mpe wrote on 2019-11-22, 23:59:

I will be using this DPT PM2022 with CM4000 cache module daughterboard as the main HDD controller. For now there is a single 16MB SIMM. I'd like to eventually add more if I can find parity SIMMs compatible with this one.

Based on a photo comparison, I believe I am the new owner of the DPT SCSI board, which is in my 50MHz build. I also possess the same (but different) model Adaptec card for floppy controller. Did you ever get to a point where you were able to run any meaningful benchmarks? I'm finding that the two boards are very similar in linear transfers (~9MB/s), but the cache makes a huge difference in access times for random transfers. The DPT board is a serious piece of kit. With the DM4000 array add-on board I can only guess it will do RAID with the correct configuration utility. I have not tried this yet, but would be fascinated to see the performance of interleaved drives.

Congrats to acquiring it!

Yes. I measured performance of the DPT PM2022 controller.

I measured 3329 kB/s linear (unbuffered) speed. For comparison Adapec AHA2742 (EISA) on the system can do almost double - 7217 kB/s with everything else being equal.

However, please note that I am not using regular SCSI-2 HDD. I am using SD2SCSI device which can pretty much max out the SCSI-2 bus.

With traditional rotating hard disks you are less likely to achieve such a high speeds and on the other hand caching aspect of the DPT controller can be more an advantage. With solid state technology the access time is pretty much non-existent so controller caching is not helping.

Can you post some benchmarks ?

Topbench
Speedsys
Doom
Drystone
VGA
Etc...

I have 18-19 MB/s on my AHA2740 on the FAST WIDE SCSI port.
It's a DX-50 too

But I don't get past 25 MBit/s with my 3COM 3C597-TX (Windows NT 4.0)
or 14 MBit/s (Windows 95)

How quick are your SIMMs?
I've noticed that my system does not run stable with 8 9x4 MBit 70 ns chips. The problems passed when I've replaced ONE bank with 2x16 Mbit + 4 MBit parity 60 ns SIMMs.
Now I use two banks with 2x16 MBit + 4 MBit parity 60 ns SIMMs.

Just noticed this old thread, do not know if anyone is interested but have a very similar build that has been in process for quiet a while now, several years to be exact, as things are just stupidly priced for such low spec old hardware, so mostly wait unit something is cheap i.e. under 60 UK pounds, a quick dirty picture is attached.

Quick specs,

Same motherboard as the original post but 64mb ram, 4167 Weitek, RTC chips have been changed and she has a AM5X86 (130+ Mhz),
SIIG IO 1831 Card for serial and parallel ports
ELSA Winner 1000 S3 928 (With 2mb)
DPT PM2022 with smart cache (64mb) and RAID module (also has the floppy port)
Running a 40x SCSI CD-ROM (50 PIN),
STD2401LW Tape Drive (saw it cheap and thought why not) (68 pin to 50 pin)
three HP ST39102LC 9.1GB 10K SCSI drives (SCA to 50 pin)
3COM Fast Etherlink XL (3C905)
SCSI SB16
Real Magic Card

The machine is mostly for DOS gaming but does play host to different operating systems when I decide/ have time to play around.

And to tell the truth I am still playing with it, future items are possibly:
Upgrade the cache chips to something faster, I know it will not achieve much but always done it in my builds, still a school yard brag.
Better CD ROM drive, annoys me it is all black, but the tray is grey.
Maybe change the case, the one it is in has been used for various builds and is rather bodged in places.
Would love to find the real ECC memory for the SCSI card but costs are stupid.
Experiment with some other processors, but it need to investigate the jumpers, limited information out there on the board, and would fail to post when I tried this before.
Would like to find some more EISA cards to play with, have an MPEG decoder laying around but cannot find the relevant config file.

simon_e_hall wrote on 2021-04-21, 12:47:

3COM Fast Etherlink XL (3C905)

pretty sure that's a PCI card. Can you explain how you get that working on a full EISA system? 🤔

This is the board:
http://www.win3x.org/uh19/motherboard/show/2523
Looks like we have a real-life pic to put there. And chipset info.

@simon_e_hall: any chance of making a bios dump? even more awesome: do you own the actual manual for it?

weedeewee wrote on 2021-04-21, 13:05:

simon_e_hall wrote on 2021-04-21, 12:47:

3COM Fast Etherlink XL (3C905)

pretty sure that's a PCI card. Can you explain how you get that working on a full EISA system? 🤔

3C509 would be an Etherlink III
ISA Fast Etherlink XL would be a 3C515.
🤔

weedeewee wrote on 2021-04-21, 13:05:

simon_e_hall wrote on 2021-04-21, 12:47:

3COM Fast Etherlink XL (3C905)

pretty sure that's a PCI card. Can you explain how you get that working on a full EISA system? 🤔

Very much an EISA card, I will double check the model number when I get home.

simon_e_hall wrote on 2021-04-21, 13:52:

Very much an EISA card, I will double check the model number when I get home.

a quick google leads to
3C597-TX 3Com 10/100 Fast Ethernet EISA Network Interface Card 😀

evasive wrote on 2021-04-21, 13:06:

This is the board: http://www.win3x.org/uh19/motherboard/show/2523 Looks like we have a real-life pic to put there. And chipset […]
Show full quote

This is the board:
http://www.win3x.org/uh19/motherboard/show/2523
Looks like we have a real-life pic to put there. And chipset info.

@simon_e_hall: any chance of making a bios dump? even more awesome: do you own the actual manual for it?

