Booting Windows NT 4 on a DEC Multia | Hacker News Reader

Booting Windows NT 4 on a DEC Multia

blog.pizzabox.computer

160 pointsfanf258 comments

In ~1997 I bought a Multia off some online auction site pre-Ebay. My local PC supplier had to special order 64MB ECC RAM for me as they didn't carry it. For a teenager at my first IT job it took me a few months to afford the RAM and was a bit before I had it running. I mainly used it beta testing Redhat Rawhide and experimenting with Linux. It was my first non-x86 system.

I did try WindowsNT 4.0 on it for a week, but it was slow and almost nothing was compatible with it.

Later I got some Alphastations but they were almost the same spec as a Multia, just better form factor.

My main PC around this time was a Sun Sparcstation 20 with a ROSS Dual Hypersparc module and 2GB RAM. I remember paying around $2000 for it in 1998, probably too much! Unfortunately the magic smoke escaped from the CPU one morning in 2000 :(

Not bad for the SPARC considering it was introduced at around $25,000. Of course, by 1998 it was only as powerful as a dual processor Pentium Pro. I think the RAM maxed out at 512MB though.

I did my thesis (and wrote code on) a sparcstation lx running an (very) early version of openbsd/sparc.

I'd often sit down and write papers at a SPARC IPX in the computer lab because nobody else knew what they were doing. It was a pretty slow computer, but the windows PCs were all on Win 3.1 at the time and that was pretty annoying. You had to boot the PC from scratch each time, while the IPX was ready to go. Anyone else remember Pitt's computing evironment around 1996?

This isn't SPARC, it's Alpha.

Holy crap that's a lot of RAM for 1998.

I hit 2GB about 10 years later when I got a Core2Duo workstation.

I think OP is mistaken, 512 MB was max in a sparcstation 20. Unless they were working at LLNL or Brookhaven Labs on something very special affording 2GB of ram in a 1997 desktop workstation would have been nearly impossible to justify financially. And probably wouldn't have been a sun, it would have been a $35,000 SGI. Such as a Onyx or Onyx2, or Challenge. Something r4400 based.

Yes, my old (but at the time still usable) windows box had 2 GB hard drive and 8 MB RAM in 1997.

I think you're right, 97 is too early for an affordable computer with 2GB. But it wasn't just SGI that had those capabilities, I bought a used HP Visualize J2240 (manufactured circa 1997-8) in the early 2000s (like 2003) and it came to me with 2GB of RAM, with a max of 4GB. This machine was also very expensive when new, in the same range as the $35K SGI you refer to.

I worked on a project in 1993 for which we bought a SparcStation 20 with 1 GB of RAM, but we used it with 2 X-Terminals to support 3 developers. At the time we thought 1 GB of RAM was crazy! I believe the machine cost us $10K. So, I can imagine 5 years later a SS 20 with 2 GB of RAM would be considered obsolete tech, but I think $1000 would not be a bad price for it. (I see Wikipedia says the SS20 maxed out at 512 MB, but I distinctly remember having 1 GB because it seemed like such an astronomical number at the time, considering my first computer had 16K.)

"My understanding of this procedure is that I would mount the hard drive on another computer and match the Administrator password hash to a known password"

For future reference, all you need to do is temporarily replace services.exe with an exe that runs "net user administrator xyz123" (to change the password to xyz123), then reboot. Then put the original services.exe back, of course.

... which would require getting your hands on a toolchain able to compile for this decades-old Windows on a pretty rare architecture.

Not really. This is pre-ASLR, so all you need to do is push a string pointer on the stack and call an address. Probably less than 15 bytes of machine code you'd need to hex-edit into an existing exe.

If you have hard drive access you can replace the logon.scr screensaver executable with the command prompt executable, then a few minutes after boot up a command prompt with system privileges will appear by magic.

Like gcc[0] or Visual C++[1]?

[0]: https://virtuallyfun.com/2016/12/09/gcc-windows-nt-dec-alpha...

[1]: https://winworldpc.com/product/visual-c/6x

It's also possible that the Linux chntpw utility might be able to change the password in the old registry file directly.

