VOGONS


VL/EISA SMP Pentium (turbocharged)

Topic actions

First post, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

This thread is to be used to report my findings with a VL/EISA Socket4 Pentium system I've been playing with. I am very interested in playing with expensive high-end toys and PC upgrades available in the 1992-1994 time frame. In this case, VL-BUS, EISA and Socket4 Pentiums. It's something hard to appreciate unless you were into PCs at that time. In 1993 the Pentium held near mythical status, and there was still uncertainty about which 32-bit card expansion bus was the way of the future. This is why this ridiculous motherboard exists (it's one of perhaps only 2 or 3 to feature both EISA and VL-bus with a Pentium CPU):

IMG_0845.JPG
This is the Acer J3 active backplane motherboard. It mainly featured in the the Acer Power and Acros line of computers, but was also found in other OEM systems like AMBRA, Apricot and perhaps a few Japanese makes as well. Several different CPU cards are available: single socket3, single socket4, dual socket4, single socket5 and dual socket5. As far as I know they are all based around OPTi chipsets, though I did hear a rumor a Triton board may exist.

J3%2520CPU%2520card.jpg

I own the dual socket4 card. Mine came with two mismatched P60s. One has FDIV and the other does not. They do not have the same sspecs. I am not sure how well it would work in an SMP enabled OS. Since I do not have the matching heatsinks for these chips (I don't want to use epoxy) I am unable to find out. In the mean time I have replaced these two CPUs with a single PODP5V133 upgrade chip. I replaced the oscillator with a 66MHz part to get the full 133MHz. The board has silk screen indicating support for 80MHz FSB, but I have not yet been brave enough to try it out. The card also supports 1MB L2 cache. I am not sure if this card does bank interleave, and I haven't bothered to test the 1MB setting yet (though I do have the chips to do so).

Performance of this board is not all that great. CPU marks are about where they should be, but the DRAM throughput is kind of shite. This is pretty consistent with most of the OPTi based boards I've tried for 386 and 486 systems. This appears to be a jerry rigged 486 design. My SiS EISA board eats this thing for lunch in this regard. However, I am hoping perhaps I can get a main memory performance boost by using a CPU with integrated L2 cache. This is possible using an adapter like this one. It's currently en-route from Germany. (Thanks to 386_Junkie for this one). I will post benchmark results when available.

DSC_4139.JPG

For VGA I have settled on #9 Motion FX 771 with 4MB VRAM. This is just about top of the line for VLB. Somehow I only managed to pay a few dollars for this card. This card is somewhat special for a VLB card because of the 220MHz RAMDAC and the VRAM which is mounted on both sides of the PCB. I attempted to run et4000W32P in my system as well, but neither of my cards would power up. This was the same problem I had in my previous OPTi based VLB Pentium system.

9fx.jpg

My disk system will comprise of an Adaptec AHA-2742W EISA based wide SCSI adapter. This one is fairly basic, but it's my understanding that caching controllers on EISA systems with lots of memory are somewhat useless. My motherboard supports up to 512MB of FPM DRAM with the CPU card I have, and I currently have 256MB installed, so I figured it should be more than enough to keep things running smoothly. The drive I will be using is a Quantum/Maxtor Atlas IV 36GB (10k).

EISA-Adaptec-AHA2742W.JPG

A stupidly complex computer deserves a stupidly long and complex soundcard. AWE32 is pretty much the perfect match. They were also stupidly expensive when they were knew. It's just a pity I sold the original CT2760 card several years back. The CT3990 is close enough though. Not sure if I will use a daughter board or not. Probably not worth the hassle in light of the hanging note bug.

ct3980.jpg

Not sure what other toys will go in this system at the moment. For now it doesn't have a dedicated chassis and I've stolen the Inwin H500 from my EISA 486 build. It will likely be another ATX conversion though.

Last edited by Anonymous Coward on 2016-01-03, 03:02. Edited 3 times in total.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 3 of 81, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

It's probably not as hot as you think. I believe those CPUs are 15W each, and can get by with just a heatsink. People made a big deal about the heat when the Pentium came out, but it was really just hot compared to a 486. Modern CPUs are complete pigs compared to the P60.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 4 of 81, by Artex

User metadata
Rank l33t
Rank
l33t

Interesting! I fired up my Gateway 2000 P5-60 and I think you're right. It's got a passive heatsink and really is only warm to the touch after an hour of gaming.

My Retro B:\ytes YouTube Channel & Retro Collection
LihnlZ.jpg

Reply 5 of 81, by Sutekh94

User metadata
Rank Oldbie
Rank
Oldbie

Heh, I was messing with my Packard Bell Legend 300CD earlier and the CPU really doesn't get that warm, for it only having a passive heatsink. Though to be fair, there is also a case fan also cooling it.

