Slashdot Mirror
'Is It Time For Open Processors?' (lwn.net)
Linux kernel developer (and LWN.net co-founder) Jonathan Corbet recently posted an essay with a tantalizing title: "Is it time for open processors?" He cited several "serious initiatives", including the OpenPOWER effort, OpenSPARC, and OpenRISC, adding that "much of the momentum" appears to be with the RISC-V architecture. An anonymous reader quotes LWN.net:
The [RISC-V] project is primarily focused on the instruction-set architecture, rather than on specific implementations, but free hardware designs do exist. Western Digital recently announced that it will be using RISC-V processors in its storage products, a decision that could lead to the shipment of RISC-V by the billion. There is a development kit available for those who would like to play with this processor and a number of designs for cores are available... RISC-V seems to have quite a bit of commercial support behind it -- the RISC-V Foundation has a long list of members. It seems likely that this architecture will continue to progress for some time.
Here's some of the reasons that Corbet argues open souce hardware "would certainly offer some benefits, but it would be no panacea."
Here's some of the reasons that Corbet argues open souce hardware "would certainly offer some benefits, but it would be no panacea."
- "While compilers can be had for free, the same is not true of chip fabrication facilities, especially the expensive fabs needed to create high-end processors... It will never be as easy or as cheap as typing 'make'..."
- "Without some way of verifying underlying design of an actual piece of hardware, we'll never really know if a given chip implements the design that we're told it does..."
- "Even if RISC-V becomes successful in the marketplace, chances are good that the processors we can actually buy will not come with freely licensed designs..."
- "Finally, even if we end up with entirely open processors, that will not bring an end to vulnerabilities at that level. We have a free kernel, but the kernel vulnerabilities come just the same. Open hardware may give us more confidence in the long term that we can retain control of our systems, but it is certainly not a magic wand that will wave our problems away."
"None of this should prevent us from trying to bring more openness and freedom to the design of our hardware, though. Once upon a time, creating a free operating system seemed like an insurmountably difficult task, but we have done it, multiple times over. Moving away from proprietary hardware designs may be one of our best chances for keeping our freedom; it would be foolish not to try."
There are several dozen teams designing RISC-V implementations. And many ASICs have RISC-V cores buried in them today. With a handful of designs being open.
The main barrier for ordinary people and software developers to have a proper R5 workstation is for there to be a market for such a chip. Right now the market is driven by the needs of ASICs, and that's not really what people are asking for when they say an "Open" processor.
“Common sense is not so common.” — Voltaire
It might find some niche even if never becomes a mainstream product, much like Linux never really took off on the desktop, but became insanely important in the server space. I suspect that this could be successful for low-cost devices that need a lightweight processor. As overall device costs decrease, the extra costs from buying a third party SoC become larger and using an old process node and an open design is going to result in some potentially significant savings.
I also think something like this has some value in education even if it doesn't do much commercially.
... but it takes a massive amount of money to design and make chips. It's not going to happen "open source" unless some very wealthy individual or organization decides to do so for altruistic reasons.
I don't respond to AC's.
One online article notes 16nm Finfet fab entry cost at $80M, 66 mask steps. You would need a very wealthy patron.
That $80M is the cost to use a fab - the cost in setting up the masks to have the fab make your processor. Building a modern fab is on the order of tens of billions of dollars.
There are several dozen teams designing RISC-V implementations. And many ASICs have RISC-V cores buried in them today. With a handful of designs being open.
The main barrier for ordinary people and software developers to have a proper R5 workstation is for there to be a market for such a chip. Right now the market is driven by the needs of ASICs, and that's not really what people are asking for when they say an "Open" processor.
Designing the architecture and logic is fraction of the engineering effort necessary to design and build a modern high end microprocessor.
I should use this sig to advertise my book ISBN-13 : 978-1501515132.
DARPA had (has?) a program to try and figure out how to ensure the computer hardware DoD is purchasing is what is actually being delivered. There are more problems with hardware than simply design and the cost of buying fab time. Validation that the design was produced correctly is not trivial in complex hardware. Opening the whole process would help solve that problem, and the DoD may have the deep pockets necessary to pay for actual hardware builds.
I believe you need to go back and re-read the REASON for the law.
The idea for it started from "lazy journalism" - which this is none of. This is a vetted technical person actually asking a technical question to the community at large. The technical question is followed up with detailed analysis on why such a question is being asked, and the ramifications of the decision if it were to be made.
