Vintage computing for fun and [no] profit

[MartinHowe]

If I could spare that I'd rather invest into a new system that can replace my current 9 year old one... Hopefully later this year.

That was my original plan, kinda; I wanted to do both, one after the other, as I wasn't sure if I could afford to buy the AlphaServer and a new computer at the same time. I thought of waiting until after mainline PC upgrade and then see what AXP hardware was on the market then. But having recently done an upgrade job for a customer at work, I realised there was no way in Hell I could afford a really good modern PC right now, as graphics cards and powerful CPU prices have gone though the roof, that is when bitcoiners and scalpers haven't nicked them all in the first place! CCL Online here in the UK (our go-to supplier for computer parts), for example, have "sold out please ask us to notify" for all RTX 30?? cards and several Ryzen CPUs.

So I decided to get the AlphaServer first and try again for a new PC in a couple of years. Please don't get me wrong, I would have done both eventually.

Having said that, it is true that the primary interest for me is because it's AXP rather than its age, so what Rachel said applies also; if I could have gotten hold of an AlphaStation with dual EV6's or even something with an EV7 (like i960MX, that one was not sold on the open market), I would have done; but prices for those even on eBay are around the £2K mark. If AXP had been upgraded like 8086 has been and there was a modern version I'd prefer to buy that, but even if I won the EuroMillions, I doubt the Chinese would sell me a copy of their "AXP rip off CPU" and a supercomputer to go with it

Finally, for the same reason as any of us working on GZDoom, QZDoom, LZDoom, Raze, EDuke32, Realm667, id's original DoomEd in a virtual NeXt Station, the Cats of Doom, or for that matter, my zdoom-dev-approved rework of the morphing subsystem a few years back ... "Because it's there "

[Blzut3]

to "WTF? Ken Olsen refused to sell Apple AXPs in 1993 for (what became the Powermac line) and brought down DEC in the process",

That's a thing? I don't think I've heard that story before.

It is however a key reason I think that x86 became and remains the dominant platform: it's available at every price point. To be fair until recently native app compatibility has been an insurmountable hurdle making x86 effectively the only option, but today with a lot of stuff being web apps it continues to perplex me that the mid range market always gets ignored. That is developer workstations. It's like they're intentionally blind to the fact that developers are going to want to deploy to the same architecture they use locally when given a choice.

Fortunately between the Raptor Computing POWER9 boards, the Honeycomb LX2 ARM workstation, and the up coming SiFive HiFive Unmatched board (and I guess the Ampere ARM workstations although those are server boards shoved in a tower so slightly off the mark), we've seen a little bit of change there. Of course people balk when they see the price since they're so used to the economies of scale of x86. Definitely the potential for interesting times ahead for ARM though.

I was actually mainly asking about the monetary aspect here.

As MartinHowe said a lot of people have hobbies that they spend irrational amounts of money on. Given that one of the things you've made known is that you don't even like building your modern computer I honestly don't think there's any way that we can hope to explain the joy of this kind of retro computing. Which is to say that your cost benefit analysis is going to be restricted to "is this hardware better than this other hardware," rather than view it from the entertainment standpoint like say the cost of buying a game or a movie. (Where the decision to buy one item isn't based on it necessarily being better than the other but simply because it's a new experience.)

Granted this particular form of retro computing is kind of niche even among people into retro computing. For me I like finding answers to often bizarre questions about hardware.

Having said that, it is true that the primary interest for me is because it's AXP rather than its age, so what Rachel said applies also; if I could have gotten hold of an AlphaStation with dual EV6's or even something with an EV7 (like i960MX, that one was not sold on the open market), I would have done; but prices for those even on eBay are around the £2K mark. If AXP had been upgraded like 8086 has been and there was a modern version I'd prefer to buy that, but even if I won the EuroMillions, I doubt the Chinese would sell me a copy of their "AXP rip off CPU" and a supercomputer to go with it

These kinds of lineages always amuse me. There are a few stories like this, for example how the Rise mP6, a fairly obscure processor that one would think would just fade away, changed hands a few times and still exists today as the DM&P Vortex86. It's like processor designs never die, they just pop up again where you least expect them.

Speaking of China, I wonder if Loongson is going to continue developing MIPS now that the company MIPS is doing RISC-V. Really wondering if that's the final blow for MIPS or if it's going to continue limping on.