I believe the format of either the registry hives or the the way the password hash is stored changed between Win2K and WinXP, so a modern tool likely wouldn't work. That's why you can use a Windows 2000 CD to boot to a recovery console in later windows versions without a password: It thinks the registry is corrupt and skips authentication.

chntpw works on all versions from NT 3.1 through Windows 10. The hive format, nor password storage format, actually hasn't changed in all the time. :-)

Mind, the trickest part of dealing with pre-4.0 is mounting the file system, assuming the OS was placed onto NTFS rather than FAT. ntfs-3g only likes the "modern" NTFS format, which is found in NT 4.0 and up.

AFAIK the SAM format changed a few times during the life of NT 4.0, including when SYSKEY was introduced.

Mind, the trickest part of dealing with pre-4.0 is mounting the file system, assuming the OS was placed onto NTFS rather than FAT.

A couple of times I've tried mounting an NT 3.51 disk in a modern Windows 7 system and while it works (you can read from and write to it fine,) once you even mount it, NT 3.51 will never boot from that disk again. It hangs during the blue NTLDR stage.

I'm not sure what would ever possess anyone to think it a good idea to write to a filesystem immediately upon mounting it, but Windows 7 does something to old NTFS volumes which renders them incompatible with NT 3.x.

Most probably the filesystem is updated (silently), like it happened historically.

In the good ol' times there were free demo CD's of Windows 2000 (90 or 120 days trial), everyone (that was running NT 4.0) tried them.

Those with NT 4.0 already up to SP 4 had no issues (but chkdsk stopped working) all the ones with earlier service packs had unbootable machine (if the NT 4.00 was on NTFS).

At the time there were a couple "workarounds" for the CHKDSK issue, one by Sysinternals/Mark Russinovich NTFSCHK and one, Mark4NTFS by M. Tartsch.

Just in case:

https://msfn.org/board/topic/169500-chkdsk-refuses-to-check-...

But no idea if the mark4NTFS can help for NT 3.x.

NT 4 prepared Win32 for synchronization with Windows 95. I hated it for that: they moved the presentation layer into the kernel (to allow features like "Active Desktop" widgets) which negatively impacted reliability and security for years to come.

Windows NT 3.1 and 3.5 were much more reliable. The CD-ROM's containing the OS from MSDN came with the Microsoft C/C++ compiler on it, somewhere they even added Visual Studio 1.0.

There was hardly any client software for MIPS and Alpha, but the i386 versions ran 16 bit versions of Ami Pro, Excel and Mosaic well. Which is probably why MIPS and Alpha NT versions never became popular, despite the fact that they were very fast for the time.

win32k has nothing to do with active desktop which was mostly shell level.

Windows 95 and NT 4 didn't even ship with Active Desktop-- it (along with IE integration into Windows Explorer) was added as part of the Windows Desktop Update that shipped with IE4, and didn't become part of the OS out of the box until Windows 98 (and 2000 on the NT series).

GDI in the kernel (win32k.sys) was a performance optimization.

Windows 2000 was always my favourite release - most of the reliability of NT4, but with the added ability to use USB.

Alpha NT didn't become popular because of price and compatibility. DEC wanted to sell Alpha in serious workstations and servers and only made some half-hearted attempts to target the consumer market. Meanwhile the consumer marker coalesced around Wintel and Intel could leverage that larger volume to reduce costs, while consumers were locked in to x86 for compatibility.

Dunno much about the history of MIPS but I imagine it was a similar story. Once Windows on x86 hit critical mass it steamrolled the rest. PowerPC was the exception with Apple, but even they didn't really start biting into bigger market shares until they switched. The dominance wasn't technically motivated (I know at least Alpha and SPARC had superior performance in many areas), it was economics and compatibility (and some shady dealings).

All the RISC architectures had superior performance over x86-32 in the '90s.

MIPS is much larger than SGI, and POWER [1] is much larger than PPC (or Apple). MIPS and POWER have been used in consoles, for example. That is volume! Especially MIPS, is used in embedded chips (together with predominantly ARM). Those are found on every mainboard, laptop, console even though the latest gen consoles run on AMD64.

Alpha however, died somewhere within DEC/Compaq/HP.

My friend argued the NSA was using Alpha processors internally, and had their own fab. I have no clue about that though.

[1] https://en.wikipedia.org/wiki/Power_Architecture

NT 3.x was mainly pushed as a server OS targeting Novell Netware. On the PC side it never made sense unless you bought one of these discount surplus Multias.