That's more than an order of magnitude higher than the NREs we were paying for the ASICs (including sea-of-RISC network processors) the last time I was doing ASICs - abouit 5 years back.
Has it gotten that expensive? I sincerely doubt it. But even if it has:
You can do your prototyping at fabs that combine the prototypes from several customers into one combo wafer, split the NREs among them, and do a small run - then repeat a couple months later, ad-infinitim. If kyour design works you've already got your mask design placed and routed, and it's just a matter of making another set where you step-and-repeat for a whole wafer. (Meanwhile you can do small volumes and proofs-of-concept with the few dozen you got from the prototype run - or even get a few more made from the old masks and just get your piece.)
Bantam Dominique roosters crow a four-note song. Once you've heard it as "Happy BIRTHday" you can't NOT hear it that way
This is what happened after China acquired AMD license to produce x64 chips in China, and acquired VIA's x86 license which VIA got from acquiring Cyrix.
The CPU license pool is cracked opened. Soon CPUs in China will be 1/4 the price of Intel/AMD but has better performance.
https://www.reddit.com/r/hardw...
Zhaoxin launched KX-5000 quad/octa-core x86 processors on Dec 28, 2017 in Shanghai, China: image, report, translation.
Zhaoxin revealed KX-6000 & KX-7000 roadmap: image, report, translation.
Other reports: golem.de, pcgameshardware.de, bitsandchips.it, phoronix
KX-5000:
Full SOC design (integrated southbridge)
28nm process by HLMC, 2.1 billion transistors
4-core / 8-core SKUs, no SMT
2.0-2.2GHz base clock, 2.4GHz max turbo
IMC supports dual channel DDR4-2400
PCIe 3.0 lanes
iGPU
integrated audio codec
ZX-200 I/O extension (chipset): SATA3.0, USB 3.1 Gen2, Gigabit Ethernet
OEM: Lenovo desktop M6200
KX-6000: 16nm tick-tock
KX-7000: new uArch, DDR5, PCIe 4.0
Related info:
About VIA & Zhaoxin: wikipedia and wikichip.
KX-5000 preview: image, report
KaiXian KX-5000 series was listed in PCI-SIG integrators list on Nov 10, 2017.
Zhaoxin KaiXian KX-5640 in SiSoftware database.
Zhaoxin ZX-C, KX-5000 series on exhibition on Nov 21, 2017 in Ukraine: report, translation.
KX-5000 CPU arch: block diagram, report, translation.
This is their first chip, and it's already faster than half of Intel's recent low cost chip.
All they have to do is sell whatever they have at 1/4 the price and Intel's China market will be shrunk by 75%, that means Intel/AMD will have loss of revenue and their cost will be increased due to smaller scale of mass production, which will lead to another round of market shrinkage.
Every industry that have underestimated China have been wiped out. Not to mention IC is one of the industry that is backed by the Chinese government to win at all costs.
Just like German and Japan's high speed rail, and soon Airbus and Boeing.
How about an open version of the Motorola 68000 series of CPU's? Those were great in the day. maybe Motorola would open up the tech on them and let them be advanced. Assembly for them was easy to learn and had a very small instruction set to learn. Learning assembly on the Commodore Amiga's was a snap with the Motorola 68000 series of CPU's.
The Truth is a Virus!!!
Not even that. There is actual expense involved. You will not be downloading a Risc-V or any other processor core and then going to a 3d printer to print it. That will never happen. The technology for 3D-printing right now couldn't even 3d-print a tube for an analog computer.
What people are using right now are FPGA's which cost 100x more than the chip core they are capable of emulating. Most of these FPGA's that are in affordable range can barely emulate an 8-bit computer. So unless you want to sacrifice the performance of an Intel i7 in favor of an Z80, no I think this is barking entirely up the wrong tree.
We have to be resigned to the fact that there will never be "free cpu"'s at all. A chip fab will not spend a billion dollars on a facility, and then produce everything for free. That's insane.
Where we can potentially break the ceiling on proprietary designs lies more in breaking the CPU up into pieces that can first be designed using multiple FPGA units, and once those designs are proven to work, license those out pieces out to whoever is willing to finance the production of the ASIC's that must not be modified from the open design. If you allow third parties to modify it, then we're back to proprietary designs again. This is one of the quite literal best use cases for GPL-like provisions. Any ASIC released, must be released with everything required to reproduce the ASIC.
A monolithic design similar to the i7 in scale, will simple not be possible without someone like Apple adopting it, and nobody else out there is willing to fight the inertia to do such a thing. Even Nintendo has moved away from RISC to ARM.
