Hacker News Reader

Comment by _ph_ | Hacker News Reader

This is something I don't understand with any cpu manufacturer which isn't x86. Why don't the manufacturers offer standard pc motherboards tweaked for their corresponding cpu? Since the mid-90ies, they would have made nice Linux platforms. I think, many Linux enthusiasts would be thrilled to run an alternative CPU platform. By offering cheap desktop solutions, they would drive the software base and interest for the more expensive enterprise hardware. Back then it was Sparc - it was still competitive in performance with Intel, later PowerPc, now Arm64.

View on HN

+1 to that. I feel his pain, having been there myself.

Back when DEC Alpha CPUs were still a going concern, I was one of the maintainers of the FreeBSD/alpha port. While DEC shipped their workstations, they were beyond the reach of OSS developers. The first "normal" Alpha board was the API UP1000. It was basically an Alpha 21264 on an AMD motherboard & it was totally awesome. API shipped me one before they were generally available so I could get FreeBSD working on it. I used it as my desktop (running FreeBSD/alpha) for years. Sadly all things Alpha were killed off in favor of Itanium.

Here is the best description I can find of it in ~30 seconds of googling: http://www.thefreelibrary.com/API+Announces+Program+to+Encou...

I fondly remember getting an AXP-PCI 33 board as a teenager, which were (at some point) really affordable [1]. It was my first non-x86 machine.

[1] http://manx.classiccmp.org/collections/mds-199909/cd1/alpha/...

"API shipped me one before they were generally available so I could get FreeBSD working on it. "

I didn't know about that. That's pretty neat. One of Alpha's main advantages was PALcode. It was close to microcode's advantages but with only assembly knowledge required. Did you or other FreeBSD developers ever do anything cool with it? And what?

Typing in the domain of that company today shows how sad the IT market works when incredible products are involved that aren't market leaders:

http://www.alpha-processor.com/

There were loads of DEC multia workstations selling cheap at some point. But even those where a pain to get booting into linux of FreeBSD.

I think the cold truth is that if they did, the non-x86 system would cost almost as much as an x86 system, and would perform worse. When you add more ports, connectors, RAM sockets, CPU socket etc., the cost of the CPU starts to becomes relatively unimportant. A cheap x86 Celeron, which is as fast, if not faster, than an ARM CPU, will only add a small percentage to your budget. And even a decent i3 won't cost that much more either, percentage-wise.

This why we only see Raspberry Pi-esque devices, I believe. ARM CPUs excel in the embedded space, which is not surprising since this is where they grew up. Their qualities are low cost and low absolute power usage (as opposed to performance per watt). This doesn't make sense for desktop systems.

There were times, when such a system might have outperformed x86 hardware (Sparc, Power). And even if not strictly competetive, people who care about writing low level software should be delighted working on something different than x86. I certainly would like to work on an Arm64 system, ideally with 8+ cores. That should be quite interesting.

May well be. I can't shake the feel that all performance we have gotten have been from smaller chip features allowing for more more heat headroom, and thus more "waste" via out-of-order, all branches at once, massive caches, etc etc etc. And once we hit the brick wall that is electron leakage, RISC or CISC will mean crap all for performance pr watt.

Never mind that in recent years both AMD and Intel has taken their offerings into the real of SoCs. AMD have largely lead the way, by moving more and more of what used to be the northbridge onto the CPU die.

The only thing keeping both out of the mobile world seems to be price pr unit. Intel bled money to get Asian OEMs to use their mobile ATOMs rather than Chinese ARM based SoCs.

The only thing really missing for many use cases is a better storage interface. The rest is gravy.

There are several single-board computers with various permutations of USB 3.0, GbE, SATA, mSATA, miniPCIe, etc.

Check out the ODroid series, Banana Pi Pro, ECS Liva, Jetson TK-1, etc.

While I have to acknowledge the RPi's pioneering contributions to the ultra-cheap SBC market, I find it immensely frustrating that it sucks up all of the oxygen on this topic. There are many, many other options at this point and many of them are better choices for the typical "mini Linux server" or "low-power desktop system" use-case.

> Check out the ODroid series, Banana Pi Pro, ECS Liva, Jetson TK-1, etc.

For which of these do regular updates (in particular kernel updates) from the mainline distribution exist?