That one vintage computer enthusiast brony.
My YouTube | My DeviantArt

Reply 6 of 81, by j^aws

User metadata
Rank Oldbie
Rank
Oldbie

Do you have any docs/ manuals for that Acer J3 board? I've got the same board but with a 486 module instead, which incidentally looks like a precursor to Slot 1 CPU cartridges.

When I get some time, I'll be looking to build my first EISA system. This era passed me by very quickly, and I ended up building PCI systems because it was going to be the next standard. Justified in hindsight, as PCI has lasted around 20 years or so. I remember EISA being a big deal back in the day, especially when reading Byte magazine.

I'm trying to ascertain whether this board needs some kind of special power supply with the extra connector in the main picture?... Also, I can't figure out whether this board has a Turbo header or not... there seems to be a jumper block "J5" that looks unlabelled? Not a big deal if it doesn't. Any info to get this great looking board up and running would be appreciated. Thanks in advance.

Reply 7 of 81, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

I do have some documents, but I think they are not 100% complete. Have you seen UKT Support:

http://www.uktsupport.co.uk/acer/mb/j3.htm

They have some jumpers listed. I think I found a second website as well, but I'll have to check my notes. I don't think I found a proper explanation for the front panel connectors, so I had to do some guesswork to get the reset and power LED in the right places. My case does not have a turbo button, so I didn't explore a possible jumper location, but it's possible it's not supported.

The extra power connector seems to be optional. I'm not using it and everything seems to be okay. It's probably not needed unless you're running 8 full length EISA cards.

Check my old posting for EISA CFG files:

http://www.vintage-computer.com/vcforum/archi … hp/t-35167.html

http://bbs.actapricot.org/files/area41/eisacfg.exe

There's even a TH99 entry:

http://museum.ttrk.ee/th99/m/A-B/32504.htm

Unfortunately it doesn't help us much with power, turbo and reset pin positions. At the very least I am sure reset is contained in the J5 block. I discovered this by randomly probing around. I was also able to get my power LED light going on J5 as well, but I am not 100% sure it's in the correct position. It could be the turbo LED for all I know. With a multimeter you can at least figure out which pins are power and which are ground. Turbo LED and switch pins (if present) are probably traceable back to the BIOS pins. The original Acer Power 9000 case seems to have a turbo LED on it, but the picture is not clear enough to see if a turbo switch is present. It's possible the turbo function was controlled entirely through key combinations.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 8 of 81, by j^aws

User metadata
Rank Oldbie
Rank
Oldbie

Nice one! Cheers for all the links, some of them I had already stumbled upon, and one of them has helped me eliminate jumper block "J7". Here's another TH99 link with slightly different but complementary info:

http://museum.ttrk.ee/th99/m/A-B/32293.htm

... However, no mention of a Floppy drive header in any of the aforementioned links, especially the above link where all sorts of other headers are labelled, which makes me think there are revisions to this board with, and without, this header. I found a PDF manual for this board mentioning this header:

Filename
152202-3.pdf
File size
854.06 KiB
Downloads
322 downloads
File comment
Acer J3 manual
File license
Fair use/fair dealing exception

Also In this manual, I couldn't find reference to J5 and J7 blocks, but thanks to one of your links above, I've found out they are for SCSI LED and optional mouse/ keyboard, respectively. And the J5 block for SCSI LED seems to have too many pins, which leaves some pins for further investigation...

BTW, which pins did you use for the Power switch connection to the case? It should be obvious, but I must be missing something...

Reply 9 of 81, by Robin4

User metadata
Rank l33t
Rank
l33t

9fx.jpg

I should clean those isa / VLB connectors with a white kind of pencil eraser.. How its now, its really oxidated..

I had the same problem on some cards of mine, but when cleaned they just looked as new again.

The pencil eraser iam using is an white one from hello kitty.

Something like this: NEW-1992-SANRIO-HELLO-KITTY-ERASER-HELLO-KITTY.jpg but with a different measurement.

~ At least it can do black and white~

Reply 10 of 81, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

The power connector is connected to the power supply. It's an ATX to AT conversion.

The apricot manual appears to be the same one I provided in my vintage-computer.com link.

The more likely reason the TH99 listing for the Apricot board is missing the header is because they simply forgot. I've found more than plenty of mistakes on TH99 over the years.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 11 of 81, by j^aws

User metadata
Rank Oldbie
Rank
Oldbie

I did a quick test and the board boots up with BIOS V1.2 R2.4, and the Dallas RTC still has the correct time! I thought I'd need to solder a new RTC, but thankfully, I just noticed it's socketed for easy replacement.

I'll put this aside for a future project when I get all the parts together. Still haven't decided if I'm going SCSI or not, nor decided if I'm going to use all these on-board connectors and save slots for other cards...

Reply 12 of 81, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Very nice build. I have a few observations/questions.

What's with the external green wires on the CPU card? Was that an expensive post-production "opps", or something you did?