Wish I had the manual for this board, being trying to dig up further information on it for a long time now, so yes, the website shows some of the jumpers, there are some around the CPU (w21, w22, etc...) that are not listed, pretty sure they change the bus speed (but not had a chance/ time to just do trial and error changes), tried just swapping the crystal once, but no joy. The board is Freetech, as the EISA config file is !FCT0003.CFG. But the BIOS ID suggests it is a Genoa computer (026156) and the chipset ID doesn't help either (1017), so my assessment is it is all logic. The BIOS is a rather generic HiFlex AMI, probably as most of the configuration is done in the EISA configuration utility, was planning to mod it/ find a Mr. BIOS image in the future, but again will download it when I have a moment.

Posts with:
ROM BIOS (C)1990 American Megatrends Inc.,
10X-E2-XX-A

40-0100-026156-00101111-1017-F

weedeewee wrote on 2021-04-21, 13:55:

simon_e_hall wrote on 2021-04-21, 13:52:

Very much an EISA card, I will double check the model number when I get home.

a quick google leads to
3C597-TX 3Com 10/100 Fast Ethernet EISA Network Interface Card 😀

And you would be correct! Just looked at my old driver list, and the part number is 03-0051-003, which would make it a 3c597, just me getting confused between my systems, think the Etherlink III is in my P2 build.

simon_e_hall wrote on 2021-04-21, 14:05:

, there are some around the CPU (w21, w22, etc...) that are not listed, pretty sure they change the bus speed (but not had a chance/ time to just do trial and error changes), tried just swapping the crystal once, but no joy.

I no longer have this build, but I was able to find out the purpose of some of these undocumented jumpers.

Will try to find my notes.

None of these pins controls bus clock as there is fixed 50 MHz oscillator on the board.

I remember there was a kill switch fo the weitek socket and others for parity checking. Nothing really interesting.

Did you ever finish building this computer and are you satisfied with its performance ?

I finished it, but no longer have it and moved to other projects. The performance was adequate - a bit on slow side. Especially in graphics - simply there is no local bus and that's the biggest issue.

I even tried to install AMD 5x86 @ 150 MHz which worked, but did not fly compared to DX2-66 on a modern 486 PCI motherboard in DOS/Windows application tests.

mpe wrote on 2021-04-21, 19:41:

I no longer have this build, but I was able to find out the purpose of some of these undocumented jumpers. […]
Show full quote

simon_e_hall wrote on 2021-04-21, 14:05:

, there are some around the CPU (w21, w22, etc...) that are not listed, pretty sure they change the bus speed (but not had a chance/ time to just do trial and error changes), tried just swapping the crystal once, but no joy.

I no longer have this build, but I was able to find out the purpose of some of these undocumented jumpers.

Will try to find my notes.

None of these pins controls bus clock as there is fixed 50 MHz oscillator on the board.

I remember there was a kill switch fo the weitek socket and others for parity checking. Nothing really interesting.

Thank you that would be great, but also confirms my suspicions with this socket, that I have probably got the best CPU for the job in there, unless I downgrade to a 66 for better compatibility, but probably will not do that at this time. However, did you identify the correct pin out for the key lock/ power LED connector/ header? Also I notice the turbo connector/ header does nothing, constantly at full speed despite what ever I do to it.

I am still not 100% on the graphics card, but wanted to do as much on the EISA bus as possible, I don't know if I am being too critical after using a Diamond Speedstar 64 (2mb) on my previous 486 build. Superscape tests between the two are only about 3~4 points out between the two and Duke 3D is playable, something just doesn't seem right, but cannot put my finger on it.

It is nice to see some interest still in these old motherboards though.

Here is a quick benchmark of it, apologies for using MS-DOS 7.10 just experimenting with FAT32 on the 9GB SCSI disk to see any issues. Hard drive performance is not completing, I assess the cache I have is upsetting the test.

mpe wrote on 2021-04-21, 20:53:

I even tried to install AMD 5x86 @ 150 MHz which worked, but did not fly compared to DX2-66 on a modern 486 PCI motherboard in DOS/Windows application tests.

Eh? A DX2-66 even on a PCI system should get its ass handed to it by a 5x86-150, even on a crappy old ISA 486 board....unless something else is being severely bottle-necked like graphics or disk.

Bus speed is the bottleneck. EISA cards are no faster than ISA and graphics performance is what's holding it back.

With ELSA Winner 1000 in the EISA slot the Am5x86 @ 150 can produce about 20fps in DOOM on this motherboard. Only ~4ps more than with DX-50 with three times slower clock. An average DX2-66 with basic VLB/PCI card can easily do 27-30fps.

Even L2/RAM latency is much slower which holds the Am5x86 back significantly. You see speedsys results above (38MB/s L2, 27MB/s RAM). I got almost identical results. Any DX2-66 on a more modern motherboard (Sis496, UMC, Ali, Intel 420EX) can easily beat that. And that's on 33 MHz vs 50 MHz bus.A 1990 tech just can compete with the one from 1995...

Of course in pure CPU ALU speed the 5x86 beats the DX2-66 any time but in those days that was somewhat less important.

mpe wrote on 2021-04-23, 11:41:

Bus speed is the bottleneck. EISA cards are no faster than ISA and graphics performance is what's holding it back.

EISA has a 32 bit data bus so would be four times as fast as an equivalently clocked 8 bit ISA bus or twice as fast an equivalently clocked 16 bit isa bus.
and it has a 32 bit address bus.