A Sub-$1000, Non-X86 Motherboard

If you’re building a computer, your options are nearly limitless. You can get a motherboard with red LEDs, with blue LEDs, green LEDs, or if you’re feeling spendy, RGB LEDs. You can get custom-milled heat spreaders in any shape you want, as long as it’s angular and screams ‘gamer’. If you want a motherboard that doesn’t use x86 — either AMD or Intel — you’re kind of out of luck. Either it doesn’t exist, or it’s going to cost a small fortune.

Raptor Engineering have just released a motherboard that isn’t x86 and doesn’t cost as much as a cheap car. The Blackbird mainboard is designed for an IBM Power9 CPU and it only costs $800. Add in a four-core CPU and the total cost comes out to about $1200. Add in some ECC RAM and you’re still under two grand. Building with a non-x86 CPU has never been cheaper. This is a significant change from earlier releases from Raptor Engineering, where just the motherboard cost $3700.

The Blackbird mainboard features dual DDR4 ECC DIMM slots, one PCI Express 4.0 x16 slot, one PCI Express 4.0 x8 slot, dual Gigabit Ethernet ports, 4 x SATA 3.0 ports, 4 x USB 3.0 ports, 1 x USB 2.0 port, and an HDMI display output.

The only reason you would build a Power9-based computer is simply to get around the black box that has become Intel and AMD CPUs. No one is really sure what’s going on in the Intel Management Engine, AMD has similar black boxes littered around. However, using a Power9 CPU has a secure boot mode and provided your computer is physically secure, you’re more or less assured you’re running your firmware and your kernel and your userspace apps. It’s security for the security-minded. RISC architecture is going to change everything.

93 thoughts on “A Sub-$1000, Non-X86 Motherboard

    1. Neither of those can be considered “motherboards”. The Raspberry Pi is a relatively low-speed SOC with very little in the way of expansion opportunities, and the Arduino is… well… an Arduino.

          1. I understand that the new Mac mini has removable RAM. I’m not going to go so far as to hope it means a return to removable memory across the board for Apple though.

    2. It’s a fundamental difference in processing power. One of those is a microcontroller (with no privilege separation) that can barely compute a trig function. The other is a very underpowered SoC someone stuck onto a board.

      The POWER machines R.E. build are designed to be directly competitive with the most powerful x86 systems. This is the stripped-down and cost-reduced version made fore someone who doesn’t need to run a computer farm. The fact that it runs completely free firmware/BIOS code is the major selling point, but the processing power makes this practical to real users.

      Want a ton of ram? no problem.
      Eight cores? you got it.

      None of that can be said of the raspberry Pi, despite a great deal of effort to re-implement the VPU firmware.

      1. The reason why Apple switched to x86 was because IBM’s PowerPC processors didn’t scale up in speed, or down in power consumption – they were too hot and too slow.

        How does a modern POWER9 compare? TDP? Speed?

        1. It was more due to IBM not bothering to make a notebook version of the G5 and the Intel switch allow Apple reallocate a lot of the Macintosh R&D team to the iphone project since they did n’t need much in the way of custom chips.
          Ironically the new world power mac was a fairly open platform as even the firmware was open source while the latest X86 macs are some of the most closed platforms there is this side of a game console.

    1. modern pipelined “CISC” systems are all either RISC or VLIW cores with decode engines bolted on the front. When we talk about microcode updates on a modern processor it’s code for the native execution engine (which can admittedly be pretty weird inside, I work on a crazy VLIW one)

    2. Well it did in in embed and mobile devices as nearly of them are based on Arm or some other RISC architecture.
      The main thing that keep X86 around was legacy code and billions and billions of dollars was spent making fast X86 chips to run it.
      To be honest things probably would have turned out better if people were willing to let go of that old code and X86 died back in the 90s.

  1. “RISC architecture is going to change everything.”
    RISC has been with us well over 20 years, and yes, it changed everything – in mobile devices. Just what further changes are you expecting?

    “you’re more or less assured you’re running your firmware and your kernel and your userspace apps.” Okay. More or less assured – that should be good enough for the security-minded. Also, says who? Says IBM? Oh yeah, IBM would never do anything sneaky.

