Great MIPS chips of the past 30 years | Hacker News Reader

Great MIPS chips of the past 30 years

alexvoica.com

51 pointsalexvoica52 comments

Unless you were doing 3D graphics, SGI computers of the 1990s were overpriced. At my first startup in 1997 we benchmarked SGI and Suns against PCs of the time, and found white-box PCs offered a clear 3 x price/performance advantage. So we kitted out our test network and datacenter with racks of white-box PCs (literally desktop PCs, not even server hardware).

I still have no idea what the advantage of Sun hardware was supposed to be, except some people were locked in, either by software or some kind of Stockholm syndrome.

Edit: Photo of our test network: http://oirase.annexia.org/tmp/testbed.jpg

Reminds me of the Softimage demo in ~96 (after Microsoft bought them) where they compared SGI and vanilla PC box. The guy explained that if the PC being 1/10th the cost of SGI, even if it was half as fast it would still be worth it. Then, to the great satisfaction of Bill Gates, near him on stage, he said that the PC was actually faster than his SGI workstation.

We've been told that the 1988 Sun 386i was quite successful due to its ability to run MS-DOS, and that the Sun 486i was canceled in 1990 because it was faster than the in-house SPARC equivalents. But it took some time for Intel machines to be competitive "big iron", say starting in the 486 days? Including running a serious UNIX(TM) or Linux. ADDED: and serious I/O, we needed EISA and then PCI and its descendants.

So by 1997 Sun and SGI had first mover advantages, but neither being well run companies they squandered them even before counting Intel's general excellence in microarchitectures and fabrication. (Plus wasn't SPARC hard to make fast due to an architectural mistake?)

One other problem SGI had, their IRIX was a notorious nightmare, which is part of their not being a well run company. Sun also lost more than a little momentum when they switched from BSD to ATT UNIX(TM), and did brain dead marketing where they boasted their UNIX(TM) was less buggy than Windows NT at a time when that simply wasn't true (and if I recall that was at a time they were too often refusing to fix some bugs).

A thesis I've just developed here is that the switch to commodity Intel hardware decoupled hardware and software, they helped you avoid getting locked into a single vendor's mistakes (and AMD helped when Intel faltered in several ways). And open source software provided you many mechanisms to get your critical bugs fixed.

About IRIX, an old but entertaining document: http://www.art.net/~hopkins/Don/unix-haters/tirix/embarrassi...

How did the management at two successful companies that had delivered great products for a long period of time get so shitty so fast?

The 90s was, in retrospect and sort of at the time, one hell of a decade.

By trying to beat MicroSoft rather than servicing customer needs.

That would be Sun, right? Particularly Scott McNealy: https://www.google.com/search?q=scott+mcnealy+microsoft (not needed SGI's special sauces, almost all my direct experience here is with Sun and DEC).

In addition, the two biggest and interlocked problems with them that I know of are their ruining their reputation in enterprise space when they messed up an off-CPU chip cache design, required customers to sign NDAs before they would try to fix it, and then blamed the customers and the supplier of the SRAM chips for their non-fault tolerant design. Instead of, you know, the IBM ethos of fixing the customer's problem.

Related, in terms of zapping the other channels of sales that might have saved them, was a total internal sales focus on enterprise sales. If you weren't making enterprise sized purchases, several million dollars worth, you were sent to a 3rd party like a VAR, but they typically wouldn't give you the time of day, they certainly were difficult to deal with when I worked with companies that bought one or two Sun workstations. Adding a web site that allowed direct purchases mostly added salt to the wound.

What this meant was that startups that needed more Sun kit than they could charge to a credit card on that web site were forced to go with Dell, who's machines weren't as good, but had the strange property of being willing and able to sell them (HP's sales operation was too screwed up at the time to be a big player here). So a bunch of companies learned how to make do with Dell servers, and the successful ones never looked at Sun when they got big enough. So they missed out on the big post dot.com sales.

A final straw when they started struggling was reducing the quality of their systems, e.g. saving one chip on a motherboard by using a chipset Marvel controller instead of the well tested and trusted Intel one. And I read that Joylent stopped buying Sun hardware when they started silently changing the lights out hardware they were shipping in their systems (IPMI or the like).

> their ruining their reputation in enterprise space when they messed up an off-CPU chip cache design

Yeah, that one was a huge. Sourcing those cache chips from IBM, and just trusting them not to knowingly supply ones that emit radioactive radiation was such an amateur move.

http://www.computerworld.com/article/2585216/computer-hardwa... http://web.archive.org/web/20020202013942/http://www.compute...