I worked at a F500 and two times DEC reps came into sell NT/Alpha, and both occasions, they couldn't get the machines to boot. I wonder how much of that NT/Alpha just for marketing, because DEC was selling a lot of x86 servers in those days.

The Pentium Pro 200Mhz was better than or close-enough to RISC processors in most standard benchmarks that most businesses seem to decide "good enough", and that was the beginning of the end of the end of the server processor wars.

NT 3.x was fine for a desktop OS. I ran both on Pentium machines as my development system.

Whatever point I thought I was trying to make came out the completely wrong way, so for anyone looking at my post history, I was drunk and disavow this comment.

I had a pet Pentium Pro server in about 1999, and it was a fantastic machine. Excellent CPU performance and a fast IO system to match.

they moved the presentation layer into the kernel

That was a nightmare, I remember opening a printer queue on a server at one customer and it instantly blue screened the machine. I'd only been on site for five minutes before I had to go and apologise to the customer.

> Finding software written for Win32/Alpha, though, is likely to be a challenge.

As someone who actually had one of these at work[1] back in the early 90s, there was hardly any software compiled for Alpha even when it was a current product. Lack of software support, was one of the reasons that WinNT Alpha never gained a foothold. Which was a shame as the Alpha hardware was pretty great.

[1] It was the only one we had, I think we got it as a freebie as the rest of the computer room was choca full of DEC VAXs.

I remember a local college loved them because you could teach 3d modeling. They really weren't used for the actual modelling but rather as cheap render farms that could run at night.

there was hardly any software compiled for Alpha even when it was a current product

I ran SQL Server on NT on Alpha in the 90s, it was actually very good, but you had to treat it as a dedicated appliance rather than a general-purpose computer because as you say, there was very little software available.

Then the industry took a weird meander through DEC StrongARM and now Microsoft is looking at Windows on ARM seriously again...

If I remember correctly there was something called FX32 that was like a dynamic transpiler from X86 to Alpha. You could run Win32 apps, and they would get faster over time as it cached translated binaries.

https://en.wikipedia.org/wiki/FX!32

I remember seeing NT3.5 running on an Alpha at a trade show around the same time period. A higher end model than the one the author posts here. The reason why it never found "consumer success" is that in late 1993 dollars it cost $7800 , equivalent to about $13,600 today. When at the same time it was possible to buy a 486DX/33 for just under $2000 in a pretty decent, though not nearly as capable configuration. That is what consumers could afford.

I remember playing with a MSDN copy of Win NT 3.51 in my cousin's cubical back at AMI in the late 90s. We had NT4 on a Pentium Pro I got to admin back in high school. I even tried to see if I could get the library computers to login off a Samba PDC instead of the NT4 Server. A lot of stuff didn't work, and group policies were pretty janky.

I like how the author also has an SGI and two Sun machines next to this one. We got a hold of some old Sun Stations in University back in 2001; paid like $12 for a crate of them. They were made in the 80s and had dual 100MB SCSI hard drives, tons of ram slots (I want to say 32 or 64MB of potential ram), and those crazy optical mice that needed the grid based mouse pads. I think I got RedHat running on mine and didn't do much with it.

I'd love to collect and work on old stuff like this, but I started to do some hard core backpacking in my 30s and my parents threw out nearly everything while I was over seas for a few years. I'd really like to stay minimal with a room full or less of stuff, so I don't see myself getting into old computing any time soon. I'll just read posts like this and read The 8-Bit guy to get that nostalgia fix.

Oh and if you're ever in Seattle and love old machines, check out The Living Computer Museum. They have actual Altos, HP UX, TIs, Apples, and DEC Alpha machines you can log into and play with. It's pretty amazing.

The Living Computer Museum is indeed fantastic. I’m partial to the mainframes but my daughter hates the loud noise in the cold room so we spend more time playing old PC games.

Group policies were introduced with Windows 2000 and Active Directory.

Sorry, I meant System Policies, the predecessors to Group Policies.

The article mentions Service Pack 6. You needed to install 6a which fixed up many faults of 6, like e.g. not being able to disable Nagle's algorithm for TCP (not funny when you wanted to send around short TCP messages).

As a kid who grew up in Massachusetts in the 80s and 90s, I'd love to go back in time and experience the competing tech that was being built in my back yard before the west coast totally won out. I had friends whose parents worked for both DEC and Wang Laboratories, and had those companies' gear in their homes. At the time I only cared about what games you could put on or write for them, but it would be awesome to poke around on them again as an adult.