    1. RISC changed a lot of things. Anyone remember SGI, MIPS, Sun Microsystems, DEC, and a dozen or so others? the primary reason for x86 adoption, IMHO was a free for some value of ‘free’) OS in the form of Linux on cheap desktop processors. When this whole tide started to turn, there was no such thing as s ‘server’, ‘mobile’, or ‘desktop’ x86 processor. They were all 386 or 486s. Pentium and the later Intel/AMD designs changed that, but then RISC was never designed for the commodity desktop either. The low end of a RISC workstation was $10-20k.

      I remember it well.

        1. It was a home computer with no major software support, and looked like a toy. Very much a home computer, not a workstation. Of course the fact it had CPU power years ahead of everything else was a nice feature, but running BBC BASIC at the speed of sound only has so much mass-market appeal.

      1. Not too well if you don’t remember that the primary reason for x86 dominance was Windows, and DOS compatibility. Microsoft already had the complete business and home desktop markets, and the combination of NT4 and the Pentium Pro killed the proprietary Unix workstation market stone dead.

      2. “the primary reason for x86 adoption, IMHO was a free for some value of ‘free’) OS in the form of Linux on cheap desktop processors.”…

        No the reason was Wondoz. People couldn’t afford a PC ***AND*** the software to go with it. So they went for a PC which they could “borrow” software for. They “borrowed” software from their employers, their friends and in some cases even went so far as to “pay” for dodgy floppies. The only reason M$ got where it is today is because of the “borrowing” and the x86 got caught in the wake.

        I remember it well.

        1. It was DOS that did it. The simple reason was software. You could pretty much get any program for the PC and that was not true with the Mac, Amiga, or ST when they were all much better systems than the PC. If you got a PC you could always find the software you needed.

          1. “much better systems” = “they all had very specific optimizations and hacks that enabled them to do one thing really well, and the fanboys then claimed superiority based on that”.

            For example, the Amiga was better than the PC becuse of the HAM-hack, which became irrelevant when the PC got VGA and SVGA graphics as an upgrade. The Mac, Amiga, ST, and pals just weren’t built to be upgraded and used as platforms for user software – they were built to be closed ecosystems with the intention of software locking people into buying the hardware.

          2. And another point being that the PC wasn’t locked to one vendor. You didn’t need to buy IBM to have a PC – even Macs really started getting popular when you could buy a macintosh clone instead of paying Apple whatever ridiculous prices they were asking. When Apple put their foot down on the clones by claiming IP rights on the ROMs and pushed the clone makers out of the market, the PC compatibles took over.

            So the value wasn’t actually free software, although that played a part. It wasn’t really the extensibility and flexibility as a platform, although that played a part – the real value was that you could buy exactly what you wanted to pay for from a healthy market of competing vendors. That put the prices down, the availability up, and made the PC a better machine on a service and support point of view – not being dependent on a single company.

          3. Luke: Of course, IBM had no intention of letting that happen. They made the hardware open, publishing the schematics and memory and I/O maps, so that other companies could make plug-in peripherals (following Apple’s example), but they fully intended for the BIOS to prevent others from making the basic machines. It was the Phoenix BIOS that made the PC an open platform, much to IBM’s dislike. And what made the Phoenix BIOS legal, was that it was written by two people. One reverse-engineered IBM’s BIOS, and from that, wrote a complete specification of what it does, but not a word of how it does it. This made that specification non-infringing. The other person then wrote the Phoenix BIOS using this spec, without ever having seen a byte of IBM’s code.

          4. The term “software” is too broad here.

            The original IBM XT made by “International Business Machines” was of course targeted at business and business software.

            If fact it was useless for games. The micro-controller in the keyboard only reported when keys were pressed and not when they were released and that made writing code for most games near impossible.

            Fair enough that changes with the AT and it’s that change that I use to get the “Ö” in my screen name directly from the keyboard but none the less IBM had that history and game programmers turned to other home computers and IBM (and clones) were more or less for business for some time.

