SGI Announces MIPS and IRIX End of Production

ramakant writes "Considering the recent news regarding their dismal financial situation, it should come as no surprise that SGI announced end of production for MIPS based hardware and the IRIX operating system. From the article: "SGI launched the MIPS/IRIX family of products in 1988. Since then, this technology has powered servers, workstations, and visualization systems used extensively in Manufacturing, Media, Science, Government/Defense, and Energy. After nearly two decades of leading the world in innovation and versatility, the MIPS IRIX products will end their general availability on December 29, 2006." IRIX has always been my favored OS, and I'll be sad to see it gone. Hopefully my O2 will survive for many years to come."

Comments

MIPS is going away?

[EvanED]

Now what narrowly-deployed architecture for which everyone runs a CPU simulator will be taught in computer organization and assembly language classes?

Re:MIPS is going away?

[OrangeTide]

Systems with a clean instruction set are apparently unpopular in the real world.

PowerPC is rather nice, but it's not as clean. (but it is easier to use)

Re:MIPS is going away?

[AKAImBatman]

Now what narrowly-deployed architecture for which everyone runs a CPU simulator will be taught in computer organization and assembly language classes?

Duh. They'll emulate the 6507 in the Atari 2600. That way they can run it on real, modern hardware! :P

Re:MIPS is going away?

[Bluesman]

It'll all be Motorola 68000.

Anything but Intel, it's just way too easy to find recent and cheap hardware and software that's i386.

Plus the textbooks won't have classic lines like,

"Today's machines now come standard with up to 4 megabytes of Random Access Memory, and this continues to increase every year!"

Re:MIPS is going away?

[DrDitto]

MIPS is not going away. They are a seperate company that now focuses on the high-end embedded market.

Re:MIPS is going away?

[Tarlus]

I'm sure they'll continue to teach MIPS assembly for many years to come simply because it's the easiest to teach and learn. SGI's dropping of MIPS won't matter since every computer-organization student is emulating it with SPIM on more common architectures, anyway.

Re:MIPS is going away?

[AKAImBatman]

To be serious for a moment, the 6502 series (which included the 6507 and 65C02) was an excellent processor architecture that was incredibly easy to learn on. Its small instruction set, focus on 8 bit instructions, and logical segmenting made it popular both in real usage (Commodore 64, Atari 2600, Nintendo Entertainment System, Atari 8-bit computers, Apple II, BBC Micro, etc.) AND in teaching.

Just about anyone can learn to program it by reading the documentation. It's so simple, it can even replace BASIC as a first language. (Although, you might have a bit of diffculty doing "Hello World" unless you understand the hardware you're programming for.)

Re:MIPS is going away?

[ajlitt]

MIPS isn't going away. MIPS is very popular for embedded video processing. TiVo is MIPS (now, at least), the PSP is MIPS, and many DVD players are based on a MIPS. MIPS is still popular because the ISA isn't patented and there are a number of compatible cores out there.

Re:MIPS is going away?

[neonprimetime]

I for one teach an undergraduate Assembly course in the Spring, and I intend to continue using SPIM in the class.

Re:MIPS is going away?

[neonprimetime]

Can I ask a legit question then, since you seem to know what you're talking about :-) What if any effect SHOULD this announcement have on current undergraduate Assembly courses that teach MIPS? Thank you in advance.

Re:MIPS is going away?

[10Ghz]

SGI MIPS-workstations are going away, MIPS itself is not going anywhere, It's still running in millions of embedded devices, and more will be announced in the future.

Re:MIPS is going away?

[LWATCDR]

I will answer. None at all.
This announcement is about the end of MIPS as a server and workstation platform. The vast majority of CPUs are not used for server or workstations. They live in toasters, DVD players, digital cameras, microwaves, and so on. In the real world very few people ever write assembly programs that run on a server or a workstation. However in the embedded space assembly is still pretty common.
MIPS isn't dead. MIPS servers are dead. MIPS lives on in many devices.

Re:MIPS is going away?

[TeknoHog]

More importantly, what units shall we use to measure CPU performance when MIPS goes away?

Re:MIPS is going away?