You noted that you have 256 MB installed now, but have you tried 512 MB? Is the 256 MB installed on the motherboard, or CPU card? Have you tried splitting half on the MB and half on the CPU card, all on the MB, and all on the CPU card to see if one configuration allows for faster CMOS memory/cache timings?

What is up with the 8 DIP-28 cache sockets mixed with the 8 DIP-32 cache sockets? There appears to only be a DIP-32 TAG. How does this work? Are there also 5 DIP-20 cache sockets below?

Which cache configuration allows up to 1 MB of L2 cache? Won't 16 DIP-28 cache sockets only accomodate up to 512 KB of cache? I wonder what would be the benefit of running 16 DIP-28 32kx8 cache modules vs. 8 DIP-32 64kx8 modules? Can you mix 8 DIP-32 128kx8 modules with 8 DIP-28 32kx8 modules for a total of 1280 kb of L2 cache? Is there any way to cache all 512 MB of memory? Will the CPU card cache memory on the motherboard, or only memory on the CPU card?

While I have never gotten into EISA, I would certainly jump at the chance to get a dual 486 system running, particularly a dual Cyrix 5x86-133 or X5-160. Even a dual Cyrix MII would be quite the novelty.

Plan your life wisely, you'll be dead before you know it.

Reply 13 of 81, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

The green wires on the CPU card are indeed patchwork from the factory.

I have not tried 512MB of memory because I don't have enough 64MB modules. I have tried running 256MB using four 64MB SIMMs and eight 32MB SIMMs. I didn't notice any problems with either configuration. The BIOS is your typical weird OEM bios (Ambra in my case) so there's really not a lot to play with anyway.

The cache situation is like on your Biostar PCI 486 board. You cannot mix and match the SRAMSs. As this is a Pentium with 64-bit memory, you have twice as many cache sockets. I do not actually know if it can interleave the cache banks, but I suspect not. In that case I will probably upgrade it to 1MB. I need to do further testing first.

1MB can only be attained with eight 128kx8 modules. This board does not support 64kx8 modules as far as I know.

I am not yet sure how much memory can or cannot be cached. I will need to run more tests.

Last edited by Anonymous Coward on 2015-01-08, 12:02. Edited 1 time in total.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 14 of 81, by feipoa

User metadata
Rank l33t++
Rank
l33t++
Anonymous Coward wrote:

You cannot mix and match the SIMMs. As this is a Pentium with 64-bit memory, you have twice as many cache sockets.

I assume you meant, "you cannot mix and match the cache DIPs".

I am confused by your second sentence referring to 64-bit memory. If you upgrade to 1 MB L2 cache, you will still be using 8 128kx8 DIP sockets, which is the same as the cache configuration on a 486.

Plan your life wisely, you'll be dead before you know it.

Reply 15 of 81, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

Correct. I goofed up (but I fixed it).

About the 64-bit cache, the issue I was trying to point out was that I don't know for sure if the board interleaves the cache like on a 486 board. Therefore if I just use one bank (8 SRAMs on a Pentium), I could possibly be taking a performance hit. Filling 8 sockets on a Pentium is like filling 4 on a 486.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 16 of 81, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Aside from the interleaving issue, there could be an issue whereby using a single cache bank may require slower CMOS timings. For example, using eight 64kx8 SRAM modules (512 KB) for the L2 cache vs. four 128kx8 SRAM modules (512 KB) on a 486. The former can handle tighter CMOS L2 timings for FSB's at 40 MHz or higher.

Plan your life wisely, you'll be dead before you know it.

Reply 18 of 81, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

Intel actually had a Socket4 chipset called the 430LX "Mercury". I believe that was one of the better ones. The SiS Socket4 chipset (501/502/503??) is supposed to offer decent performance as well. So far I can't speak very highly of this OPTi chipset, mainy because of the crumby memory performance. In the PC Magazine review they said this motherboard equipped with a P66 didn't perform much better than their P60 systems...not that there would be much of a difference anyway. From what I gather Pentiums weren't really worth it until pipelined burst cache became available.

If you want to know how different this board performs than an Intel 430FX with async cache, probably quite a bit slower.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 19 of 81, by j^aws

User metadata
Rank Oldbie
Rank
Oldbie
Anonymous Coward wrote:

...
For VGA I have settled on #9 Motion FX 771 with 4MB VRAM. This is just about top of the line for VLB. Somehow I only managed to pay a few dollars for this card. This card is somewhat special for a VLB card because of the 220MHz RAMDAC and the VRAM which is mounted on both sides of the PCB. I attempted to run et4000W32P in my system as well, but neither of my cards would power up. This was the same problem I had in my previous OPTi based VLB Pentium system.
...

Can this Tseng issue be related to the motherboard BIOS and/or VGA BIOS?

FYI, I get the complete opposite with the 486 Socket 3 module for this motherboard, i.e. I can get an ET4000W32P working but not an S3 968P - both VLB cards. IIRC, this motherboard requires different BIOS versions for the Pentium and 486 modules.