    2. The beauty of this is that you don’t have to trust IBM. You can go through, read, and edit the source on your own, and run whatever compiled version you want on the actual hardware. Just try that with the IME or AMD’s version of it.

      Oh, and the HDL for the chip is even available. If IBM was being sneaky someone somewhere would be blowing a news story wide open: https://www.reddit.com/r/hardware/comments/9zveed/raptor_blackbird_microatx_power9_motherboard/eaeigvl

      And on that topic, seriously, when’s the last time you saw the hardware OEM and the CPU vendor both discussing this kind of detail in public? It’s been far too many years in my opinion!

      1. This is not true. The very CPU’s silicon is as closed as it gets and IBM will never let you audit or inspect it. That includes the microcode inside, which is not updateable unlike on x86, so you can’t even try to crack the encryption and reverese-engineer it. If you are into tinfoil, you can expect just about anything inside the Power9 silicon.

        BTW, the article doesn’t mention one thing – you need a special heatsink for this motherboard and CPU, which Raptor Engineering sells for $75 (not very nice price). You have to add that to the budget. Overseas shipping and “order handling” is also expensive outside USA but that’s another issue.

        1. Curious, since IBM licenses this very core to third parties for manufacture, from HDL to RTL. Did you even read the link posted above? Or are you making an assumption that IBM is acting like Intel and AMD (hint: no one outside of the company ever sees the RTL, let alone HDL, for their processors?)

          Even the ring data is open (HCODE). Sounds like you just want cheap consumer hardware and want to feel good about running other people’s hard work for free on your owner-hostile, insecure purchase.

          1. “Curious, since IBM licenses this very core to third parties for manufacture, from HDL to RTL.”
            Can you back this claim? Are there any CPUs from other vendors built on that?
            And even if there were, that doesn’t mean the HDL/RTL is open. IP gets licensed all the time, but when it’s under NDA and contractual obligations to keep it all secret, it stays closed.

        2. Yeah but you can’t run a CPU through a decompiler or a debugger. IBM claim that the chips they sell are the same as the designs you can read the HDL for. But if they were lying you’d have no way of knowing. You can read the BIOS, sure, but what about the hidden “nark engine” that runs it’s own code from internal ROM and has full access to hardware including RAM and network ports?

          Admittedly I’ve no idea if there is a “nark engine”. Intel and AMD grudgingly admit theirs, but again you’re only trusting IBM’s word there isn’t one.

          Of course, it can be hard to prove something doesn’t exist. But my point is, even with all this openness, you don’t really know what’s actually in the chip. You have to trust IBM. No reason to suspect them past the obligatory ones (they’re a big corporation, the world’s governments are sneaky fuckers), but that’s not undoubtable trust.

          1. If you have both the HDL and RTL, along with the core design itself (which is part of the license) it is very possible to decap a chip and verify what is being made matches the design. IBM (along with every other large server CPU vendor) does that as a matter of routine.

            If you think we’ll be seeing printable chips in the next half century that you can just make in a basement, better start working on a new version of the ZX Spectrum software, because that’s about all the performance you can expect even if it becomes possible (there are some really nasty chemicals involved in silicon doping; there is no way Joe Public will ever be allowed to use them).

            The ability to actually verify a high performance CPU is worth far, far more in practice IMO — IBM is theoretically allowing this (and has already opened the firmware so that anyone can verify and modify it); Intel, AMD, and most ARM vendors not only admit to having management processors but have very clearly stated they are for DRM and that no one can audit or verify them (let alone the CPU silicon). My understanding is that even Google cannot get IME source or the ability to audit the CPU, and if they can’t, who could?

  2. “Building with a non-x86 CPU has never been cheaper”… twenty years ago, you could buy standard AT/ATX boards with DEC Alpha 21066 (AXPpci33) or Motorola PPC603/604 CPU (Atlas) for less than $1000. Good times…

  3. “Building with a non-x86 CPU has never been cheaper”… twenty years ago, you could buy standard AT/ATX boards with DEC Alpha 21066 (AXPpci33) or Motorola PPC603/604 CPU (Atlas) for less than $1000. Good times…

    1. On general purpose applications RISK is good. Interestingly, hardware can be more power efficient than software, so a purposed designed CISK processor, with an instruction set designed for one specific application, can be more power efficient. But that is quite a tight set of requirements, but one that should be kept in mind in an era of growing quantities of application specific FPGA CPUs.

