How picky is NT 4.0 with CPUs?

[bakemono]

You mentioned it was a SCSI disk. If you don't also have the SCSI card that it was originally connected to, then the most likely result of trying to boot it is an "inaccessible boot device" BSOD when it can't find the card. If you do have the card, then I'd say ignore the naysayers and try it, you might get lucky.

 

[hunterjwizzard]

It is physically *possible* that I have the original card. I have at least one PCI 50 pin SCSI card who's origins I cannot trace. Its plausible it originally came out of that machine and then survived a lengthy series of hardware purges.

But I am not entirely sure I want to role those dice.

 

[GiGaBiTe]

Not to mention I saw NT 4.0 machines being run on Hospital equipment as late as 2014. Because it just worked.

Hospital equipment running ancient operating systems is less because "it just works" and more because the system is attached to a multi-million dollar machine where the control software was only ever validated to run on a very specific hardware kit.

The hospital I worked at in 2019 had the MRI machine running on an SGI Octane2. The radiotherapy unit a few rooms over used a Windows 95 box with all sorts of specialized addon cards.

To replace the SGI Octane2 and the Windows 95 box would also require replacing the multi million dollar machines they're controlling. Nobody is going to touch those systems and own it if someone ends up being harmed or killed by them from an unforeseen malfunction caused by messing with stuff that doesn't want to be messed with.

A prime example is the Therac-25, where it was directly based on an older Therac unit, but had bits of it changed in software and had some hardware interlocks removed. Due to extremely poor programming by arrogant engineers/programmers and the removal of the hardware interlocks, 6 people either died or were significantly harmed from software errors and manufacturer incompetence causing operator errors that lead to the events. The software bugs had been reported by operators, but were ignored.

Such was the incompetence of the programmers, they had a single error that stated "error - a radiation dose too high or too low was administered".

 

[Eudimorphodon]

The hospital I worked at in 2019 had the MRI machine running on an SGI Octane2. The radiotherapy unit a few rooms over used a Windows 95 box with all sorts of specialized addon cards.

A friend of mine that was doing consulting work in hospitals about a decade ago had to go on a scavenger hunt to repair a DOS based system. Stuff in these kind of environments lasts *forever* because, yeah, the systems they're a part of were certified as a whole and you can't just start upgrading things piecemeal without having the whole thing re-certified.

That a thing that happens to powered by NT that was built to do a very specific thing in a controlled environment is still capable of doing said thing 15 years later isn't exactly impressive when you think about it. Computers are *supposed* to be reliable and produce repeatable results if you don't change any variables. Consumer PCs are the utter trainwreck they are because "doing their job" is such a rapidly moving target.

 

[VERAULT]

Thats just the NT4.0 setup. There were ancient MS-DOS controller pc's inside of lots of things as well.

 

[GiGaBiTe]

There are still bump bars for kitchens that run DOS on DM&P Vortex86 CPUs.

 

[hunterjwizzard]

Nobody is going to touch those systems and own it if someone ends up being harmed or killed by them from an unforeseen malfunction caused by messing with stuff that doesn't want to be messed with.

Actually people do touch them. On a regular basis, in fact. Thats how those million dollar machines stay running for so long: regular maintenance. Its a highly specialized breed of technicians who service those million dollar machines and the PCs that run them.

This is also how I got a call one evening for a floppy drive in 2019. What funny is the guy actually knew very little about "computers" in general. He made six figures servicing medical tech and control systems, but he knew how to read a service guide and replace a part thats out of spec. The computers themselves are typically air gapped and do nothing but run that specialized software. The control computers are worth tens of thousands of dollars despite relatively modest specs because the manufacturer has had to keep a supply of validated replacement parts on hand for 30+ years.

Really fascinating world. Or really boring, depending on who you ask.

 

[GiGaBiTe]

Actually people do touch them. On a regular basis, in fact. Thats how those million dollar machines stay running for so long: regular maintenance. Its a highly specialized breed of technicians who service those million dollar machines and the PCs that run them.

You're quoting me out of context.