[vincecate]

That MIPS is not dead is both true, and a good point. However, the Zilog Z80 is not dead either. But we don't get very excited about it any more. When we got our first one, it was way cool, but today most people would not even think about the fact that Zilog still sells Z80s.

Re:MIPS is going away?

[Duhavid]

I dont know... It was pretty crappy.

1 register you can use that isnt used by something
else ( ax ) ( bx was used for something, cx
was used in some "counting" instructions,
dx would have the most significant 16 bits
of a multiplication ( ax would have the lower,
now that I think of it.... dx:ax would be
the full 32 bit result ). So, ax that, 0 registers
you can use that arent used by something else.

Addressing ( "long" ). 20 bit addressing bus,
16 bit system. So, you load a register with
a 16 bit value, and wink wink, it gets shifted 4 places
left. 20 bits it is now. Then, load another
register with another 16 bit value, which is added
to the other you just did, viola, a 20 bit address.
Dont ask me how the two registers you just loaded
related to each other, I am trying to repress that.

Re:MIPS is going away?

[AndyboyH]

Who the HECK puts cpus in a toaster?!?

How do you think the toast gets cooked overwise? It ain't going to brown itself...

Meanwhile, on eBay

[Intron]

204 items found for SGI.

Good times for collectors.

Irix was cracker paradise

[phayes]

Irix itself wan't that much worse than any other *nix of the same time period, but none of the varied tools, 3D bells & whistles that SGI bolted on were designed with security in mind. The only way to avoid getting hacked into was to remove it all before connecting it to the net, but once you removed it there was little point in buying one.

Oh Gosh

[sarathmenon]

Now I have to dump my IRIX, right after I dumped SCO UX. This just isn't fair!

Too bad - MIPS was pretty

[Just Some Guy]

My computer architecture class textbook was based on MIPS, and after messing around with 68k and x86 assembler for years, its assembler was like a breath of fresh air. It had a truly elegant design, or so I thought, and it's a shame to see it die.

Alpha, MIPS, and others - where are you now? x86-2^x is pretty much all that's left for general-purpose programming these days (although Sun might have something to say about that), and that's too bad. Kind of like how you can't be a great programmer without ever having seen Lisp, you can't be a great chip designer without ever having known something that doesn't run IA32 code.

Re:Too bad - MIPS was pretty

[BrentRJones]

Consolidation of operating systems is not good for computer science. Variety increases knowledge just as travel expands the mind. Everything is heading for the day when there is one primary OS on one primary HW platform. Let's hope that creativity spawns many new creations.

Re:Too bad - MIPS was pretty

[mnmn]

MIPS was great and still has life left in it. However ARM has been bulldozing its way through recently in the higher embedded markets where MIPS was strong. Even AMD sold Alchemy eventually.

The embedded market was getting crowded, which is a good thing. The survival of the fittest gave us ARM instead of us being stuck with assembly codes like the PIC and x86.

Re:Too bad - MIPS was pretty

[be-fan]

It's quite true in a sense, though of course it depends on your definition of "great". If "great" is just someone who is very good at his/her job, then yes, you can be a "great" programmer without ever using Lisp. However, if you define "great" to mean somebody who really has a higher-level perspective about programming as a whole, then I'd have to say that you can't be "great" without experiencing Lisp or something like it.

Put simply, different languages represent different areas in the design space of programming languages. C represents one area, C++/Java/C# another set of closely-grouped areas, Ruby another, Python another, etc. Lisp represents a very large, and to many people used to C++/Java/etc, very novel portion of the design space, as does ML and its kin. A truely great programmer, then, must not only be proficient in the usage of a specific tool that represents a specific point in the design space, but must have a perspective of the whole design space. He must be able to look at specific solutions, and realize when they are just instances of a higher-order, more generalized principle. The only way to gain that perspective is to explore the design space, and mastering Lisp is a way to explore a very large and unique part of that space.

