Slashdot Mirror


CEO of Centaur Discusses x86 Strategy and Linux

An anonymous reader writes "This fascinating interview with Glenn Henry, founder of VIA processor subsidiary Centaur Technology, discusses the founding of Centaur, its strategy and products, and why Linux is fundamental to his company's success. Additional topics covered include: how to produce an x86 clone with a few million dollars and a few dozen engineers; the embedded x86 market, and how it compares to the traditional ARM and MIPS based embedded market; why Centaur doesn't compete with AMD and Intel so much as enable x86 to reach new markets; how Linux is enabling greater hardware functionality; the urgent need for pervasive security -- and much more!"

17 of 135 comments (clear)

  1. Wondering by swordboy · · Score: 4, Interesting

    I've been wondering why some company like this doesn't create a "network appliance" specification for all of us to hack on. It would be nice if I could just go buy a Netgear router and roll my own Linux installation. I purchased a Toshiba Magnia SG10 some time ago when they were a couple hundred bucks during the end-of-life period. For a 566mhz Celeron with an honest-to-goodness hard drive and switch on the back, it was hard to go wrong. I immediately wiped the stock Linux OS and rolled FreeBSD on there.

    Wouldn't it be MORE profitable for companies like the aforementioned Netgear to do this? What am I not seeing? Centaur: help us out!

    --

    Life is the leading cause of death in America.
  2. Is x86 the best chip to use by millahtime · · Score: 4, Interesting

    Is an x86 or clone really the best chip to take to markey this way? Linux will run on other processors and the x86 isnt' the best archicecture. There are processors that are more efficient, use less power and can run linux.

    Although, I'll admit some of those embedded boards that I have seen are pretty cool and easy to use.



    P.S. I know I can't spell. That's why I'm not an english teacher.

    1. Re:Is x86 the best chip to use by squiggleslash · · Score: 3, Interesting
      It's a nice idea, but there are still a hell of a lot of binary-only packages for Linux (and GNU/Linux) varying from display device drivers to video decoders (such as Real.) Unless your chip is likely to capture a significant portion of the market, it's not terribly likely that many of those binary packages will be ported.

      On top of which, as long as it runs ix86, platforms with the chip will not be limited to Linux.

      --
      You are not alone. This is not normal. None of this is normal.
    2. Re:Is x86 the best chip to use by iguana · · Score: 5, Interesting

      As an embedded Linux developer moving from x86 to ARM, there are good reasons to stick with embedded x86.

      - My Linux development PC runs x86 so stuff compiles/runs on my development machine will tftp over to my embedded board and run immediately.

      - The GNU x86 toolchain is WAY easier to use than the cross GCC tools. All GNU (Linux kernel and gcc especially) projects target x86 first and everyone plays catchup. Bugs are found/fixed in x86 first. Features work on x86 first.

      Reasons to move from x86 to ARM.

      - ARM is designed as an embedded CPU and x86 is designed as a desktop CPU. So the ARM has a very simple memory map, you don't have to worry about working around 15 year old cruft (why the foo doesn't a modern x86 boot straight to protected mode?).

      - It's hard to find an x86, even one ostensibly designed for the embedded market, without the "kitchen sink" problem. Our product with >500 parts with an SiS x86 dropped to 300 with an Atmel AT91 ARM. Most of the passive parts on the SiS board were to turn OFF crap we didn't need (IDE controller, SD controller, video controller and so forth).

      - ARM is cheaper. ARM is smaller. ARM uses less power. ARM is cooler (temperature). ARM is cooler (it's a very nifty chip design).

      Just my opinions, lightly flavored with experience.

    3. Re:Is x86 the best chip to use by iguana · · Score: 4, Interesting
      The embedded world is still largely about ASM routines and tight little loops that dont leave the cache, etc, etc.. You can write and test your fancy-shamcy hand-assembled routine at your leisure on your desktop computer....

      Not really. There isn't much assembly necessary on a modern embedded platform. Even the little 4- and 8-bit processors have specialized C compilers available. Assembly is necessary for startup but that's about it; then we just jump to main(). I think Windows game programmers write more assembly than most embedded systems folks do.

      Embedded systems development is just like any other development. How can I get the best product out as quickly as I can? ASM doesn't get me there quickly enough.

      That's why using embedded x86 can be a good fit. It may not be the best technical solution, but we'll get to market quickly because the tools and available leveragable source is so good.