And there's nothing specialized about maintaining old hardware, those companies are just legalized extortion, and one of the reasons healthcare costs are spiraling out of control. Like military contractors, they'll charge $1000 to even show up, another thousands in diagnostic fees and $10,000 to change a hard drive.

I have customers that I maintain old hardware for and they appreciate the fact I don't rake them over the coals like those other companies do.

 

[Zare]

PCs aren't actually running these machines, I don't know what exact machines we're talking about, but I presume medical/industrial. Usually, 99.9% certainty, they aren't controlled by that PC. The PC is an UI and MCU orchestrator. Microcontrollers do the work. Or else you'd be writing real-time software for Windows 9x/NT which is nearly impossible.

A lot of machines used QNX for orchestrator PC because they even wanted lowest latency possible in UI response.

To remove the old PC from machine one could make an implementation against the MCU if it has proper specification.
However good luck with hiring people to write software per specification in today's "development market". Average programmer hasn't seen a proper spec in his career.

As much as I despise NT I would never argue that the admins that acquired the necessary skills to run these giant NT deployments didn't earn their money. And NT *was* a stable and reliable OS when it was installed properly on reliable hardware; if you were setting up a computer to be a server or appliance it did the job. (We had an NT-based telephone system at the startup, a Pentium II-grade PC in a rack-mount case with like 20 ISA slots holding analog PBX cards, and, sure, it worked. But we didn't set it up, it was an appliance.) But it was a *terrible* OS to troubleshoot if anything went wrong and could be *remarkably* fragile if an administrator just took a single wrong step. Most elite Windows NT techpriests I've met have basically admitted that the solution to most critical failures was to wipe it and start over.

I've brought many Linux and BSD back online from their death.
There is not a single Microsoft OS that's easy to troubleshoot. Consider back then you could log all the kernel drivers being loaded to a file, but you couldn't view any file if OS won't come up. Live Linux CD when it came out, was the biggest tool to debug non-bootable Windows installations ever.

This is a horrendous choice by Microsoft.

 

[hunterjwizzard]

You're quoting me out of context.

Oops, sorry.

And there's nothing specialized about maintaining old hardware, those companies are just legalized extortion,

But those technicians I'm talking about don't service JUST the old PCs in there. They are there to fix the entire machine. The PC is just one component of it(and arguably the one the tech knows the least about).

 

[Eudimorphodon]

There is not a single Microsoft OS that's easy to troubleshoot.

There is a reason I used the word “Techpriest” when describing Windows admins. Don’t get me wrong, plenty of them are smart people, but let’s face it, the thing they have to maintain is so obscure and unknowable they are pretty much members of the Adeptus Mechanicus, applying the essential oils and following the ancient complex rituals in hope that the mysterious mechanism will work another day.

I can’t think of a single problem I’ve had in years that I couldn’t fix with Linux or xBSD, or at least precisely nail down what the root cause is, with a few minutes of determined Googling and picking through log messages. Windows, well… if you’re extremely lucky you might find an answer that’s *not* cross your fingers and reinstall it (IE, a registry edit or something) 19 pages deep inside a Dell discussion board apparently written by the God-Emperor himself as proof to the faithful that his spirit still lives inside his rotting corpse for those who truly seek him. You sure as hell won’t actually get help from Microsoft.

 

[rmay635703]

I remember NT4 refusing to install on NexGen cpus. It detected them as 386. Worked fine if you installed on a different machine and moved the hard drive over to the NexGen system and ran wonderfully.

Seems like the original 6x86 wouldn’t work either.

 

[T-R-A]

Seems like the original 6x86 wouldn’t work either.

Nope. I had it running pretty decently on a 6x86 in the mid 90s. The only real PITA issue I had was getting Dial-up networking to work. Installing the "then-current" Internet Explorer (4.0) fixed that for whatever reason. I was a Netscape fan and never used IE.

 

[Chuck(G)]

I've still got it (NT 4.0 server) installed on a 6x86 system. Resides on a CF card.