There's a saying that "to a man with a hammer, everything looks like a nail". This refers to a general notion that our tools limit the ways in which we think about solving problems. Let me give you a concrete example. Lisp has a feature called multi-method dispatch, in which the target of a polymorphic call is decided by the types of more than one of the arguments. To someone who has only ever used a language with single-dispatch (ie: C++/Java/C#'s virtual methods), the very idea of using multiple-dispatch to solve a particular problem never even comes to mind. He makes due with what he has, using techniques like the "visitor pattern", and sits content thinking that this is the best he can do. Somebody who knows Lisp, on the other hand, might still have to use the visitor pattern (because his boss forces him to use Java), but he'll realize that its just a way to do double-dispatch in a single-dispatch language, and that increased understanding of the nature of the solution will help him write better code.

SPARC?

[Grendel Drago]

I learned on SPARC; I thought everybody else did too...

Re:SPARC?

[EvanED]

Penn State uses MIPS, I think Cornell uses MIPS (at least the musical I read based off of their org course seemed to indicate so, at least from the hardware point of view), I'm pretty sure Wisconsin uses MIPS. Heck, the aforementioned CPU simulator (SPIM) came out of U. Wisconsin. And those are the only three places I have any inkling about.

I do know that PSU *used* to teach SPARC in a standalone assembly course, but that was later combined with the org class and at that point changed to MIPS.

(BTW, an addendum to my original post, I know that there are plenty of SGI machines around, so "narrowly-deployed" is probably too harsh. That said, the only time I've run any MIPS code we wrote in that class was on SPIM.)

Re:SPARC?

[E-Lad]

I think MIPS was the popular arch to learn asm on. Here at UMBC, MIPS is still what the assembly programming courses revolve around. In the mid 90s, SGI/IRIX was popular in (american, at least) universities. This course is pretty much one of the only reasons why we keep a few O200s around (including a 24-CPU Challenge XL... well, okay, it's now 16 CPUs, because we seem to be seeing one CPU board die each year). It's funny because back in the 90s, the Challenge XL was billed to faculty as a high-speed research computing server, which it was - at the time. Some of the old timers believe that's still is true today, probably because they just don't know better. 16x 200Mhz CPUs ain't all that, no matter what arch you're on.

Hopfully we can convince the CS dept to move their course off of MIPS so we can push these aging servers off the end of the loading dock. SPARC or x86/64 would be the alternatives here.

Re:SPARC?

[LWATCDR]

X86!
using the X86 to teach assembly language is like using Perl to teach object oriented programing.

No need to move off MIPS, just use an emulator.

Re:SPARC?

[jd]

You can't confirm that! You're not Netcraft! :)

Re:SPARC?

[drinkypoo]

Using x86 to teach assembler is great! It means that once you've struggled and fought with that piece of shit, everything else will be easy. By contrast, MIPS is too easy. After that, everything else will be harder :)

Re:SPARC?

[mnmn]

x86 assembly is more useful than most other assembly. Once you learn it you can further learn x64, MMX etc and make fast drivers and codecs that you can (1) sell (2) get a job through.

I learned x86 asm around 1994 mainly because there was nothing else for a 15 yr old with a PC, and because x86 even back then was pervasive enough.

I was trying to build a boot code virus using instructions and code taken from a BBS server.

I failed to infect my own computer.

Re:SPARC?

[swillden]

By contrast, MIPS is too easy. After that, everything else will be harder :)

My first assembly language was VAX. For those who are unfamiliar with it, the great thing about VAX assembler was that there was an instruction for everything. For example there was a machine instruction that performed a quicksort. The old joke was that you could write any program with a single instruction, if you could find it and figure out how to use it.

Re:SPARC?

[hey!]

Jokes aside, the thing about VAX assembly programming is that the instruction set was probably one of the last ones designed where human readability was considered a critial factor. I swear that VAX assembler was almost as easy as coding in C.

Programming after all is a matter of mastering idioms. Good programming is often largely a matter of choosing sensible conventions and sticking with them. The thing that kills you in the system is lack of orthagonality. Broadly speaking, what I mean by this is that it's important that when you write a program that looks like it ought to be syntactically valid, it should be. Furthermore, it should work more or less the way a sensible programmer unfamiliar with the language would guess it's supposed to work. I'd rather write programs in VAX assembler than Microsoft Transact-SQL dialect because of orthagonality. T-SQL is the most non-orthagonal language in widespread use; you're constantly having to look in the documentation to see if such and so will work in this context.

