In early 1996 I purchased a DEC Multia, a 233 MHz Alpha 21066 box withWindows NT. There was also an older 166 MHz model. They also offered the166 MHz model with a bigger disk as an Easy Web Server.
DEC offered the 166 MHz model specifically for use with Linux(and without Windows NT) under the name UDB(Universal Desktop Box). The pricing was quite attractive.
The UDB is no longer being sold by Digital. However, there are a numberof distributors still selling the remaining inventory.
The UDB only comes with a few sheets of drool-proof paper showing howto plug in the cables. More technical information is available in the 116-pagemanual "Multia MultiClient Desktop Service Information", DECpart number EK-MULTS-IN.C01. This manual is fairly pricey (about $85 inthe US). Fortunately it is now available in machine-readable form:
NOTE: these have identical contents. Please don't download morethan one. It would just be a waste of your time.
DEC has finally made the firmware upgrade and failsafe loader disk imagesavailable on their FTPserver.
If you are not having problems with your UDB, I don't recommend attemptingto install the firmware upgrade. Most UDBs seem to have been shipped withthe most recent firmware, and there appears to be very little differencebetween the old and the new firmware. Some people have reported problemsinstalling the upgrade.
The UDB requires memory to be installed as pairs of "by 36"SIMMs, also known as "parity" SIMMs. It does not use EDO, andI'm not sure whether EDO SIMMs would even work. I've been unable to locateany "by 36" EDO SIMMs in any case. Note that "logic parity"SIMMs, which do not actually store 36 bits per word, will not work.
Some configurations of the VX41 and VX42 models are supplied with unusual12 megabyte SIMMs. These SIMMs will only work in the first pair of SIMMsockets; if you try to use them in the other sockets, or in another system,they will only be recognized as 8 megabyte SIMMs.
The UDB doesn't actually use parity in the conventional sense. The 21066microprocessor has a 64 bit data bus, with 8 additional bits for ECC(Error Correction Code). If you are running Linux, you may from timeto time see ECC messages in the syslog, which look like this:
Nov 6 10:20:23 houhah kernel: lca: machine check (la=0xfffffc0000208490,pc=0x120011250)Nov 6 10:20:23 houhah kernel: Reason: correctable ECC error (short frame, retryable, dc_stat=0):Nov 6 10:20:23 houhah kernel: Correctable read error to b-cache occurred at address 56e9260
If these appear only rarely, they are not a cause for concern. It justmeans that the ECC is doing its job. On a system with only parity, insteadof being corrected and logged, the error would cause a system halt. Andon a system with neither parity nor ECC, the error would simply go undetectedby the hardware. It could cause a crash, or it could cause some data corruptionwhich might go unnoticed for a long time (especially if the data got writtento disk).
If you get lots of ECC correctable error messages in the syslog, orif you get uncorrectable errors, you have cause for concern. This can becaused by defective SIMMs, mis-seated SIMMs, or a defective motherboard.The first thing to try is removing and re-seating the SIMMs. If that doesn'tsolve the problem, I would suggest trying a different set of SIMMs.
The UDB was generally sold configured with either a 2.5 inch or a 3.5inch internal SCSI disk. This was usually denoted by the last digit ofthe full part number of the unit. For example, a "VX42B-F2" hasa 2.5 inch disk, while a "VX42B-F3" has a 3.5 inch disk.
Units incorporating a 3.5 inch disk have a different PCI riser cardand mounting bracket. It is not possible to install a PCI expansion cardin these units.
The UDB power supply is not capable of powering a second internal harddisk, and may not be sufficient to power even a single high performancedrive, such as a Seagate Barracuda. I strongly recommend use of an externaldrive case.
If you do manage to install a 3.5 inch drive in a unit which originallyhad a 2.5 inch drive, you'll have to remove (or at least disconnect) the2.5 inch drive.
As far as I know, there no 2.5 inch SCSI disk drives having storagecapacities substantially exceeding 1 gigabyte. The UDB does have an internalIDE interface, with a 44-pin connector for both the IDE signals and power.This connector is commonly used with 2.5 inch IDE drives for portable computers.It is only a plain IDE interface, not EIDE, so it doesn't have any fastDMA or bus-mastering modes. The necessary 44-pin IDE cable isn't very common,but apparently may be purchased for $10 plus shipping as part number FC530from
Hard Data Ltd. Telephone: (403) 456-151011060 - 166 Avenue email:harddata@bigfoot.comEdmonton, AB, Canada - T5X 1Y3
VX40 (166 MHz) | VX41 (166 MHz) | VX42 (233 MHz) | description |
---|---|---|---|
W9 | W12 | Mini console (auxiliary) | |
W8 (default) | W10 | Normal power-up sequence | |
W7 | W11 | Causes the system to attempt a floppy boot fail safe load (FSL) | |
W6 to W3 | W6 to W3 | Under SRM, these jumpers determine the resolution driven to the graphicsdisplay | |
W2 | W2 | When connected to a switch, causes a reset or halt (see J8) | |
J8 | J8 | When set 2 to 3 causes a reset, when set 1 to 2 causes a halt requestCPU interrupt | |
none | W9, W8, W7 | CPU speed (see below) |
W9 | W8 | W7 | speed (MHz) |
---|---|---|---|
1 | 1 | 1 | 66 |
0 | 1 | 1 | 100 |
1 | 0 | 1 | 133 |
0 | 0 | 1 | 166 |
1 | 1 | 0 | 200 |
0 | 1 | 0 | 233 |
1 | 0 | 0 | 266 (unsupported) |
0 | 0 | 0 | 300 (unsupported) |
Note: 1 is jumper removed, 0 is jumper installed.
