8 ms·

Great MIPS chips of the past 30 years

51 points by alexvoica 11y ago

niklasni1 11y ago
> 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.
- rwmj 11y ago
  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-billion-chips-since-1991-where-did-they-go http://www.anandtech.com/show/7909/arm-partners-ship-50-bill...)
- alexvoica 11y ago
  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.
  - axeldeboer 11y ago
    > 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 https://en.wikipedia.org/wiki/Lexra [1] https://en.wikipedia.org/wiki/Loongson#MIPS_patent_issues https://en.wikipedia.org/wiki/Loongson#MIPS_patent_issues
    - alexvoica 11y ago
      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-free-mipsfpga-design-for-academia http://www.anandtech.com/show/9194/imagination-announces-fre...
      - axeldeboer 11y ago
        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 http://opencores.org/project,ion [1] https://en.wikipedia.org/wiki/Research_exemption https://en.wikipedia.org/wiki/Research_exemption
    - hajile 11y ago
      I guess that's one reason why we need a completely patent-free ISA. http://www.eecs.berkeley.edu/Pubs/TechRpts/2014/EECS-2014-146.pdf http://www.eecs.berkeley.edu/Pubs/TechRpts/2014/EECS-2014-14...
- kabdib 11y ago
  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?
  - kjs3 11y ago
    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.
    - alexvoica 11y ago
      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.
      - kjs3 11y ago
        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.
rwmj 11y ago
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 http://oirase.annexia.org/tmp/testbed.jpg
- agumonkey 11y ago
  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.
- hga 11y ago
  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.
saganus 11y ago
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.
- linuxlizard 11y ago
  Probably Postscript and PCL rendering in firmware. Requires a lot of memory and CPU. Current printers use custom ASICs with specialized rendering hardware.
- kjs3 11y ago
  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.
- flomo 11y ago
  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.
- prodmerc 11y ago
  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.
kjs3 11y ago
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.
- alexvoica 11y ago
  I left out R8000 because it was used only for a short period of time and in a limited range of systems.
  - kjs3 11y ago
    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.
to3m 11y ago
I wonder how many other CPUs are named after an extremely useful thing they don't have.
- alexvoica 11y ago
  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.
gcb0 11y ago
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 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!

51 points by alexvoica 11y ago

> 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.

rwmj 11y ago
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-billion-chips-since-1991-where-did-they-go http://www.anandtech.com/show/7909/arm-partners-ship-50-bill...)
alexvoica 11y ago
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.
- axeldeboer 11y ago
  > 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 https://en.wikipedia.org/wiki/Lexra [1] https://en.wikipedia.org/wiki/Loongson#MIPS_patent_issues https://en.wikipedia.org/wiki/Loongson#MIPS_patent_issues
  - alexvoica 11y ago
    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-free-mipsfpga-design-for-academia http://www.anandtech.com/show/9194/imagination-announces-fre...
    - axeldeboer 11y ago
      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 http://opencores.org/project,ion [1] https://en.wikipedia.org/wiki/Research_exemption https://en.wikipedia.org/wiki/Research_exemption
  - hajile 11y ago
    I guess that's one reason why we need a completely patent-free ISA. http://www.eecs.berkeley.edu/Pubs/TechRpts/2014/EECS-2014-146.pdf http://www.eecs.berkeley.edu/Pubs/TechRpts/2014/EECS-2014-14...
kabdib 11y ago
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?
- kjs3 11y ago
  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.
  - alexvoica 11y ago
    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.
    - kjs3 11y ago
      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.

rwmj 11y ago

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 http://oirase.annexia.org/tmp/testbed.jpg

agumonkey 11y ago
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.
hga 11y ago
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.

saganus 11y ago

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.

linuxlizard 11y ago
Probably Postscript and PCL rendering in firmware. Requires a lot of memory and CPU. Current printers use custom ASICs with specialized rendering hardware.
kjs3 11y ago
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.
flomo 11y ago
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.
prodmerc 11y ago
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.

kjs3 11y ago

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.

alexvoica 11y ago
I left out R8000 because it was used only for a short period of time and in a limited range of systems.
- kjs3 11y ago
  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.

to3m 11y ago

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

alexvoica 11y ago
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.

gcb0 11y ago

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 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!