Re:Shame

[MightyYar]

IRIX is System V-based, and thus probably encumbered with SCO nastiness... I wouldn't expect it to be open sourced. Perhaps the parts that were developed by SGI could be, though. They already released XFS under the GPL, for instance.

FOSS

[Anonymous Coward]

I think they should release IRIX under the GPL and let the community maintain it!

Re:FOSS

[archen]

I'm not sure where the logic for something like this comes from. Like there is an infinite ammount of people who will work on every project ever abandoned. If IRIX was so wildly popular, I doubt SGI would stop working on it. I'm sure there's lots of useful code in there, but I'm also sure it's littered with stuff that has a patent on it as well. SGI is still a company that seeks to survive (I would assume), and isn't doing so well. They are in no position to work on figuring out licence issues to put IRIX under the GPL.

Well this is the first of the old school Unix's to fall that I can think of. AIX and Solaris will probably be last. The ones maintained by Hewlet-Compaqard will be next in line after the death of SCO derivatives.

Re:FOSS

[Burdell]

The Alpha and Tru64 Unix are going away first. The last order date for a new AlphaServer is October 27, and (despite earlier Compaq and HP promises and guarantees) Tru64 and its related technologies die with the Alpha.

Re:FOSS

[hackstraw]

I think they should release IRIX under the GPL and let the community maintain it!

I believe that is called Linux. SGI has already released bunches of IRIX to Linux including ccNUMA code and XFS and I'm sure other goodies as well.

Re:FOSS

[turgid]

It's just Another UNIX System Vr4 variant that runs on SGI MIPS hardware. You get all the same and more in Solaris (swap MIPS for SPARC) since it had much more R&D. You also get the same and more in Linux, but without the official Sys Vr4 codebase.

Re:FOSS

[Paul Jakma]

I'm still waiting for opensolaris' complete sources to go through

See cvs.opensolaris.org. Every bit that Sun can release has been released.

ARM is patented

[tepples]

ARM?

MIPS is popular because it's unpatented (except for a few less common instructions, which aren't taught in Computer Organization and Design anyway). A common term project in computer architecture courses is to design a reduced implementation of the MIPS architecture on an FPGA; some students go beyond this and end up with Plasma. The ARM architecture, on the other hand, is still patented.

(arcem isn't maintained, from the looks of it, but it's a neat pure-hardware-level ARM platform simulator.)

The most popular ARM platform simulator nowadays seems to be VisualBoyAdvance.

Re:ARM is patented

[dgatwood]

The ARM architecture, on the other hand, is still patented.

Those patents should all have expired by now, at least for the original architecture. Patents filed prior to June 8, 1994 have a term of 20 years from filing date or 17 years fro issue date, whichever is greater. ARM1 was in development testing in 1985 and shipped in 1986. Unless some of those patents too more than four years to be issued, they should be in the clear by now. Of course, you'll have to do a search to be completely certain, but....

The thumb instruction set, on the other hand, does have currently active patents, I believe.

A discussion of this issue can be found here.

Make patches available

[pathological liar]

What would be awesome is if they made *all* the patches available after the EOP/EOL period. As of right now there are a lot of them that are restricted to folks with support contracts. Ideally they would make the core OS available as well instead of just the overlays, but I'm not going to hold my breath on that.

It'd be nice though.

IRIX==Motorbike.

[sbaker]

I always remember talking to some vendor at a usenix conference a few years after the birth of IRIX. We were talking about the relative benefits of SunOS (Solaris as it is now) versus IRIX. The guy said that using IRIX compared to SunOS was like riding a motorbike compared to driving a car. "It's fast, it's a rush, it's more fun than you can possibly imagine - but it's easy to fall off - and when you do it hurts a lot more!"...that pretty much says it all.

I spent a large fraction of my most productive years sitting in front of a million dollar computer with IRIX in my face. It was pretty good - but with SGI's market share shrinking and Linux getting so mature, it makes sense for them to dump the hideous cost of maintaining an entire OS by themselves. For SGI, it's a good decision in desperate times.

We split from using SGI to off-the-shelf PC/Linux about 5 years ago - about as soon as nVidia's graphics got good enough for our needs. A PC costs about 1% of an SGI with similar horsepower...QED.