[MartinHowe]

@Blzut3: Regarding Ken Olsen and his "Oh sod the proles, worthless bastards to a man!" attitude to Apple:

DEC: The mistakes that led to its downfall, University of London, date unknown, page 7
At Digital Equipment, Ken Olsen Is Under Pressure to Produce, Wall Street Journal, 2011
DEC's Final Demise, Forbes, 2000
Bonfire of the Vaxities: Avoidable mistakes that took DEC from penthouse to gutter, Techmonitor, 1994

Pity about that supercomputer, though - it might even run Planisphere 2 at a decent frame rate

[MartinHowe]

Time for a progress report. The NVidia GeForce 8400GS with 512MB VRAM works fine in a PC and on the AlphaServer with its NT bios (AlphaBios); the VMS/UNIX bios (SRM) hangs when initialising the card. I grabbed a USB FDD and some discs to create a bios update floppy set on my home computer then updated SRM to v6 and AlphaBios to the corresponding version; thank ${DEITY} someone had archived all of DEC's firmware FTP website, as SRM was only at v5.4.

Of course, the latest firmware, from 2001, identifies itself as Compaq rather than Digital (at least it didn't say HP ). I would use MILO (a LILO look-alike, this is the Linux bootloader that was used in the early 1990s before GRUB came along, so I have used it before) but there isn't a MILO bootable image for my machine as everyone was expected to just use SRM, rather than start MILO from AlphaBios. I may be able to build a MILO for it, if only I can find the MILO source, which seems to have gone missing from the internet.

Updating SRM didn't help much, but at least it ignores the card if it's on the higher numbered PCI bus, so I have my puny Permedia 2V as the default card and am hoping to get the Nvidia working as a secondary card once Linux is installed. I've never tired a dual-monitor or dual-card setup, so that;s another learning curve to undertake.

PS: It's instructive to note that Digital was doing a variation on "press F8 to flash firmware" in 1999 - long before PC manufacturers stopped using arcane boot floppies, DOS apps, WinPhlash, etc.; DEC was way ahead of its time in this regard.

[Blzut3]

Dual card or single card multi monitor basically makes no difference these days in terms of setup difficulty. That said once you figure out if Nouveau is even going to work on Alpha the thing to look into would be just blacklisting the Permedia card from the Linux kernel. Then you'll just have a setup where one card is used at the firmware and then auto switch once the kernel boots up. (Well auto might not be the right word since it's likely the Permedia will just hold the last frame.) Basically use the same techniques the vfio community uses to dedicate GPUs to VMs, just in this case the Permedia isn't going to do anything.

[Redneckerz]

got it for £60

Ouch! The price for these old cards is truly insane.

I think this is the point where I seriously have to ask how much it is worth to get some obscure vintage computer working.
Please enlighten me, what's the fascination here - I could understand if this was usable for running some old software, but what can this thing actually do...?

Getting old things to run things they aren't supposed to or you wouldn't expect they could do. Something similar would be like a SGI Indigo2, maxed out, running Quake 3 or having it running a OpenGL build of Doom, which it technically should as ports exist to that platform that have OGL support.

Or seeing dynamic colored lights on 3D Blaster VLB (3D Labs Glint based). Or Doom 3 style stencil shadowing and lighting on a card that does not have shader support (Severance: Blade of Darkness does this.)

Its an itch for a specific kind of people. With limitations bores creativity.

Some kind of endorphin gets released whenever I do <3d game art thing> for a virge, pcx2 or whatever of similar capabilities that modern shadery hardware can't do... (even with all the 'retro aesthetic' filters in the world)

Exactly this. Always love your posts around obscurity things, lei! Learned a great bunch about GPU's that way.

Works beyond hardware aswell. Ive recently taken a keen interest in getting Esselfortium's TRANMAP hacks compiled for easier access. One of the few things GZDoom can't do beyond supporting reading a TRANMAP lump since the effects you can make are primarily for 8 bit software and OpenGL cannot emulate that kind of thing (And why should it) properly.

[MartinHowe]

Time for another progress report. I am using an old Debian version (5 aka 'lenny') on my Alphaserver because only that and Gentoo seem to support Alpha AXp at all nowadays.

Debian 5 is from 2012 and hopelessly out of date. Gentoo is a nightmare to install and currently their 'bootstrap the install' CD crashes on my computer. So I have tried to do it from the Debian, given that I have multiple HDDs. However, this requires starting a basic version of Gentoo in a kind of VM called a chroot, and my Debian kernel is too old. The guys at Gentoo suggest all kinds of things; cross compiling a kernel over a network, installing Gentoo on another computer that has SCSI capabilities (I have none), and so on. While I realise from their POV they are being helpful, and I have learned some useful stuff from them that I am using now, this is way too much trouble when all I want is a working installation.