Depends on your use case. I believe the Banana Pi Pro is supported by Debian Jessie with the stock kernel if you can do without video and audio out and a handful of other features you're unlikely to need in a headless server. See http://linux-sunxi.org/Linux_mainlining_effort - Banana Pi /Pro is A20, Jessie ships Linux 3.16. (Audio and unaccelerated graphics are supported by newer mainline kernels, in theory at least.)

The ODROID-C2 (their AArch64 offering) was added to the Linux kernel device tree in March: https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux....

But Hardkernel does seem to have a lot of additional patches in the kernel that they ship: https://github.com/hardkernel/linux/tree/odroidc2-3.14.y

There was a motherboard for the StrongARM SA110 CPU, I still have one. Fits in a standard PC case, takes PCI expansion cards, connectors for IDE etc. in the standard places.

Sun made a Sparc motherboard too but I don't know how easy it was to buy one.

Sun's "SME" division sold the Ultra AXi and AXMP boards, which were 100% standard ATX boards with PCI slots.

The Ultra 5 and Ultra 10 were basically PCs with SPARC chips.

31 Oct 2015 Titanium AM5728 Mini-DTX

Four 3Gbps SATA connections

Two PCI Express root complexes

http://www.ti.com/devnet/docs/catalog/thirdpartydevtoolfolde...

Huh, that actually looks standard.

Approx $590.

I can't get a clear idea if this is 64-bit or not though.

Also, with this board, I don't see anywhere obvious to bolt a heatsink - and at 1.1GHz, I'd be be reluctant to do overnight compiles on the thing.

Looks like it uses two Cortex A15s, which is 32 bit.

http://shop.elesar.co.uk/image/data/brief-ti22.pdf

https://en.wikipedia.org/wiki/ARM_Cortex-A15

It's not necessarily cheap to provide and support a full desktop system, especially given the very small minority of people who might actually be interested in it.

> Why don't the manufacturers offer standard pc motherboards tweaked for their corresponding cpu?

Because there's no standard interface. Each SoC has its own set of pins, which changes every generation for every manufacturer. Each board would essentially have to be redesigned for each CPU, meaning that the idea of having a separate "motherboard" would be completely pointless, as it'd be totally coupled to the CPU. You could have motherboards with pre-soldered, rather than socketed, SoCs, but then again, manufacturers would essentially have to put as much work into each revision of those as they would into a true PC motherboard that has a much broader audience.

The economics just don't work out.

My old StrongARM motherboard has the CPU soldered in place.

Most current SoCs don't have a PCI bus so there isn't very much you would be able to add on. A manufacturer could provide one or two SODIMM sockets instead of soldering DRAM onto the board though.

> Why don't the manufacturers offer standard pc motherboards tweaked for their corresponding cpu?

No expertise and x86 is baked deeply into PC's.

1) No Expertise

ARM manufacturers tend to synthesize logic a lot. They don't tend to pull out circuit simulators and design things by hand. As such, the advanced techniques behind fast memory, CPU and graphics interfaces simply aren't there.

2) x86 is baked into PC's

People forget that there is a lot of "initialization"-type code on things that plug into buses. ALL of that code is x86. The Alpha PC's had to include emulators for x86 at boot time in order to init the old PCI cards.

Chicken and egg. It costs more than you think, for probably fewer geeks than you think.

ARM seems really well positioned, if there has ever been something to square off with Intel, it's ARM, you can get cheap SoCs and play with them. The software ecosystem is growing so maybe someone will throw down and be willing to put together a full ATX system, I suspect it will cost more and perform less than an Intel based product though.

RISC-V might have this edge if a manufacturer picks it up. Maybe the SiFive folks are listening.

I'd definitely be interested in running a system like that, if it had PCIe ports to run a supported GPU, perhaps a fancy NIC or a RAID card.

If they would make a PC-format RISC-V board with standard attachments, I would buy several in an instant. Though I know it's not free to license all of those bits, this being a major expense on x86 boards.

Quite simple.... there is no market for them.

There is a commodity PC market. There isn't a commodity ARM motherboard market as the type of devices ARMs go into aren't PC's, they're devices.

Yes... there is a hobbyist market. But that isn't worth investing in.

> Yes... there is a hobbyist market. But that isn't worth investing in.

The home computer market was originally a hobbyist market.