It's inevitable, the only question is at what rate the materials in your chips will emit them, and what are the consequences. Smart companies spend a little extra for parity/EEC, they don't blame their customers for machine room environmental non-optimality unless they're really sure that's the problem. He also didn't deny the "sign the NDA if you want us to (pretend to) address the problem".

Bottom line, he and Sun didn't own the problem, didn't take care of their customers until too much bad PR was generated, and again dealt with it at least as much as a PR problem for Sun than as a problem for their customers.

A bad idea when there are other companies out there offering better enterprise level stuff. Or as a DEC book put it, looking back on the frothy '70s minicomputer market, the companies that won did an adequate job of everything, didn't drop the ball on any of the essential things like documentation and support. Certainly no one bought DEC equipment for their disk drives....

I've never heard that the 386i was successful by any metric. Most big Sun clients had a couple sitting in a storeroom because they would basically give you a couple when you renewed maintenance to generate interest. I was at a Sun reseller at the time, and I can't recall perhaps a single significant 386i deal in the whole run of the product. But that may have just been my corner of the world and somewhere else there was a huge demand for an otherwise unremarkable i386 PC that cost 4 times more than what you could get from everyone else in the market.

It would be your corner of the world, I think. As I recall, it was a big hit in finance, where they wanted to run existing MS-DOS stuff, including I'm sure Lotus 1-2-3, and had demands beyond what MS-DOS could (easily) do. And of course it preceded Windows 3.0 by a couple of years.

I worked for an early cable high speed internet provider in the mid-late 1990s. Amongst other things, we used SGIs to power the proxy-cache servers at the edge of our customer networks. In some places, we even had SGIs on racks down in manholes where some of the headends were located. I also remember some SGIs up at our northernmost outpost in, I believe, Fort McMurray, Alberta. One of them had a hardware failure but they were under service contract and some poor SGI tech (or whatever company they had subcontracted to) had to make a day-long drive up to fix it.

Not quite. There were no 64-bit PC's back then or even a quality PC workstation Unix.

Your alternatives were things like HP workstations or a DEC AlphaStation which were fairly cheap. Eventually apps started showing up on NT but they were garbage for quite a while.

Once Linux was available, along with the Pentium, we never bought another big-name Unix server except for some high-end database stuff.

That's not really true at any level. There were decent i386 SVR3 releases before 1990 and some very nice SVR4 releases soon after (particularly, Dell). I deployed a lot of them as an alternative to workstations. HP and DEC were not cheap in that timeframe, though you could argue they were more capable. And while not a "PC", a 64-bit desktop showed up in the form of the R4000 in 1991.

In 1997 Irix was still vastly better than Linux on the development front. Mips Pro C++ compilers could actually handle the STL and had useful error messages. Irix came with debuggers that you could use. If you were doing perl however there was no benefit.

Things changed very rapidly though. By 1999 Irix was languishing badly.

All the machines in the picture were running FreeBSD which was our development platform. We were doing kernel & networking development work and BSD was much better documented because of the second Stevens book.

Userspace was a lot of crappy C++ code using the bits of the STL which worked across gcc / whatever Sun's compiler was called / Windows VC++ (which was not a large intersection at that time).

I was doing 3d graphics. We had broadcast television on analog at the time with engineers that did not know how to use computers running the show. The cost of an O2 SGI workstation was £10k or so - you could buy a broadcast spec clock for that sort of money. Digital tape machines - Digibeta - cost £40k, plus you would always need more than one.

That whole industry went race to the bottom. Internally you also had people not willing to learn the new PC tools. So there was a while before the PC inertia got going. On SGI machines doing CPU locked tasks, e.g. frame rendering from a render queue there was a simple cost equation with using PCs, off the shelf non SCSI disks were another saving, as was RAM, a tenth the price on the PC. Memory for SGI machines would need board level authorisation!

So with many SGI use cases the CPUs would be doing very little. The graphics pipeline would be doing all the real work and you would hit your polygon limit in there. Only if you went CPU intensive - software rendering - would you Max out the CPU.

Most SGI use cases were in capital intensive industries where proper money was paid for proper professional kit. Those 6K£ D1 monitors? Yes you could have a wall of TVs instead but that would not happen. So white box PCs took a while.