    4. Re:Is x86 the best chip to use by Anonymous Coward · · Score: 5, Interesting

      I honestly fell in love with the PPC in college .. It's a lovely architecture

      PPC is nicer than X86 but it's hardly a "lovely architecture." In fact, it's one of the uglier RISC designs.

      - Backwards bit numbering scheme. The MSB is referred to as bit 0. Normally, it's just a matter of convention and makes no difference but there is some inconsistency in the ISA about this. For example, shift counts are stored in the 5 LSBs but indirect segment register indices appear in the upper 5 bits (probably because of the MSB convention.)

      - Some really complex instructions which increase processor complexity. Rotate-with-masked-insertion? That's not RISC. IBM's POWER chips actually need to break down some of the instructions into smaller ops, similar to X86.

      - Flags. It's nice that most flag computations are optional and are only done if explicitly requested, but flags don't belong in a RISC architecture. The way PPC implements them is even more complicated than on most CISC machines: A flag register with 8 4-bit CC fields and additional flags in a separate XER register! (??)

      - A bizarre MMU. The paging scheme isn't as efficient as other architectures, even including X86. Apparently this is IBM legacy cruft from a previous architecture (possibly S360, can't remember.)

      As far as RISC architectures go, Alpha was probably the best by far, and the actual implementation was great starting with the EV6.

  3. I wonder by foidulus · · Score: 4, Interesting

    if they could take this technology and turn it into a "portable pc gaming console" of sorts(of similiar size and shape to the gba) In the article it states that they can run at 533 MHz fanless with a worst case power consumption of 2.5W. Wouldn't it be neat to create a gaming console(kind of like the phantom pc gaming console, only portable) for this with flash cards being the "cartridge"
    Though licensing for the abandon-ware would be a pain. As would trying to standardize the input across a large number of games. But still, it would be friggin' cool to play leisure suit larry while you are bored in class!

    1. Re:I wonder by foidulus · · Score: 3, Interesting

      Yeah, but with the small screen you are limited to on a portable, do you really want anything fancy? I'm not sure I could play unreal tournament on a small screen. There are a ton of old games that you cannot find anymore(Leisure Suit Larry, a whole load of freeware games that came on this cd I bought 10 years ago) that were a lot of fun, but not very portable. Plus, a lot of times you had to mess around with "low memory" and whatnot just to get them to run. It would be nice to have a console that just "worked" with all those games.
      It's too bad that the companies who made those games either:
      a) don't exist anymore or
      b) don't want to open source the games even though there is almost no chance of the games bringing any more revenue.

  4. Reminded me Transmeta by nomad63 · · Score: 2, Interesting

    Well, not exactly the same idea maybe but the story is the same as transmeta to me. One good upside to transmeta, they did not start with linux on mind.
    I am assuming their product did not make too much inroads into the industry otherwise I would have expected to hear about their products before this time and from 95 till 04, it is anawful long time to stay in hiatus.

    --

    __________
    The more I know people, the more I love animals
  5. Open source designs? by Trigun · · Score: 3, Interesting

    The biggest problem with most hardware is that it is exotic. I'm not talking about PC's, and routers have become cheaper with micto-atx mobos with dual ethernet onboard. Laptops and PDA's are where there seem to be a lack of standards. There are no displays which can be plugged into your PC without special driver circuits. Why not create a LCD that has a standard interface that is compatible with PC hardware now?

    If you standardize the equipment, then you will drive prices down, and you won't be stuck throwing out a PDA that has a broken screen, or has eol'd. Just upgrade it.

    1. Re:Open source designs? by Trigun · · Score: 3, Interesting

      Flat cables aren't that difficult to work with. Proprietary flat cables are a pain to work with.

      As an example, I have an IBM thinkpad 380Z with a broken lcd cable. Simple enough, right? Replace the cable. They're mostly standard connectors. Except the cable has a variety of surface mounts on it, and can only be replaced with a $125 replacement cable. If it didn't have the surface mounts, I could use an old IDE cable to replace it. If it had a driver circuit that wasn't built onto the mainboard, I'd have a vga display to play around with.

      All I've got now is a broken laptop.

  6. centar by millahtime · · Score: 3, Interesting

    Centaur is positioned to tap the largest market in the industry- the sub-$400 PC market. Right now Centaur is designing a processor that will run over 2 GHZ.