As for MIPS, the equation is the same one Apple had to face down. Performance = Horsepower per CPU / Price per CPU -- and whilst your own solution can win on horsepower, you can't beat the price of whatever is made in the largest quantities...and it's the same deal as with IRIX - when you have to cut costs, designing your own CPU isn't the smart way to go.

Sad - but inevitable.

It's a UNIX system. I know this!

[earbenT]

Lex's skills are useless now. :(

Such a crowded graveyard, big deal.

[museumpeace]

Apollo, DEC, Amdahl, Prime, RCA, Remington Rand, GE, Univac, Perkin Elmer, MassComp, Concurrent Computer, Compaq, Sequent, Encore, Xerox, Scientific Data Systems, Wang, GO corporation...and so many more.

The only lesson you could profit from in all this carnage is knowing when to sell your shares, when to find a good merger rather than waiting for the bankers to hold a fire sale of your patent portfolio.

Re:Such a crowded graveyard, big deal.

[1lus10n]

True, but we could still use technology driven companies like DEC, Sun and yes even SGI.

When left to their own devices most of the large computer companies (IBM, HP, Dell, even Intel, AMD, Cisco etc etc) do very little revolutionary or insightful things. They usually tread water with minor "improvements" until someone comes along and kicks them in the pants (see: IBM vs Apple, IBM vs DEC, Intel vs AMD etc etc) with some better technology.

If all we have left are the "big guys" where is the next revolution going to come from ?

Re:Pick an OS with staying power

[jgrahn]

I feel sorry for the person that picks an OS dependent on a corporation for its existence. When there is only one "Sun" to nourish your OS "ecology" it is much more likely to wither away - eventually. I picked an popular open source OS for this very reason. RedHat may die but it will take a unprecedented disaster to also kill off Ubuntu, Debian, Slackware (especially Slackware), SUSE Linux, etc.. My intellectual investment is safest with Linux.

This is UNIX! You're supposed to be able to take your ecology with you to Linux, or Solaris, or OpenBSD, or wherever. With some pain, admittedly---but little more pain than if you're migrating from, say, RedHat to Debian.

If you read all the way to the bottom

[Anonymous Coward]

For those of us who still work at SGI and continue to support this product line,
also of importance at the bottom of the article is:

SGI is also committed to offering our customer a full level of support needed to protect their investment in SGI products. End of support (EOS) for MIPS/IRIX products is currently scheduled for no sooner than December 2013. SGI Technology Solutions is continuously evaluating the demand for extended support and may consider longer extensions if necessary.

Re:Sad

[RAMMS EIN]

``It seems innovators, the "product guys", have a danger of being overrun by companies with more agressive marketing. The technology lovers, the hackers, don't always make it in a world run by economics.''

Nor do those who price themselves out of the market.

Re:Pick an OS with staying power

[iggymanz]

oh, we should all be sorry for people who chose mvs or vm/cms in the last few decades? SunOS 1.0 was released in 1982, and today we have SunOS 5.10 aka Solaris 10. we'll have to see if there's Linux distros in 24 years before we can really pass judgements on longevity compared to Sun's stuff. BSD is still widely used too, and fifteen or more years older than GNU/Linux distros depending on how you look at things, so I'd say any BSD based stuff has even more safety than Linux (and Linux keeps borrowing good stuff from it and verse vice)

IRIX was obviously going away.

[Medievalist]

SGI ported their graphics code to linux years ago, so that they could eliminate the cost of maintaining their own unix variant.

Even chkconfig reasonably standard in mainstream linux distros. IRIX is not worth the effort.

They can now concentrate on their core competency, which is presumably better graphics hardware than their competition.

I guess Erwin will have to start shopping for spare parts on ebay...

Re:IRIX was obviously going away.

[Mr. Hankey]

They do make clusters as well, which I suspect is where they're really going to dig in. I work in a HPC environment and there are some fairly large SGI systems (how's 10240 CPUs sound?) in the building next door that scale quite well.

Re:Pick an OS with staying power

[Anonymous Coward]

I am an anonymous coward.. err.. don't really feel the need to create an account. But, as an SGI employee I can attest to Linux scaling very well on SGI. How about 1024 Itanium2 processors as a single system. No, this is not some cluster machine. One copy of linux runnuing on 1024p with many terabytes of memory.