SGI computers had value, so people paid $$ for that value that they couldn't get anywhere else. For those people, it wasn't overpriced. Like many companies that do really well, they eventually squandered that edge.

To answer you question... why did people used Suns for several reasons. These reasons are not in any quantitative order.

1. Religion - many people watched people smarter than them achieve success with Sun. Given their growing market size, they had a bigger religion. So whenever somebody tried something different and failed, it reinforced the religion. This lasted for quite some time (til early 2000's). Once the internet hit (1995,1996), just believing in this religion was a successful move.

2. Educational market. Sun gave educational institutions discounts. Students and Professors liked Suns and that created momentum when they left. Remember, even to this day, some people will stick with a system just because they know it better.

3. Sun systems were typically more data center friendly than SGI, HP, PCs, and especially SGI. They had serial consoles with OpenBOOT for lights-out-management - this was huge. They also stacked up nicely and fit into racks easily. If you knew what you were doing, you could do everything remote.

4. Many solutions had Sun picked as their configuration. For example, if you wanted to run Oracle or Veritas, you must buy Sun. It was not uncommon to see Oracle, Veritas, Sun, and Cisco all over the place in any data center in the valley. That was the template.

5. Much lower risk for hardware integration at scale. The hardware was targeted for Unix and scale. You knew that you could put 8 gigs in that Sun. You knew that the EMC RAID was going to work. So, spend the $, and move on to the problem. This was not the case with PCs. Sometimes that Intel motherboard, or hard-drive, or mis-spec'd RAM burned you on PC hardware. This could cost a lot of time and $$.

6. OS Maturity - they had just enough of an edge on the competition here. With a small amount of work, you could have something that could reboot quickly and stay up. The monitoring and tracing tools were much better. With skilled people, you could really make these systems work really well. This was in a time when many kernels had bugs and couldn't stay up for longer than 24-48 hours under normal load. SGI Irix was a great design, but many skilled people still had issues. I remember HPUX requiring weekly patch sets. The internet really exposed how bad these OSs were for server usage.

7. Java - once Java took off, that created a halo effect for Sun hardware.

This combination created an enormous wave of $$$ to go to Sun.

Sun had its warts for sure, but SGI, HP, DEC, etc all blew it compared to Sun.

BTW, I was always more of a NeXT guy (we all had our religions), but I respected and liked Sun solutions. However, once I saw what a Pentium Pro combined with the maturity of FreeBSD (1997), I knew the workstation guys were in serious trouble.

I've been doing 3D and film/video at the time on SGIs. While true that CPUs gained in performance on other platforms, IRIX machines were 64-bit, had great 3D accelerators (OpenGL was SGI's in fact) and - most importantly - they had bus bandwidth which was very important for large amount of data image processing required. Also, proprietary I/O was available for video and film which was only starting to come out for PCs (FAST and TARGA cards were getting cheap). It's all funny considering there was a void in the market. SGI (and discreet for video and A|W and Softimage for 3D) was great, if you could afford it, for high end broadcast and film and PC was nearly usable for video, but not for film still. Mac was great for video editing, but not for 3D... etc. Amiga was great for everything for video (not film), but it was, at that point, dead for several years.

SGI got lazy. It had the market for video and audio in its pocket. mostly because of what you said: I/O bandwidth! even if the RISC cpu was garbage (despite what the uninformed article says), moving from PC to SGI was like adding a SSD to your machine! it was night and day when working with video.

Then microsoft come around and used its obnoxius tactics to take hold of firms such as Autodesk and others. which all dropped unix support to focus on windows. that was the nineties, when CGI quit being a small niche market to be the holywood standard, i.e. a very profitable niche market.

apple scrambled at the last minute and got some companies too. the most important at the time was shake or tremor (never remember what was the product and what was the company) and quickly reworked the products to only use quicktime video exclusively, which obviously killed it down the line. because quicktime was, is and always will be garbage. i think at that time they got what is now their video suit btw. but i'm not sure.

anyway, SGI ended up with no excuse for people to pay the premium price, not even on video editing. So everyone moved to PCs or macs. most of the custom software started to be written for linux (e.g. renderman) which SGI tried to corner by moving its server line to linux. but after itel and nvidia crushed them in performance /watt and /price, they just died for good.

tl;dr; there is nothing special about MIPS. i give that assembly for mips is much more comfortable than risc. but meh. you always have to code for the lower common denominator anyway.

You're thinking Shake by NothingReal (they had a product called Tremor also, but it wasn't released, hah). Apple did what Apple does. They bought the company and stopped development for linux and windows and stagnated development for Mac also. This made a lot of people very angry.