  4. oh wow a power9 CPU … and in 4 cores i was totally unaware they made them in quad core variants normally they are 12 or 24 core making me wonder how old they are
    the only reason one would use a power9 CPU as far as i can tell is the use of NVLink and multiprocessor configurations and this motherboard has none of these options. Software support is going to be a problem also there is zero evidence that power9 CPUs are any more secure

    also i think that the RISC comments are snarky as power9 is NOT a risk based microarchitecture

    1. Power9 has two types of cores, smt4 and smt8, where a core is actually 4 or 8 processing units interconnected for multi-threading.
      A physical chip has either 24 of those smt4 cores, or 12 of the smt8 cores, all on the same die.
      They call each of their single threaded processing units a “slice”, and you always get 96 of them.

      So what the article calls 4 or 8 core would be more akin to a single intel cpu hyperthread.
      What you call 12 or 24 core the number of 4 or 8 way CPU units.
      The 96 slices is more akin to logical processors that you would see in /proc/cpu or task manager.

      As for their age, about two years old. And yes they are RISC, a direct descendant from the PowerPC and RiSC CPU ISAs from the 90s.
      In fact that is what the “R” in “Power” stands for: Performance Optimization With Enhanced RISC

      1. So a 4-core SMT4 Power9 chip has 16 threads? Also, from the Wikipedia article, it looks like the 4/8/18/22/24 core variants are all the same die, just with different numbers of functional/enabled cores.

      1. Intel x86 CPU architecture has been RISC since the mid 90’s.

        Why would they continue to use a slow cumbersome CPU (hardware) architecture when all CPU manufacturers were competing for CPU performance / cost ratios.

        An Intel x86 compatible CPU breaks x86 CISC instructions into microcode which runs on a RISC processor.

        All of this is just for compatibility. The the coder the CPU is x86 CISC but at the micro-architecture level the hardware is RISC.

        So the next question is …

        Is any of this actually true or am I just a very convincing troll?

        1. Is that still the case? RISC was handy in the recent past because simpler designs meant clock speeds could go up and up. Now we’ve hit the wall on that, getting more done per cycle is the only way to advance.

          Certainly x86 chips used to contain a RISC core and a translation layer, but is that still true? x86s rely more and more on ludicrously complex hardware. Since MMX they’ve added more and more of it. MMX the CPU feature, not the year. Since we’ve hit the clock speed limit, but also have great oceans of transistors to use, it makes sense now to build chips as complex as possible.

          I’m sure I heard somewhere that Intel went back to CISC a few chips ago.

    1. I think that’s unlikely – the UPNP issue is within broadcom *router* firmware, not mere ethernet ports.

      You’d be slightly less wrong to assert that raspberry pis (BCM2835 etc) are on fire sale due to a broadcom vulnerability.

  5. Hey does this mean I’m going to finally get my power PC development board that IBM and Motorola promised me in 1997? I’ve been waiting a long time for that… it would be a merry Christmas indeed if it actually showed up

  6. But can it run Crysis at max settings?

    In all seriousness, the promise of new hardware is great but there’s effectively nothing to run on it unless you’re prepared to do some serious programming.

    1. When I get some money and buy one of these, I hope to laugh maniacally as I run whatever I want through the glorious twin powers of “emulation” and “having tons horsepower on tap” (thank you, Turing completeness and enthusiastic programmers!) Because they have a *nix install, and from a *nix install I can compile a great many things. I hope to find that compiling and switching to a minimalist distro from Raptor’s default distro (Red hat? I forget) isn’t too difficult.

      Heck, I’ll probably be able to run Rocket League on that thing – sure it’s nothing new, but I enjoy it and it’ll be running 2 nested emulators deep. If I get 60FPS I’ll smile like a madman.

      The future’s now, and I’m excited!