W6 | W5 | W4 | W3 | Standard | Resolution | Vertical Refresh (Hz) | Pixel Clock (MHz) |
---|---|---|---|---|---|---|---|
0 | 0 | 0 | 0 | VGA | 640 x 480 | 59.94 | 25.175 |
0 | 0 | 0 | 1 | ERGO VGA | 640 x 480 | 72.79 | 31.50 |
0 | 0 | 1 | 0 | SVGA | 800 x 600 | 60.32 | 40.00 |
0 | 0 | 1 | 1 | ERGO SVGA | 800 x 600 | 72.19 | 50.00 |
0 | 1 | 0 | 0 | VESA 1024x768 | 1024 x 768 | 70.07 | 75.00 |
0 | 1 | 0 | 1 | Digital 1024x768 | 1024 x 768 | 72.033 | 74.367 |
0 | 1 | 1 | 0 | Digital 1024x864 | 1024 x 864 | 60.00 | 70.00 |
0 | 1 | 1 | 1 | 1280x1024 60Hz (S3 86C928) | 1280 x 1024 | 60.10 | 110.15 |
1 | 0 | 0 | 0 | 1280x1024 66Hz | 1280 x 1024 | 66.51 | 119.843 |
1 | 0 | 0 | 1 | 1280x1024 72Hz | 1280 x 1024 | 72.556 | 130.808 |
1 | 0 | 1 | 0 | 1280x1024 72Hz (S3 86C928) | 1280 x 1024 | 72.05 | 130.08 |
1 | 0 | 1 | 1 | 1280x512 Stereo | 1280 x 512 | 139.336 | 130.804 |
Note: 1 is jumper removed, 0 is jumper installed.
The UDB has two serial ports, but only one DB-25 connector. DEC sellsan appropriate splitter cable (P/N BC25X-1F), but it is fairly expensiveso if you are handy with a soldering iron or crimping tool you may wishto make your own.
COM1 is wired normally. COM2 is wired using the EIA-232E secondary channelpin assignments where they exist, and other pins (arbitrary?) for the rest.
Signal | COM1 pin | COM2 pin |
---|---|---|
TxD | 2 | 14 |
RxD | 3 | 16 |
RTS | 4 | 19 |
CTS | 5 | 13 |
DSR | 6 | 23 |
signal ground | 7 | |
DCD | 8 | 12 |
DTR | 20 | 11 |
RI | 22 | 25 |
I don't recommend overclocking, but many people report the 233 MHz systemrunning fine at 266 MHz. Note that the early 166 MHz parts (21066-AA) werenot capable of being run at these speeds; the faster parts resulted froma redesign in a smaller process. If you have a VX41 with a 166 MHz 21066-BB,you might be able to run it considerably faster.
Here are all the variants of the LCA that I have information on:
Chip Design | Part Number | Rated Speed | Notes |
---|---|---|---|
21066 (LCA4) | 21066-AA | 166 MHz | |
21068-AA | 100 MHz | for embedded systems | |
21066A (LCA45) | 21066-AB | 233 MHz | |
21066-BB | 166 MHz | ||
21066-CB | 100 MHz | replaces 21068 |
Mitsubishi, a licensed second-source vendor, claimed to offer 200 MHzversions of the original 21066 (LCA4).
For UDB, the Linux Kernel should be configured as a NoName board. Tomake the sound system work, choose only the Microsoft Sound System supportand the /dev/dsp and /dev/audio support. The settings should be:
I/O base | 530 |
---|---|
IRQ | 9 |
DMA | 3 |
Before you buy a UDB, be forewarned that the seemingly high clock ratedoes not mean that the machine is a screamer. For non-floating-pointtasks, the 233 MHz model seems to offer performance comparable to a 100to 120 MHz Pentium. However, the floating point performance is fairly good.
Last updated April 1, 1997
Copyright 1996, 1997 Eric Smith