>I think, many Linux enthusiasts would be thrilled to run an alternative CPU platform.

forget Linux enthusiasts. Whole countries, like Russia or China, have national security interest in it - beside everything else, Intel chips has VNC server built-in on the silicon :). China has been making great progress in building its national computing platform, though not sure whether it has reached the desktops of their government employees. Wrt. Russia - it again lifted the ban on buying foreign hardware for security services, time to upgrade has come again i guess :) I mean, even such guaranteed markets and money available is still not enough to even come close to Intel.

Russia mostly uses SPARC: an ISA used from embedded to high-end NUMA's that was first to open up. Their mainframe CPU's are called Elbrus. Fujitsu has their own that they work with Oracle on. There's also a MIPS company. They also cloned and improved Itanium but sold that to Intel. Reason they avoid ARM, like I tell everyone else to, is that ARM charges huge licensing fees (up to $10mil I hear) with royalties. MIPS charges around $700-900k. SPARC requires a $99 registration fee. See why Japs and Russians use SPARC? ;)

The Chinese first, powerful CPU was an Alpha clone called Shinwei. Their next one was MIPS line called Loongson (or Godson). It supports x86 emulation at hardware level. So, that MIPS CPU is their main strategy for getting off Intel. I don't know if Russians have a strategy for that. But that's what the two are doing in terms of CPU's.

EDIT: China also built its own state-of-the-art fabs with a 28nm coming online recently. The Russians have fabs but not that many or that good that I know. That's two areas they should worry about.

Russia seems to be not very interested in ARM.

https://en.wikipedia.org/wiki/Baikal_CPU

VNC on silicon? Do you have sources for that?

It has different names. AMT is one of the earliest implementations.

There is a separate chip with its own OS, connecting to the video chip and the nic. No driver needed. Can not be permanently removed.

https://en.wikipedia.org/wiki/Intel_vPro#VNC-based_KVM_remot...

http://www.meshcommander.com/meshcommander as a client to the Intel AMT and maybe ME, which if you do some more research, you'll see is indeliby baked into most newer Intel hardware.

I assume people aren't making non-x86 desktop or workstation systems because they cost a lot of money to develop, even just a motherboard. If the system cannot run Windows, you've just made the whole idea economically infeasible.

I think the issue is more demand than supply. I'm currently working on an embedded software project where the arm based SoC supports 2 x4 pcie buses, 3 usb ports, sata and ddr. The evaluation board looks like a normal motherboard with 2 pcie slots, a ddr slot and usb and eSata ports. And it was running ubuntu.

People know how to build non-x86 desktops. Building non-x86 desktops is not difficult to do or expensive. People just don't want it.

What's the price of that evaluation board? Can the CPU run faster than in embedded scenario thanks to better cooling?

A comparable example: https://boundarydevices.com/product/nitrogen6x-board-imx6-ar...

This kind of thing is stuck in a volume trap: you need to produce enough units to make it cheaper than a PC, but nobody will buy it until you do. In the meantime, people are buying ARM tablets and 'mini-PC's from various Chinese OEMs, but they're non-modular units.

The modularity of the PC is almost an accident of history.

So that's what all that Windows RT stuff was about. I'm a bit ashamed I didn't realize Windows (proper) doesn't run on ARM. I'm assuming MS really wants to be able to run Windows on different types of processors but can't practically make that happen and keep backwards compatibility. It sounds like they've shot themselves in the foot. Unfortunately, they were stepping on all our heads at the same time.

Windows NT was written to be cross-platform, and it was actually developed on RISC chips -- hence the HAL (Hardware Abstraction Layer). Unlike Linux, it wasn't written on and for an x86 processor....

At one time, commercial Windows NT PCs were sold with DEC Alpha processors.

However, the volume of x86 PC sales drives huge economies of scale that make it hard for other platforms. This actually benefited Linux in particular, but it also encouraged Apple switch from its own hardware to PC-based x86 hardware, which resulted in cheaper Macs.

Microsoft has never developed an OS specifically for Intel x86 chips (ie Windows NT and CE: the basis for DOS was bought in) to try to avoid being wholly dependent on Intel. Intel has also supported alternative operating systems to avoid being wholly dependent on Microsoft. But neither is immune from market forces.

I still have an old NT 4(?) install disc with x86, MIPS, Alpha, and PPC arches.

