[MrWeedster]

Hi.
Recently i got luck and found and affordable Compaq/Digital Personal Workstation 500au.
I wonder if it's possible to use standard REG ECC SD-RAM with it?
A test with two sticks of Kingston KVR133X72RC3/256 "standard PC REG ECC RAM" failed. The memory gets detected at boot, but the system crashes shortly afterwards.
Now i wonder if it's just an issue with PC133 or that "standard PC REG ECC RAM" simply just doesn't work with the system.

Does anybody have experience with "foreign" RAM in an PWS?

Also: Do standard radeon 7500 pci cards work in them? I have Tru64 5.1B running atm.

[johnnym]

Recently i got luck and found and affordable Compaq/Digital Personal Workstation 500au.

Congratulations! A true classic.

I wonder if it's possible to use standard REG ECC SD-RAM with it?
A test with two sticks of Kingston KVR133X72RC3/256 "standard PC REG ECC RAM" failed. The memory gets detected at boot, but the system crashes shortly afterwards.
Now i wonder if it's just an issue with PC133 or that "standard PC REG ECC RAM" simply just doesn't work with the system.

I believe I've made the same experience: it starts promising but crashes in OS. But I am not sure what I have installed in my PWS 500au currently. I can have a look tomorrow, it is in storage.

When looking at the Digital Technical Journal, Volume 9, page 56, figure 7 there already seem to be buffers between the memory slots and the 21174, so maybe additional buffers on the memory modules are making problems. Figure 8 shows what looks like a buffered ECC DIMM beneath a B-Cache module - which looks like it implies those could work, too - but I believe figure 7 only shows unbuffered ECC modules installed on the MLB.

Also: Do standard radeon 7500 pci cards work in them? I have Tru64 5.1B running atm.

Not sure, mine had an ELSA one in it, but I didn't use it. Maybe it will work with AlphaBIOS running. I know that newer AlphaStations like the DS15 could use Radeons, but if the PWS's SRM can work with them I don't know. Well, you alway can access the machine via the serial console when running the SRM, so maybe just give it a try.

[MrWeedster]

Thanks for the info.
I dropped in a standard Radeon 7500 PCI from the PC sector. It does work somehow, but just as standard-VGA adapter with 640x480.
Regarding the RAM maybe it's an issue with different timings. I'll try with the kingston sticks alone.

[johnnym]

I dropped in a standard Radeon 7500 PCI from the PC sector. It does work somehow, but just as standard-VGA adapter with 640x480.

How is it shown in SRM if at all?

Regarding the RAM maybe it's an issue with different timings. I'll try with the kingston sticks alone.

I have two modules of each of the following unbuffered ECC DIMMs installed in my PWS 500au, providing a total of 704 MiB of RAM:

• MemoryWorkx MXK128/DCPW43, 64 MiB - unfortunately no info on the WWW about the DIMM, but the chips are KM48S8030BT-F10 good for 100 MHz
  
• SEC KMM374S403CTL-G0Q, Compaq part # 268308-002, 32 MiB, PC66 compliant but according to the datasheet can run at up to 100 MHz
  
• Kingston KVR100x72C2/256, 256 MiB, 100 MHz
  

[johnnym]

Hey, looks like these double-height memory modules really exist for Personal Workstations:

https://www.ebay.de/itm/133484664080
https://www.ebay.de/itm/225365908048

To my eyes these look very much like buffered ECC DIMMs (except maybe for the square chips in addition to the buffers). No idea, why the buffered ECC DIMMs from PCs won't work correctly in a PWS. Maybe I should retest with some PC-100 modules.

EDIT:
One those above has part number 54-25158-DA, which seems to be a 256 MiB (32MBx72) 83MHz module according to https://www.mkd-gmbh.de/gebraucht.php. There is also a PC100 ECC registered module listed manufactured by Camintonn.

Other thought, maybe buffered/registered DIMMs require a specific jumper/dip switch setting?

E.g. a Sun Blade 150 would only run with buffered ECC (PC) DIMMs with a specific jumper set on the system board but otherwise only worked with unbuffered ECC DIMMs. Strangely the jumper is labelled wrongly so implied the opposite IIRC.

@MrWeedster:
Any luck with other buffered ECC DIMMs?

[johnnym]

Short of any other results, I did check my inventory for maybe compatible buffered/registered ECC DIMMs and though some didn't work, some did indeed work - at least for the time frame I tested them!

I don't remember where these two originally came from, but they did work in my PWS 500au and it survived both kernel boot (Linux 6.x) and operation (7za b -md=2m, openssl speed -elapsed) with them installed:



Those are CENTON CCB256M/REG133, though not identical modules (different PCB revision and different SDRAM chips). Unfortunately I couldn't find anything about them on the WWW, neither could I read out their SPD EEPROM for their timings from the PWS nor from a DL380 G2 and I think I don't have any other systems that can use those DIMMs, except maybe my Blade 100. But I doubt I can read the EEPROM there with Linux. So yeah, those are buffered ECC PC133 modules with 256 MiB each and they worked for me.