    A 2 GHz machine that is x86 compatible but will it have all those nifty other features that windows can use to speed it up? If not, it would be much slower than AMD and Intel.

    Although, the thought of being able to use a chip like that for some embedded app would be pretty neat and pretty powerful because you don't need all the extra stuff that windows uses.

  7. Out of skew by b0lt · · Score: 3, Interesting
    Additional topics covered include: how to produce an x86 clone with a few million dollars and a few dozen engineers;
    Didn't Intel CREATE the x86 with less than a million dollars and a few dozen engineers?
    --
    got sig?
  8. I Wish for RISC by RAMMS+EIN · · Score: 2, Interesting

    I think VIAs mini-ITX range of products are really cool. However, I wish there was an equivalent based on ARM, MIPS, PPC (seemingly any kind of RISC) CPUs. Those architectures always seem to beat an x86 of similare performance when low power, low heat production, low cost, small size, etc. are concerned. I care about these things. I don't need a super fast computer, so I don't want to pay more for a chip with a messy architecture that needs more power, needs cooling, and has a large die.

    Sadly, the nice, small boards and CPUs that I would like to have are hard to come by, and you pay a price penalty for that. Then, the next best thing is a sort of mainstream x86 line that aims for the same goals, and that's exactly what Centaur is doing (and Transmeta, although they seem to have failed to satisfy even themselves).

    So, if anyone knows where to get cheap RISC systems (a few hundred euros tops) in the Netherlands, please tell me.

    --
    Please correct me if I got my facts wrong.
  9. Great low power (thus heat & noise) processors by tji · · Score: 2, Interesting

    I am a long time user of their C3 processors.. His quote about the current state of the product illustrates my main reasons for using it:

    The product we're shipping now, the C5P has a top speed of 1.4 to 1.5GHz, today, but the sweet spot is 1GHz. We have a fanless version at 1GHz. We also sell all the way down to 533 or even 400MHz, for low-power applications.

    To give you an idea about the 1GHz version we're selling today, the worst case power -- not "typical" or "average" power, which other people talk about -- our worst case power is 7 watts, which is low enough to do fanless at 1GHz [story], and no one else can do that.


    To do a fanless CPU in a small case, you really need to be under the 10 Watt range. Their CPU's do this nicely. Compare that to 80W+ for current Intel and AMD workstation processors. ( Intel's Pentium M has good power spec's, but it is very hard to find chips & boards for end user purchase. Most Pentium M boards are intended for embedded or industrial use, and are priced for OEM quantities). Add a 2.5" hard drive (at ~ 2.5 Watts vs. 15 Watts for a 3.5" drive) and you have a nice low power Linux server, which takes up very little space and can run almost silently.

    I have been using an 800MHz C3 for about three years now. I run it fanless, with a big heat sink in a medium sized case. It has been completely reliable, and plenty fast for my DSL Linux services (WWW, SMTP, FTP, VPN, DNS, NTP, etc.) + LAN SMB services.

  10. Future processor innovations? by tji · · Score: 2, Interesting

    At the end of the interview, he mentions the crypto hardware they've put in recent processors.

    six months after we first started shipping our product with encryption in it [story], we have three or four operating systems, including Linux, OpenBSD, and FreeBSD

    This is a really great capability.. hardware random number generation, and ridiculously fast AES crypto (VIA claimed 15Gbps AES. That's probably for on-cache data. But, it's screaming fast anyway).

    Then, he give a little teaser about future CPUs:

    Our next processor -- I haven't ever told anyone, so I won't say what it is -- but our next processor has even more things in it that I think will be just as quickly adopted by the open source software world, and provide even more value.

    I wonder what this will be, more crypto - like public key accel, or a new direction? As an HTPC user, I would like to see some better multimedia capabilities. The MMX/SSE stuff is nice, but it doesn't cut it for the heavy lifting needed for HDTV MPEG2 processing, or WMV HD processing.

  11. Not Midrange performance by Nazmun · · Score: 2, Interesting

    I keep on having to point this out on Slashdot as many people don't realize how bad the performance of a via c3 is. The 1.0 ghz system is often times twice as slow as a 667 mhz Celeron in certain tasks. I'd label this as an ultra low range processor not midrange. I've seen simimlar benchmarks with a 400 MHZ p3 beat it handily. Benchies: http://www.tomshardware.com/cpu/20020605/c3-07.htm l

    --
    Hmmm... Pie...