VGA (640x480 in 16c) and 14" CRTs.

Edit: To give an idea, Kodak's Photo CD supported resolutions up to 2048x3070 (BASE*16, true color). In 1992/93, maybe earlier.
And that's merely the consumer version.
The maximum resolution of the format was 4096x6144.
Again, in 1992/93. In the 386/486 days, before 586/Win 95.

Re: Graphics card for Windows 3.11 and Nt3.51 in high resolutions (1600x1200 and higher)

rasz_pl wrote on 2023-01-13, 01:39:

:]
TLDR: there is nothing that would run well on a 386 and require more than 8MB of ram. Same deal with 486 and ~12MB of ram.

That's quite a statement to make. You know every single use case of a computer with a 386 during its production run, of course, yes?

megatron-uk wrote on 2023-01-13, 15:41:

rasz_pl wrote on 2023-01-13, 01:39:

:]
TLDR: there is nothing that would run well on a 386 and require more than 8MB of ram. Same deal with 486 and ~12MB of ram.

That's quite a statement to make. You know every single use case of a computer with a 386 during its production run, of course, yes?

*today, unless you have a time machine 😀