This brings back so many fond memories from the late 90ies. I dual booted Windows NT 4 and Red Hat Linux for many years on an old Pentium 166 with an Adaptec scsi controller (remember that?). It was wonderful. Could never afford an Alpha machine though. WinNT was good stuff, miss the good old blue boot screen.

Yeah I remember having a server with an Adaptec SCSI controller. It only booted with Linux 2.4. When we attempted to boot Linux 2.6 I had to drive 300 km to the datacenter to fix it. "trying harder", it printed. Fun stuff.

I ran NT4 on a MIPS Magnum R4000 for a while. It worked okay. Could run 16bit x86 Windows/DOS stuff in a weird sort of translation/emulation.

All my Multias (I owned 2-3 over the years) usually ended up with RHEL on them; nobody in the nerd crowd I was in wanted to run Windows on them. They ran so hot that if you sat it horizontal it would keep your coffee cup warm if close enough.

I believe someone got VMS running on one after some hackery with the firmware/BIOS.

I bought an Acer PICA (https://en.wikipedia.org/wiki/Acer_PICA) back in the early 2000s, obsolete by then, just to play with it and Windows NT 4. MIPS still seemed exotic, and there was little free software to use on the platform. I think a port of Office existed, and maybe an early Internet Explorer.

I'm not sure what the point of the ARC ports of NT even were, except Microsoft's attempt to keep its fingers in as many pies as possible and crowd out competitors.

They weren't sure if x86 was going to be competitive in the long run at that point, so they were porting to all the major architectures. I even heard rumors of a SPARC port of NT 3.51 but I've never seen it in the flesh.

Solaris even ported to PPC (specific RS6K boxes) for Solaris 2.5.1; I've got a copy of that ISO image around somewhere but never had the right hardware to run it.

I had an Alpha Multia back in ‘98. I remember having to scour eBay for the appropriate SCSI adapters since DEC used something proprietary. And even with the adapter, I never got the Linuxes of the time to recognize the HDs.

The key to running NT4 for Alpha was DEC’s FX!32. It crosscompiled x86 code to AXP. Thanks to Alpha’s better design this made my 233mhz 21066 a lot faster running x86 code the my Pentium MMX 233.

I realized the only other time I'd heard of the Multia was from Slashdot posts from nearly 20 years ago.

https://tech.slashdot.org/story/99/08/02/0219235/free-multia...

PSA: don't click the link to the linux store in the Slashdot article. Someone super dodgy owns that domain name now.

I had an alpha (miata, AKA Personal Workstation 500au, I think) running NT on my desk for a while, while we waited for space to open up in a rack..(1) The NT on it seemed quite usable, thanks to the FX!32 emulator. When running x86 compiled stuff, it felt mostly as fast as contemporary high-end PCs, once you'd opened the x86 program and used it for a while.

(1) We used FreeBSD/alpha, so we always bought the "NT" version of machines to save $$$ on the DEC UNIX license fee. Eg, just like today, the Windows version of a machine was cheaper.. FreeBSD did not support the ARC firmware needed by NT, but you could easily switch the machines to SRM in order to load FreeBSD.

I miss the simplicity of the NT / 2003 era of servers. You could make setup as simple or as complex as you now. Now you have to have a BSA to configure so many arbitrary standards just to connect to a NAS.

I knew someone who had a DEC running Windows NT around 2000. They admitted it didn't work as well as they would have liked.

I have Windows 2000 RC2 running on an Alpha :) It's really neat, you can even execute x86 programs.

So at the time, to compile software for Windows NT Alpha, was there an Alpha IDE and compiler available (something like VC++) such that developers were writing and debugging locally, or would developers usually cross-compile from an x86 machine?

To answer my own question, now that I'm not on my mobile: it looks like there was an Alpha version of Visual C++ 2.0, as well as a cross-compiler for the 68k Macintosh! [1] I was firmly in VB land at the time (as a ~10 year old) so this bit of history eluded me.

[1] https://accu.org/index.php/journals/1771

I messed around with one of these I got in college at an off lease / salvage type place... The boot loader on those was really cool and because it was an alpha, you had a license for digital UNIX...

Reply on news.ycombinator.com