NT4 was the last to support MIPS & PPC, but only x86 and Alpha got SP3 and beyond.

Digital/Compaq did have a beta build of Win2k for Alpha, but the plug got pulled right before the RTM.

Did the code ever escape in any form?

Windows itself ran fine on ARM. Windows RT failed for a couple reasons: there was no backwards compatibility for x86 Windows software, and there were a couple artificial limitations in place that made ARM-native desktop software impossible. Essentially, Windows RT threw out the good part of what makes Windows Windows (decades worth of third-party software) in an attempt to make a "dedicated" tablet OS a la iOS or Android.

True. However, Microsoft thought users would be OK with Microsoft Office (bundled free with Surface RT) and sandboxed tablet apps.

Windows RT threw out the good parts of backwards compatibility, but it also threw out the bad parts: viruses and malware, software downloads from dodgy websites, being your own IT manager etc.

Some would consider this a reasonable trade-off, but the market didn't buy it. Odd, really, since Windows RT has all the benefits of a Chromebook, but also offers much more.

If MS had enabled Domain Join & Group Policy, the RT devices could have taken a lot of enterprise sales. At the time, there wasn't a great story on any tablet around user account management; RT could have been that platform.

Agreed. It would also have filled the hole for school testing, which is what generated sales of Chromebooks....

The Amiga people kept getting new PPC's built for their systems. Almost nobody bought them. Gaisler open-sourced his Leon3 SPARC CPU and I.P. under GPL. Sun did same for OpenSPARC T1 & T2. Nobody built anything or pooled together money. MIPS and ARM stayed in embedded space mostly with big vendors (esp SGI) switching to Intel/AMD since they performed better per dollar. That along with licensing cost meant nobody was doing a desktop CPU there since it be a financial sinkhold. I still think Cavium's Octeon III and ThunderX could become great desktops with a PC board. Yet, they show no interest in that outside cloud servers maybe. PA-RISC and Alpha similarly died since no demand to pay for increasing expense on CPU's. SuperH, which powered Dreamcast, stayed in embedded with higher-end one doing multimedia products.

So, not only are there no desktops, everyone that tried either sold jack and/or went out of business. Your theory hasn't been true for a long time. Linux and FOSS types talk a lot but buy very little if it's not x86. A desktop could easily be made in Shenzen or whatever based on existing components, esp Leon3 or Leon4 SOC. eASIC's mask-programmed tech already supports Leon to crank them out quick and cheap. Peripherals could be handled with cheap microcontrollers if necessary that have onboard HW accelerators for them. The tech is there just no suppliers right now and usually no actual buyers.

Something like this: http://b2b.gigabyte.com/products/product-page.aspx?pid=5422#...

That's the $1000+ server market. In Germany it sells for 850 EUR : http://www.heise.de/preisvergleich/gigabyte-mp30-ar0-9mp30ar...

That's perfect in every way except price.

The CPU is 2.4GHz and 8 cores.

It can handle 64GB ECC DDR3.

It has onboard VGA since GPU support (which typically involves x86-specific initialization instructions at boot time) will be spotty.

Remote management (dedicated NIC) w/ iKVM and VGA.

4xSATA3

2x 10G SFP

What's not so good: only USB 2.0 - two external ports and one header. Not sure how many root hubs and what goes where. That said, you can use eg a USB3 HDD (or do almost anything else involving USB3) via the 20Gbit Ethernet to another box, so...

I'd settle for a standard power/backplate, maybe about 0.6x the size of the PC backplate that's used from ITX through ATX boards... then have the MB size be that, squared... which is bigger than most enthusiast devices (rpi and similar). Something with a fairly standard mounting interface, and power adapter would be a good start.

From there, a standard interface for booting and mounting the OS and additional devices. There'd be a lot of work in that, and a lot of competition, but would be nice to see some work in the open for a more standard platform for ARM devices.

They were available in Germany mit 90ies. Nobody bought them.

Vobis Highscreen Alpha 5000, based on alpha processor 21164PC

Hardware companies are really really traditional in their business thinking and do not understand virality or 'priming the pump' at all. They do a straight-up market size analysis, conclude that the (current) market is small, and move on.

Their margins are low, and it's super expensive to tool up a factory, so they are (intelligently) going to be somewhat risk adverse.

Reply on news.ycombinator.com