 

[Robbbert]

I've got half a dozen NT 4.0 installs, but they are almost all for the same CPU (500MHz). There's 2 meant for a 450MHz machine, and if accidently plugged into the 500MHz machine they won't boot. So, the CPU is important.

No Device Manager, so installing hardware is a right royal pain if you don't know exactly what's in the machine.

No FAT32 support, but you can install a little something that adds it.

No USB-stick support, but again there's drivers available.

By default there's no scroll mouse support, but once again there's drivers around.

I've installed the FAT32, USB, and scroll-mouse support on all my installations with complete success.

As others have said, if everything is done properly it is a stable OS, ideal for servers in unattended situations that need to run for months or years continuously.

NT was the first Microsoft OS to support a domain, but it was rather lacking when compared to Active Directory (first appeared with W2K) .

 

[Zare]

If NT weren't Microsoft it would've gotten nowhere. First release of NT had more impact in business sense than technological, it was crap compared to Unix, but the fact that Microsoft is serious about going enterprise was a clear indication of what was about to happen. They'd slowly use inferior technology and proprietary stack to ensure market growth because they had already started to conquer an adjacent market of personal computers.

As others have said, if everything is done properly it is a stable OS, ideal for servers in unattended situations that need to run for months or years continuously.

I highly disagree with that last part - "unattended situations that need to run for months or years continuously", because that's the basic definition of a server.
If there wasn't a single hardware platform a server OS can run on for years, it would've been a defective product to start with.

NT4 runs on way less hardware platforms than even Linux or FreeBSD/NetBSD of the day. It is also tougher to configure and maintain, it is absolutely a nightmare if you want to change the base hardware like mainboard to another model. It is also slower than them.

There is nothing ideal about NT4, it's not a good server product, but one that's capable of running Microsoft ecosystem services. And that's the only thing that matters.

Microsoft used FreeBSD for a great amount of their infrastructure especially internet-exposed one, not NT 3.x or 4.

 

[Svenska]

PCs aren't actually running these machines, I don't know what exact machines we're talking about, but I presume medical/industrial. Usually, 99.9% certainty, they aren't controlled by that PC. The PC is an UI and MCU orchestrator. Microcontrollers do the work. Or else you'd be writing real-time software for Windows 9x/NT which is nearly impossible.

Which is exactly what people have done.

I know of a dual-CPU system which told NTLDR that the machine only has one CPU. A custom NT kernel driver then brought up the second CPU and ran a bespoke PLC real-time emulation for real-time control of semiconductor equipment. Some oscilloscope vendors are known to have used very special graphics drivers on their NT-based offerings to allow the FPGA circuitry to directly draw into the framebuffer and bypass the OS latency.

Both solutions do not survive OS upgrades, but they combine hard real-time guarantees with the freedom of using cheap PC-style hardware, existing file systems and development knowledge and tools.

 

[njroadfan]

I've run NT 4.0 without issue on 65nm Core2Duo machines. I followed this fairly complex guide to build the install CD: https://bearwindows.zcm.com.au/winnt4.htm#3

Actually worked quite well considering the machine lacked PS/2 ports. I installed that USB support driver and it supported keyboards and mice without issue.

 

[Eudimorphodon]

If there wasn't a single hardware platform a server OS can run on for years, it would've been a defective product to start with.

I’m sure plenty of the older folks here remember hearing the stories of, say, Novell Netware servers that ran for 16 years straight and were only shut off because the whine from their worn-out disk bearings became intolerable. (Or the even juicier ones about insert-vendor-here servers that were so quietly reliable nobody noticed for years when the closet they were in got walled off.) There is indeed nothing particularly impressive about a computer sitting in one place running the same software for years, particularly if it’s just something like serving files.

 

[czunit]

Fun to read all of this NT history. I remember when I came to the IEEE Computer Society in 1994 they had Windows 3.11 and on their workstations and Novell servers/Sparc 10/20 systems.

Windows 3 was a nightmare to install properly, and getting a system that was *reliable* required a massive amount of custom ini files, weird drivers, and the like. It was a total mess: You can install it of course but getting it to run day after day without crashing was a serious balancing act.