The Gentoo folks' ethos is having fun facing the challenge of getting Linux working on a system and making it as lean and mean as possible (hence the name: Gentoo is the fastest swimming penguin); however, to me, the working Linux is a means to an end and not and end in itself; the challenge I want to fight is getting a half-decent form of Doom engine on the thing; the system itself should 'just work'. It would be a bit like telling the SAS that they are wusses because they don't spend a day in a forge and make their own ammo before going into battle But why should they? That's what armourers are for! I mean, I have no interest in Gentoo beyond (a) it is modern and (b) it runs on Alpha; if there was an up-to-date Debian derived distro that runs on Alpha, I probably would never even have heard of Gentoo Despite that I do enjoy some of the things they do; it's just that right now, it's a lower priority than Doom unless I win the lottery and can retire and spend as much time as I want on Gentoo before doing Doom.

So it seem like the best course of action is to upgrade each package of my old Debian system by compiling from scratch, at least until I can build a modern kernel and any associated dependencies. Maybe even get a nearly modern Debian in the process. It seems also that for direct frame buffer access (equivalent to DirectX in Windows) for Permedia 2V graphics card, my kernel is too new; yes you read that right, too new! So if I want OpenGL working properly for games, I'd have to find the 2.3 kernel source, rip the relevant kernel driver sources, and build them as a module.

Logically, this means starting with the build tools and a few other essentials; I have built the latest versions of tar, automake, autoconf, m4 and libtool from source and am currently building GCC 4.8, which is required to build the latest GCC. Then I'll have to use the latest GCC to rebuild (a better copy of) itself and all the other stuff. Mustn't forget to do make as well. I guess in a sense, I'm using the old Debian as a sort-of 'live cd'.

I'll let y'all know how I get on ... building GCC, even for just the c and c++ compilers is probably an overnight job ... I may be gone some time

Some kind of endorphin gets released whenever I do <3d game art thing> for a virge, pcx2 or whatever of similar capabilities that modern shadery hardware can't do

Ah, yes, I seem to remember seeing your username or a variant of it on VOGONS or somewhere like that

[Blzut3]

a kind of VM called a chroot,

A chroot isn't remotely a VM. It's literally what it says on the tin, changing logical root directory to another one for a process. This allows you to, for the most part (some issues with selinux/apparmor have to be corrected if used) use the user space applications and libraries of another system under your current kernel space.

Pretty much all distros are built around this idea for their installer. Haven't set up a gentoo system but they may make it more obvious what they're doing, but Debian/Ubuntu has debootstrap and Arch has pacstrap which does the work of setting up the bare minimum packages into a chroot to operate the package manager. Once this is done further package installs are done from inside chroot, and then some clean up is done at the end before you reboot into your working system.

rip the relevant kernel river sources, and build them as a module.

You forgot the step in the middle there: rewrite it due to the many API changes over the years. Basically no chance that you're going to get acceleration on the Permedia card going.

Logically, this means starting with the build tools and a few other essentials; I have built the latest versions of tar, automake, autoconf, m4 and libtool from source and am currently building GCC 4.8, which is required to build the latest GCC. Then I'll have to use the latest GCC to rebuild (a better copy of) itself and all the other stuff. Mustn't forget to do make as well. I guess in a sense, I'm using the old Debian as a sort-of 'live cd'.

I'm surprised you had to rebuild those basic tools. I know they're old but RHEL5 is older and didn't give me much trouble in building new GCC or whatever. Obviously with GCC 11 you need a newer intermediate compiler, but just saying that I didn't need to built autotools or make. Certainly doesn't hurt though.

Also you don't need to build GCC 11 with itself, the build system does that for you (provided you're not building a cross compiler which you aren't, or you explicitly disabled bootstrapping). Which is part of the reason why it takes so long.

[MartinHowe]

Thanks, @Blzut3, for the useful info. (I knew chroot wasn't a VM, but was unsure exactly what it was and that was the nearest analogy I could think of.) For building the Permedia driver, I had hoped that since 2.6 isn't that much of a jump from 2.3 it would be possible I rebuilt the other tools simply because it seemed a good idea, as when previously building prerequisites for things like chocolate doom I got a lot of "toolname-xxx or higher is required" and wanted to avoid that.

I had no idea the GCC system will re-build itself in those circumstances, but thanks for that, as it's been 12 hours and still going I've also attempted a cross-compile from my (relatively modern) Dell 490; never attempted a cross-compile of anything before and have probably screwed it up somewhere along the way, but hey, this is a big experiment

EDIT1: I have found an online PDF of the programmers manual for the TI version of the Permedia card's chipset, but it's likely not exactly the same as a pm2v. Will be interesting to read, though.