In addition I did pull two from each of the following buffered PC133 256 MiB DIMMs out of another DL380 G2 and all of them worked, too, i.e. same as above: kernel boot OK and operation OK with all of them installed (first pair in DIMM slots J1, J2 (bottom ones, see the DIGITAL Personal Workstation 433a Upgrade Guide (EK–ALMIA–KU. A01) pages 14 and 15 for details) and so forth):



The top and bottom ones have a HP/COMPAQ part number 127005-031 and are:

* top: hynix HYM71V32C735HCT4-H PC133R-333-542
* bottom: Micron MT18LSDT3272G-133E1 PC133R-333-542-B2

The middle one is actually also a Micron MT18LSDT3272G-133E1, though with PC133R-333-542-B1 instead. It has a HP label attached with (specification?) D8266A.

Here's what you see on the system console and in SRM when all of these are installed:

Code:

ff.fe.fd.fc.fb.fa.f9.f8.f7.f6.f5.CPU 0 speed is 2.00 ns (500MHz)

ef.ee.ed.ec.eb.f4.ea.e9.e8.e7.e6.
ewa0: link up : Negotiated  100BaseTX: full duplex
e5.e4.
Digital Personal WorkStation 500au
Console V7.2-1 Mar  6 2000 14:47:02
>>>
>>>show config

Firmware
SRM Console:    V7.2-1          Mar  6 2000 14:47:02
ARC Console:    5.70
PALcode:        OpenVMS PALcode V1.20-16, Tru64 UNIX PALcode V1.22-18
SROM Version:   v5.90

Processor
DECchip (tm) 21164A-2   Pass   500 MHz  96 KBytes SCache
2 MB BCache
PYXIS ASIC Pass 257

MEMORY


Memory Size = 1536Mb

Bank      Size/Sets   Base Addr
------    ----------  ---------
   0        512Mb      00000000
   1        512Mb      20000000
   2        512Mb      40000000


BCache Size = 2Mb

Tested Memory =  1536Mbytes
[...]

NICE! I think I'll continue to use these modules for my PWS 500au. 

I also found another compatible DIMM type when disassembling one of my IBM B50s. Those are buffered PC100 256 MiB DIMMs from IBM which work the same way as the six modules above and also together with four of those six modules above. They are stacked DIMMs, so actually have twice the amount of SDRAM chips on both sides:



IBM FRU number is 01K1142 labelled PC100 322-622R

As written above I didn't use all or most of the memory when testing with 7z. But I might give that a try when I find the time.

Apart from that all these DIMMs worked much more stable than anything else I tested, except for the non-buffered ECC modules that did equally well and which were installed prior to these ones.

---

So to sum that up, the following buffered ECC DIMMs work in the PWS (500au):

• CENTON CCB256M/REG133 (256 MiB/PC133)
  
• HP/COMPAQ 127005-031 (256 MiB/PC133) (see next post)
  
• HP D8266A (256 MiB/PC133) (see next post)
  
• IBM 01K1142 (256 MiB/PC100)
  

[johnnym]

So, I had another go at this and got new information:

Though the six 256 MiB PC133 registered/buffered ECC modules (two of each of the Compaq/HP related modules from the last post) I had installed in my PWS 500au work through POST, OS boot and "light" testing (i.e. 7za b -md=2m), they unfortunately seem to fail when using more memory (i.e. bigger dictionaries) during 7za b runs. When this happens, the system freezes and becomes unresponsive.

This also happens when testing them separately (meaning two identical modules) in SRM with the memtest tool (see the SRM console reference, page 97 and following for details and usage), which hence might be a good and quick way to determine any "good" modules for a PWS.

So these three (or two when considering the part number or specification) module types are no good for the PWS.

The IBM and CENTON ones though do work in SRM's memtest and also work during 7za b runs unless you hit a situation where the Linux kernel (6.3.x) is not able to handle a paging request.

Testing in SRM could be done like that (example from testing the Centon modules):

Code:

>>>show mem


Memory Size = 512Mb

Bank      Size/Sets   Base Addr
------    ----------  ---------
   0        512Mb      00000000


BCache Size = 2Mb

Tested Memory =  512Mbytes

>>>date; memtest -bs 100000 -rb -p 1 && echo "OK"; date
10:58:11  June 23, 2023
OK
10:58:38  June 23, 2023

>>>date; memtest -bs 100000 -rb -p 10 && echo "OK"; date
10:59:34  June 23, 2023
OK
11:02:01  June 23, 2023

First check that all memory was detected, then do a single pass test which should take less than 30 seconds, then run a multipass test.

So I'm now down to 1 GiB at the moment using the two IBM (PC100) and the two CENTON (PC133) registered/buffered ECC modules.

I also tested the unbuffered ECC modules from my earlier post in SRM and can confirm that the machine survives the testing with them.

[MrWeedster]

Hey Johnnym.
I forgot about this topic at all, your message in discord reminded me i had to do something
Found cheap Micron MT18LSDT3272G-133E1 PC133R-333-542-B2 on ebay and bought a pair. Sadly no luck, the machine recognizes them but freezes at memory test/boot/memtest in srm.
I guess i stay at 128MB till i find cheap modules eventually

[johnnym]

Found cheap Micron MT18LSDT3272G-133E1 PC133R-333-542-B2 on ebay and bought a pair. Sadly no luck, the machine recognizes them but freezes at memory test/boot/memtest in srm.
I guess i stay at 128MB till i find cheap modules eventually

So are you actually running ECC only SDRAM DIMMs so far?

[MrWeedster]

i guess so, yes