We also had a strange donation: An NCR 3500 system which was a big cube box about 5 feet tall that held a lot of 1gb SCSI drives and *4* 486/50 CPUs with I think 32mb of memory. I played with this system, it ran NCR's brain dead System V Unix and was doing nothing (NCR gave it to us) so I tried loading NT 3.5 on it.

What an amazing OS! Unlike Windows it ran overnight, over the weekend, it just RAN. So we used it as a spare file server and to do things like CCMail mailbox compactions, where I could run several jobs at once on the 4 CPU system. Note it did have a special HAL that I had to install for the processors, and it had another one for its system plane and the MicroChannel bus.

When Windows 4.0 preview came out in 1995 I was blown away. It was solid, reliable, AND had the newer interface. I immediately flattened the 3500, loaded it up, and it worked perfectly. All 4 CPUs worked, all memory recognized, and we could now use the system to do stuff.

That "Stuff" came quickly. Anderson was still with UIUC and he had developed this thing called "Netscape Commerce Server" which had SHTTP and HTTPS ability. I took one look at that pre-release code, and we loaded it on the 3500 so we could put SuperComputing 95's registration up on the Web and ACCEPT CREDIT CARDS FOR REGISTRATION!

After getting that all ready I was writing the PERL code to handle the page submission and realized *I* would be the one who would have to type the damn cards into ICVerify. And I looked at the DOS version of that credit card processing software. And looked at the 3500 with its perfect DOS emulation and 2 serial ports. And back at the stone age ICVerify software....

And got to work building a queue interface so the web site could write the card to a temp file, and another PERL script would take the file and jam it into ICVerify, then read out the result and put it back into the queue for the Web server to acknowedge payment.

BANG! I created one of the first secure E-Commerce systems on the web. Complete with HTTPS:, a software based WAF (the queuing system did all the checks to make sure someone wasn't trying to hack in a refund) and OS controlled isolation of the ICVerify from the Web server.

The rest.... was history. We ran SuperComputing95, then other conferences and everyone LOVED it. Then we did $99 renewals online, and expanded the NCR to 256mb of memory. Then I re-wrote the IEEE CSLSP (Computer Society Library Subscription Program) which at the time came with the stupid IEEE LSP which consisted of a huge disk changer of 100 or so CD ROMs and a big clunky computer for colleges. We took the 10 CD's that had the Computer Society content, wrote an SGML-HTML converter, used DYNAWEB to convert Tek Math to GIFs (in real time) and had a system that could now serve the whole library system via the internet from our office to Colleges without a massive chunk of junky pioneer hardware...

This required the 3500 to get some.... disks.... But we had space in the box so we built several RAID arrays of 4.3gb SCSI disks and placed a massive order with NCR for their 3550 upgrade.

Which consisted of a SECOND Microchannel bus (which we split the disk controllers onto as the 3500 had incredible internal bandwidth), 256mb memory upgrade with L3 cache, and most importantly we replaced the 4 486/50 with a total of eight Pentium Pro/200 systems on 2 processor boards. Each board had two sets of 2 PPros with their level 1 cache and a separate 1mb L2 cache that went to one of the two memory lanes to the memory boards. Each MicroChannel likewise would connect to one of the two memory lanes. So we had 8 CPUs (later upgraded to 16) that allowed people to hit the web server, request an article, and one of the Pentium Pros would decode the SGML, another one or two would render the TEKMath into HTML in real time all in parallel. And since there were dual MCA's one could handle one periodical while another handled a different one at the same time.

As load increased I could watch the system bring more of the processors into bear and watch as the Microchannel busses balanced out the load. We kept Windows NT on that system all the way till I left in 2000, with 16 Pentium Pro CPUs (upgraded to the ODP 333mhz chips which did work as the system treated the Pentiums as pairs of CPUs) and it darn near never went down due to crashes.

I wonder what happened to that NCR3550. It was a helll of a workhorse and was the source of much of what we call E-Commerce and Digital Libraries......