EDIT2: Cross compile failed with "cannot compute suffix" - that's 'filename extension' to you Windows plebs - "for object files". Oh well, it was worth a try; was a last minute idea before leaving for work; this time I will actually have to properly read the fine manual

EDIT3: Self-hosted compile going on the AlphaServer for nearly 15 hours and it's linking now - time for some heart attack pills just in case

EDIT4: Feel the power of the Dark Side of the Force! (or Google as it's known these days): http://www.vgamuseum.info/images/doc/3dlabs/permedia2/

[Blzut3]

I had hoped that since 2.6 isn't that much of a jump from 2.3 it would be possible

Don't forget that prior to 3.0 they were still using the old school versioning scheme where everyone treated the first two numbers as sacred and thus never incremented ever. The 2.6.x series saw lots of big changes throughout and thus why the version scheme changed to a more arbitrary one.

EDIT2: Cross compile failed with "cannot compute suffix" - that's 'filename extension' to you Windows plebs - "for object files". Oh well, it was worth a try; was a last minute idea before leaving for work; this time I will actually have to properly read the fine manual

I definitely can't say I recommend building a cross compiler unless you really know what you're doing. There are lots of ways to do it wrong and plenty more ways to get a working compiler that still produces binaries that don't work on the target system. That particular error is the unhelpful final error, you'd have to look in the relevant config.log to find the real error. Probably some issue with your alpha cross linker.

[MartinHowe]

Well it doesn't really matter now; I was able to get the previous Gentoo minimal install CD to boot on my system; there's graphics corruption, so it's question of blindly getting networking up so I can SSH in and so on. So I'm now unpacking the tarball with the system to be chroote'd into (I set it up wrong last time anyway). I'll let y'all know how I get on.

[MartinHowe]

Time for another update. I have abandoned attempts at Gentoo for now, it's too much hard work for too little reward; I wouldn't mind if at least the things in the handbook actually worked as advertised, but there's definitely an attitude of "spend hours of Googling to find how to work around a flaw in the product" and not enough of "sorry, we'll fix it when we can". This is at least arguable for edge-of-the-world stuff like Alpha AXP, but some of it is unrelated to my architecture by nature, plus I've seen others on x84 or ARM have the same issues and that really annoys me; basic system-setting-up should at least work as advertised on mainstream platforms and problems that have been around for a while should really have some sort of guided mitigation than "Google". Also, and this isn't anyone's fault, compiling anything beyond a large single file app takes ages; even with threading and both processors, building GCC took 15 hours.

Some of the advice and encouragement from the Gentoo people was helpful and I have implemented some of it; there are a lot of helpful guides on Google for some things, namely setting up a cross compiler; I now have an x86-64 to EV56 cross compile toolchain and have built my first kernel It actually works! I haven't yet got it to link to the userland on the existing Debian installation, but I can at least get a Busybox and access the real filesystems:



Of course, if I could have afforded a better system I would; the AlphaServer 53XX is a 12XX in a different coloured case and set up by default for running Windows NT; however, the 12XX itself is a 40XX in a smaller case that can take up to 2 processors; the 40XX can take up to 64. Imagine owning a 64-processor Pentium 4 machine in 2003 - and playing DooM on it There are even companies who supply and support these old machines for folks who are running things like a 30 year old hospital scanner that only runs on Tru64 Unix or Windows 2000. Well, I know what I'm buying first when I win the lottery

[KynikossDragonn]

Imagine owning a 64-processor Pentium 4 machine in 2003

That sounds like a fire hazard, and a good way to blow out a circuit breaker too.

Speaking of power, do you know how much wattage your Alpha uses?

[MartinHowe]

I haven't yet got it to link to the userland on the existing Debian installation

Er, well, I seem to have promised a lot of demons my Eternal Soul tonight ... just as well I'm an atheist



The short story is I copied the Debian Lenny initrd.img file, included the kernel modules for the modern kernel
and a few script fixes, then wrote it back with my kernel's initrd name. A couple of issues, but at least it boots the
Debian 5 userland now

[Blzut3]

The short story is I copied the Debian Lenny initrd.img file, included the kernel modules for the modern kernel
and a few script fixes, then wrote it back with my kernel's initrd name. A couple of issues, but at least it boots the
Debian 5 userland now

That was probably the hard way to do it. Not sure why you couldn't just use update-initramfs (or whatever the equivalent was back then) after installing your kernel and modules to the file system, but whatever works. Out of curiosity any particular reason you went with 5.4?

Glad to hear you got a working cross compiler for the kernel at least. Hopefully getting it to actually build user space applications won't be too much more work.