Mac had the advantage, at the time, to be the only host platform for AVID editing products. AVID was the only game in town back then if you were doing anything remotely serious. Apple made AVID angry by entering their own Final Cut into the arena. This also made Adobe angry because Final Cut main dev was Premiere originator (Premiere was still shite back then). This made a paradigm shift for Apple because it was known as 'the' Photoshop and Video (editing) platform. AVID prepared their suite for Windows and Adobe concentrated more on Windows as well. At the same time (more or less) Microsoft bought Softimage (the other big 3D name, along with A|W) and made them do a Softimage|3D for Windows NT (it was IRIX only) to showcase their new OS. Lightwave came to NT as well and Autodesk bought discreet (IRIX ediitng and compositing powerhouse) and joined it up with their own kinetix group (3d studio and 3dsmax) - and Autodesk is more or less Windows only shop. This nexus of bad things finally came together against Apple years after, when dev cycles were a few gears ahead, and suddenly Mac was no longer 'the' graphics platform. Windows was, and to an extent Linux (in larger shops/systems). In parallel, Microsoft shafted SGI as well when they teamed up to make a successor for OpenGL called Fahrenheit, but opted out and bought/made DirectStuff.

Fun times.

I see SGI machines and software with nostalgia now, but when I remember those outrageous prices it makes no sense for them to be alive with that kind of politics. For example, in Maya (their 'new' software, made by their subsidiary A|W) you had two additional modules (Artisan and OptiFX) which were in five figures each. These two modules are standard offerings in that software now that aren't even mentioned/used all that much. Everything was a daylight robbery with them. EVERYTHING. Nice machines, all well - but robbery.

On the other hand you had Amigas which were capable machines with software and it ultimately failed as well for different reasons. One was a high end robbery, the other was - they didn't know what they wanted with it I guess.

One thing I miss (and I don't at the same time) is that we had a heterogeneous mix of hardware platforms and OS' back then. Now it's only a heterogeneous mix of OS', and even that is questionable. ARM is on the horizon, but I doubt we will see that kind of a wonderful mess we had back then on the desktop side.

right you are.

the kinetix line was mostly for 3d. for video editing all of autodesk/discreet line was based on flamable names. flint, ignition, flame, spark, etc... that's why the effort was called Fahrenheit. if i recall correctly.

and couldn't agree more with the high end robbery. I mostly worked with those boxes while contracting in my teens to help university departments. most of them had SGI boxes just because they had to burn they yearly budget to not get a cut next year! ...nowadays they buy mac pros. which tells a lot about the state of apple out of portables. and which market jobs had in mind with Next.

Amiga's was being used on Babylon 5 first seasons. Also, I have Amiga A1200 with animations for TV and a movie across the whole 90's. So not was too dead.

Yeah, PCs were cheaper than Unix workstations in 1997, but a Ford Pinto is cheaper than a Ferrari...

Linux had serious shortcomings in 1997. It was OK as a web server or mail server or for home-grown number crunching software, but the NFS implementation was full of bugs, and none of the industry-specific commercial software had been ported. Windows 95 was really, really bad - it crashed all the time, barely had a functioning network stack and had no multiuser functionality.

BMCs and KVMoIP hadn't been invented, so for PC-based servers you needed a keyboard, monitor and set of hands for each rack. SATA hadn't been invented either and with PATA you were limited to two drives per controller, no external drives and it was always a crapshoot whether any particular permutation of drives and mobo would actually work. Expansion slots were ISA or mixed ISA/PCI so you got to have fun with IRQ, DMA and I/O address settings. The mechanical design of PC enclosures was awful to mediocre. Sun mechanical design was a little better (retaining springs, etc) and SGI mechanical design was much better (tool free servicing). ECC and parity RAM were exotic on PCs but ubiquitous on Unix workstations, and the operating system actually did the right thing after a parity error rather than either ignoring it or freezing up.

Also, regardless of the raw CPU performance, Suns running SunOS seemed more responsive under heavy load than PCs running Linux. I'm not sure why, and I don't have numbers to back it up, but that's how it seemed at the time.

PC BIOS is still worse today than Sun OpenFirmware was in the 90s, but we put up with it because it's good enough and for many applications there's no other reasonable game in town.

There was SCSi, EISA, VESA bus, Unixes for PCs, windows NT, OS/2 and NetWare.