Then there is a 4096 processor machine without manky infiniband interconnect or myrinet nonsense that you can run MPI programs against.

And if you really want a big cluster machine, how about 10240 processors addressable via MPI over infiniband.

Re:Why aren't they selling x86 and Linux?

[kimvette]

Uh, they did, and that is what killed them.

Re:Why aren't they selling x86 and Linux?

[Animats]

They tried that. Didn't help.

SGI also tried making overpriced Windows desktops. That also flopped. Nice cases, though.

No problem: COMPAT_IRIX in NetBSD/sgimips

[hubertf]

You can run Irix binaries on NetBSD/sgimips. See http://www.onlamp.com/pub/a/bsd/2002/08/08/irix.ht ml for more information, and check out the NetBSD port's page at http://www.netbsd.org/Ports/sgimips/.

PowerPC is superscalar. ARM isn't.

[tepples]

If you can figure out a way to implement the instructions in a way that differs from ARM's patented implementation, then you are free to do so.

If you can figure out a way to implement LZW or RSA or MP3 or any other patented codec in a way that differs from the patent owner's patented implementation, then you are free to do so. Unfortunately, no other way exists because the claims on those methods are rawther broad.

Of course, ARM has a built-in advantage due to its simplicity. Unfortunately, IBM engineers had a huge ego trip and demanded to build yet another RISC processor -- the PowerPC.

PowerPC was also built to scale to multiple functional units per thread, such that they can run a load, an integer arithmetic, a floating point arithmetic, and a branch at once. Are any ARM implementations superscalar? XScale sure isn't. Or are you talking about a massively multicore CPU, some sort of squared octopus with 64 ARMs?

Re:PowerPC is superscalar. ARM isn't.

[drinkypoo]

A judge also ordered that all their players get seized from a trade show.

Yes, based on a preliminary injunction, which is apparently quite easy to get in Germany. Their evidence has not actually been presented in court.

i.e. you lose.

Re:PowerPC is superscalar. So is ARM

[Anonymous Coward]

Are any ARM implementations superscalar? XScale sure isn't.

XScale is a very old ARM implementation. Cortex A8 is indeed superscalar. http://www.arm.com/products/CPUs/ARM_Cortex-A8.htm l

Re:What about Tivo?

[demon]

The MIPS core that is used in the TiVo Series2 systems is a Broadcom design, licensed by Broadcom from MIPS Technologies. I see no reason that that would jeopardize Broadcom's licensing arrangement... though I could be wrong.

Re:Pick an OS with staying power

[thanasakis]

Um, at least Solaris makes great efforts to make sure that an application that runs on Solaris.X will be able to be transfered seemlessly to Solaris.X+1. Thats one of the reasons there's all this legacy staff lying aroung in various directories. If you look at Solaris man pages, there's usually a note about whether the interface is stable and whether it will remain in the next releases. Even the output of commands tends to be relatively stable across releases. And of course there are cases of drivers (for example network cards) that are compatible across Solaris.8,9,10 because they were written according to the guidelines. Even the migration to 64 bits (on SPARC) has been done a decade ago. So, if you invest on it, chances are that your software will probably be able to be transfered unchanged to the next version, when it is around.

Linux is great, don't take me wrong, but in what way your intellectual investment is safe, when the entire landscape is in continuous flux? I mean, APIs are changing back and forth, kernel modules come and go across minor kernel releases, each distro has its favorite places where commands and files are placed. Not to mention the 32 to 64 bit migration which is in a terrible mess. This is not MHO, read July's Linux Journal editorial and laugh about it.

The truth is, I don't think you are not interested in any kind of intellectual investment, you are just betting on a horse because someone told you that in the end there can be only one. Well, guess what, "they" 've been saying the same since 1985, and today there is still Windows,Linux,all the BSDs,Solaris,AIX,HPUX etc etc. They aren't going anywhere anytime soon, and, really, it is great to have such a diverse ecology.

(Although I may have sounded harsh throughout the post, I want to stress that this is all just friendly advice -- kind regards)