    2. Yes, nothing to run except all of the of the GNU/Linux world. GCC and the Linux kernel have had support for years. And from it you have distros like Debian, Ubuntu, and Fedora. It’s an official release architecture for Debian as of Debian 8 (Jessie) from April 2015. And, of course, there’s absolutely *nothing* in the Debian repos, right? You’d need to do “serious programming” hahaha :)

    3. These aren’t meant for ordinary desktops. They’ll run high-end stuff, just IBM themselves have a huge ecosystem of hardware and software (and all with their own silly alternative terms for things like hard drives). Nothing most people would ever see, but one field is big databases, IBM have tons of expensive “solutions” for things like that.

      What is this particular board aimed at? I dunno, maybe somebody who admins a POWER system at work, and wants to be able to tinker with stuff at home? Or a way to get IBM’s weltamschaung into smaller businesses? Selling to BIOS conspiracy theorists?

      Point is there’s plenty of software for it, just nothing you’ve heard of. Me neither, mostly. But there is a whole IBM world out there, with real inhabitants.

      1. I don’t think “conspiracy theorists” is correct. That implies that the threat posed by a malicious firmware stack is non-zero and that people are reacting to a non-existent threat. Nothing could be further from the truth; for just one tiny example take a look at just the most recent flaw in the IME to be disclosed: https://www.theregister.co.uk/2017/11/09/chipzilla_come_closer_closer_listen_dump_ime/

        Now imagine if you relied on that firmware stack (which every Intel user does, daily, with no replacement possible due to Intel fiat) and were compromised because Intel made parts of it very highly privileged (far more than they should be) for DRM purposes. Do you think you will have a winnable case against Intel for loss of trade secrets, harm to business, etc.? Of course not; it’s just (black box, signed) firmware and therefore you would have no recourse whatsoever. It even remains to be seen if any monetary recourse will be available for people affected by Intel’s proven hardware defects.

        Might be a good idea to read up on malware before dismissing these concepts out of hand. There was a time when a red pill / blue pill attack would have been dismissed too as “conspiracy”; they are now a well known attack against vulnerable systems. Firmware is just the next step down, and the final residence for the most advanced APTs today.

    1. Want a CPU with designs freely available through openPOWER, with full open source firmware and BIOS code?

      This things was made to be fully audit-able and secure, running only what you tell it to. The excellent performance is a nice cherry on top.

    2. I tried to buy a Synquacer ages ago. The supplier literally told me I can’t have one. They were supposed to get back to me once they had availability. That never happened. So the supplier, the only supplier, did not have any availability what-so-ever and when I asked about them it was so foreign a question that they had to get back to me. So I bought MacchiatoBin instead. It works. Don’t buy the SingleShot (like I did).

    3. It’s 24 cores, but reading the fine print it’s 24 Cortex-A53 cores running at 1GHz. 5W only gets you so much. I would be interested in an ARM workstation, but it has to be at least comparable in performance to mainstream commodity hardware. Sitting around waiting for builds gets old really fast. The ThunderX2 workstation would fit the bill, except it costs somewhere around $10K.

  7. The real market for this is as a development system of course, so you don’t have to spend $20k for a bottom end Power9 system just to compile and test code on. But it’s still very cool. Internally IBM ported Windows to Power4/5 over a decade ago which is amusing. But the Power port of Windows will never leave their R&D labs. Back in the day I used to have a loaner PowerPC Thinkpad that ran AIX 4.1. It wasn’t going to break any speed records, but it was awesome to have a commercial unix system I could cart around in my backpack.

      1. “Not IBM, Microsoft. They also ported WIndows NT to Sparc, MIPS, Alpha”… No M$ never ported NT to ALPHA. They stung DEC for a huge licence fee for NT 3.1 so they could put it on their ALPHA then brought out NT 4.0 and wanted more money for that. You got to M$ – they are consistent.

        1. Quoting Wikipedia: “In order to prevent Intel x86-specific code from slipping into the operating system by developers used to developing on x86 chips, Windows NT 3.1 was initially developed using non-x86 development systems and then ported to the x86 architecture. This work was initially based on the Intel i860-based Dazzle system and, later, the MIPS R4000-based Jazz platform. Both systems were designed internally at Microsoft.”

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.