Indeed. In 1995 I think it was I bought my first "custom" machine, an EISA dual Pentium with a 2 GB HP SCSI hard drive running NT.

> R3000 sold extremely well (over 1 million units were produced).

Although I know it's not directly comparable, I was at an ST event recently, and the representative said they were making one million STM32 ARM microcontrollers per day.

And even that is only a fraction of the 10 billion ARM cores made each year (source: http://www.anandtech.com/show/7909/arm-partners-ship-50-bill...)

I know, we've shipped about 800 million MIPS CPUs in the last year alone.

But for that time in history, a million processors was a big milestone. You also have to remember that in 1988 MIPS was a three-year old company battling established vendors.

> You also have to remember that in 1988 MIPS was a three-year old company battling established vendors.

And 10 years after that MIPS has become an embittered outcast, trying to sue potential partners into the ground [0].

Luckily, after yet another 10 years MIPS has grown up and started acting somewhat sensibly [1].

[0] https://en.wikipedia.org/wiki/Lexra [1] https://en.wikipedia.org/wiki/Loongson#MIPS_patent_issues

Both MIPS and ARM sell CPU IP therefore I'm sure anyone here can dig up several examples of patent infringements that ended up in court.

I think what sets us apart from the competition is the fact that a MIPS CPU is now completely open and free for university use, including production-quality Verilog code and tools from Xilinx.

http://www.anandtech.com/show/9194/imagination-announces-fre...

It takes quite some big brass b^H^H^H^H^H^H^H PR skills to downplay Lexra disaster that (together with market crash and other events around year 2000) costed MIPS itself quite dearly in both money directly and more importantly in market position. Need I remind you that was right around the time of ARM's spectacular rise: Quite literally MIPS Technologies being too busy with this litigation is the only good explanation why it was blind to the real threat: ARM7TDMI taking embedded/telecom market by storm. Only burgeoning xDSL market (and good availability of established binary-only MIPS codebase for it) that saved MIPS in early 2000s.

Not to mention quite egregious fact-twisting I see in your reply on two counts: First, that wasn't really a "patent infringement" - first lawsuit was about trademark infringement - which Lexra arguably was guilty of, but quickly backpedaled, at which point case felt apart. Then MIPS, unhappy with that outcome, launched second, even bigger lawsuit that was technically about patent infringement, but Lexra never actually used the patent in question, and specifically marketed their design as free from lwl/r and swl/r instructions, making the whole thing a theater of absurd going on for years where plaintiff was alleging that Lexra was was yes, not implenting those instructions, but somehow facilitating their clients' emulating those. Hardly that can be called "ended up in court". Whopsie-daisy... "ended up", right...

Providing official Verilog code for an older MIPS core is certainly a massive feat and deserve an applause, but then again there are plenty of other HDL implementations of CPU cores around, MIPS ISA cores included [0]. And patent law have a safe haven for academia [1], so "completely open and free for university use" is just hot air, sorry. Still, it's "official", so there's value in that, of course.

[0] http://opencores.org/project,ion [1] https://en.wikipedia.org/wiki/Research_exemption

The recently released official MIPS core has a standard bus interface so it will be much easier to extend it in student projects than any of the OpenCores designs.

Quite literally MIPS Technologies being too busy with this litigation is the only good explanation why it was blind to the real threat: ARM7TDMI taking embedded/telecom market by storm.

Case in point: in the first half of 2001 I worked at Lucent on a board that had something like 288 or more dial up modem channels (this was once a part of Ascend Communications). It had a bunch of specialized Analog Devices chips doing the heavy lifting, a bunch of ARM processors controlling them ... and one MIPS processor doing system housekeeping.

I wasn't around to witness the Lexra situation so I can't/won't contradict your observations.

However, I think there are a few other reasons why MIPS did not ride the wave of mobile like ARM did.

First of all, MIPS Technologies acquired mixed-signal design house Chipidea for $147m and then sold it to Synopsys for $22m after a rocky two-year integration process.

Secondly, MIPS management focused on markets like networking and home entertainment (set-top boxes, digital TVs, etc.) which did not enjoy the explosive growth of mobile.

To address the second part of your comment about MIPSfpga, this is not "an older core". It is a current-generation CPU capable of running Linux. It is used today in the Microchip PIC32 and the Samsung Artik 1 MCUs - two products that were released within the last year. Furthermore, most of these open cores implement MIPS III or IV architectures from two decades ago whereas MIPSfpga is MIPS32 Release 3. In addition, MIPSfpga implements industry-standard interfaces which make the core much easier to use on an FPGA.

I guess that's one reason why we need a completely patent-free ISA.

http://www.eecs.berkeley.edu/Pubs/TechRpts/2014/EECS-2014-14...

My housemate and I used to joke about CPUs someday being so cheap that they would "fall out of cereal boxes as the prize."

The STM8, a chip about 5X more powerful than the first computer I owned, has a variant that is under 17 cents in quantity. Doesn't quite qualify as mommy-buy-me-that-cereal prize material, but it's damn close.

The number of 8-bit CPUs sold every year must dwarf the number of 32-bit ones, yes?

If found it a bit difficult to get accurate sales numbers of how all 32-bit embedded processors compare to all 8-bit embedded processors. However, there are still tens of billions of Intel 8051 (introduced in 1980) derived processors/cores/IP shipped every year, so it's safe to assume the whole market is still vast.

I can tell you that ARM and MIPS ship about 5bn 32-bit MCUs per year combined. I can't give you a statistic for 8- or 16-bit MCUs though.

Yeah...I get the same handwavey number for ARM & MIPS; its a crazy successful market. And it's clear that 32-bit is the future, for some value of future.

That said, trying just to understand how many 8051 derivatives is almost impossible. There are many dozens of vendors who supply chips, cores & other embedded IP, many of whom don't even call it 8051 any more. There's probably a few tens of them within 100 feet of you. Much less the 20 or so other significant 8-bit families.

It's a shame that hard data is only available from high dollar market research firms, but I suppose they did the legwork and deserve something for the work.

Wow, what does an HP LJ4000 laser printer need a processor that's also used in an SGI workstation for?

I looked into the Wikipedia article and although it's a big printer, I don't really understand how does it compare to a workstation's performance.

Probably Postscript and PCL rendering in firmware. Requires a lot of memory and CPU. Current printers use custom ASICs with specialized rendering hardware.

Rendering to a piece of paper based on a Postscript/PCL/etc definition is similar to rendering to a screen. Both a video subsystem and a printer want to do that fast.

Until a few years ago, I owned a LJ4000, and it still had to stop and think for a bit if you were printing a picture or something complicated. Earlier laser printers could be incredibly slow - an Apple LaserWriter took about 30 minutes to print one page of a 'zine' we were were working on.

I remember the LaserJet 4000 and the 4550 (color beast, the startup sequence sounded amazing with all those motors and the always-on fans). Worked flawlessly for hundreds of thousands of pages.

It had expandable RAM, and you could send files to the print queue inside the printer via LPT or Ethernet, and it did PostScript rendering inside, which was quite convenient when printing lots of stuff (reduced the amount of data you had to send, and no issues with interrupted printing). I guess that's why it needed its own processor.

i can't stop laughing at that article.

> Even though these days MIPS is best defined by the low-power [...] all desktop and server processors released since the 1990s have borrowed heavily from many of the energy-saving RISC philosophies introduced three decades ago.</blockquote>

maybe the CPU itself was power efficient, but the actual machines from SGI were far from it.

for example, nekochan is the de-facto forum for sgi/irix for a long time. here is someone begging to give away a onyx2 (top of the line, latest model, R12k sgi powerhorse) because they moved to a new office and didn't have a wimpy 250V 60A outlet laying around on the new place

http://forums.nekochan.net/viewtopic.php?t=16720600

> [...] our move to a new office. Since we have moved here we don't have enough power to test it as it requires 60 amps at 250v (2 30 amp 230v connectors).[...] The current offer is 250$ and pickup.</blockquote> --- 2009!

I wonder how many other CPUs are named after an extremely useful thing they don't have.

I think the original intention of the Stanford group was to highlight pipelining, so the "without interlocked" aspect should be seen in that context and not interpreted literally.

Until the late 1980s pipelining had been drastically underexploited by both CISC and RISC architectures.

No R8000? Odd, multi-chip "FP performance at all costs" variant used in some of the SGI Power series. Made SGI a legitimate supercomputer vendor for a couple of years.

I left out R8000 because it was used only for a short period of time and in a limited range of systems.

I get it, but it really did make MIPS an undisputed scientific computing beast. Probably I just have a soft spot for it as it was the chip that made me really understand that we'd be able to do what was once considered supercomputer-only tasks on something that was from a budget perspective a departmental or even workstation platform.

Reply on news.ycombinator.com