You can buy RISC-V asic's currently https://www.sifive.com/posts/2017/10/04/sifive-launches-first-risc-v-based-cpu-core-with-linux-support/
But you're looking at a chip that is so slow, that you would not be sticking this in servers or desktops, it would be relegated to hobby devices.
Did a lot of people really say that Linux would never compete against "real" operating systems in 1991-1992? But what's the connection anyway?
First person: "You can't travel faster than the speed of light"
Second person: "They said the exact same thing about traveling to the moon".
You don't follow world news do you.
China built 25,000km high speed rail in 5 years, through deserts, glaciers, mountain ranges, forests, how many km have the Germans built?
Chinese trains have become so good that Germany's Deutsche Bahn wants to buy them.
According to DW columnist Frank Sieren, the railway can no longer afford to give preferential treatment to German companies.
http://www.dw.com/en/sierens-c...
Chinese train technology rolls into Germany
http://www.chinadaily.com.cn/w...
China is on track to build high-speed rail in just about every corner of the world
https://qz.com/292321/china-is...
France and Germany now have to team up to compete with China
France-Germany rail merger aims to take on China
http://money.cnn.com/2017/09/2...
The deal aims to counter China's growing clout in global rail markets. Beijing stepped up its efforts in 2015 by merging two big companies into state-backed giant CRRC, which describes itself as "the world's largest supplier of rail transit equipment."
The exact same thing was said about Linux in 1991-1992, that it would never compete against "real" operating systems like Solaris, ULTRIX, and others.
That is not my recollection. There was a demo of X11 running on SLS Linux at the 1992 SUG meeting, and the folks from Sun were giving each other very concerned looks. They clearly saw it as a serious threat.
Let's assume for a moment we had a rousing speech from the ghost of John F. Kennedy saying that this community should commit itself, to achieving the goal, before this decade is out, of creating an open processor, and installing it safely in a computer. And Jeff Bezos thought it was a good idea and committed to writing a blank check to make it happen.
And enough of the the few thousand people in the world who can ground-up design a processor have willingly donated their time to the effort, and have made a perfect, error-free processor with very little physical testing, and one or two of the few-dozen-at-best CPU fab plants in the world have committed their time to retool their assembly lines to decrease the output of Intel and AMD and ARM and Qualcomm chips to make a few hundred thousand of this OpenProc. Also, we're assuming that all of this is done such that there are zero patent infringements from the existing guys, and thus at no point are there any lawsuits from Intel or AMD.
We're already comfortably in 'not happening' territory, but let's keep going.
These CPUs need to fit into motherboards somewhere, right? I mean, the implication here is that we're looking for desktop and server chips. They're not going to work in standard Intel or AMD sockets, I'm assuming...so on the heels of designing an open processor, we need an open motherboard to fit it (which again, avoids any and all litigation as it's being designed). Somewhere in that process, we also end up with an OpenNIC and an OpenSoundBlaster and OpenSATA and FreeUSB and FreePCIe et al. Also, someone codes a ground-up open UEFI or BIOS or something that interacts with all of this hardware properly and without issue or conflict, because any issue faced in this scenario becomes the biggest possible nightmare to test. Also, Foxconn agrees to produce this MagicMobo alongside standard, more profitable units.
Now, we've got all that hardware and can get to a boot device. What are we booting? Linux successfully compiled for this barely tested hardware using a compiler that assumes all the specs are, in fact, working as intended? Okay, great! now let's get some more software on it, because a full Linux distro, even something as relatively-simple as DSL or Puppy is going to require all of its software to be recompiled, so it's yet another race to start porting over applications, with some applications never leaving x86 due to a lack of developer interest.
Everyone, everywhere, ever, has willingly done their part to support this new architecture. Now, to convince people to use it. Who, exactly, would that be? Some software developers and hobbyists, I mean sure, but then who? End users, even tech savvy ones, are going to be wary of an architecture where the best case scenario is a subset of standard Linux software, to say nothing about the countless Windows and OSX titles, niche hardware, and lack of laptop iterations.
Maybe if it were heavily optimized for database loads it might have a bit of a niche there, but now you have to have someone's name on it. Who is going to be the OEM to sell these machines? Companies aren't buying motherboards and rackmount cases to start using these as servers; Dell or HP or Lenovo will have to get on board, which is rough when Intel has been their poison of choice for so long.
So, in summary, even if everyone volunteers everything they need at every step of the way, what is the expectation? A niche market at best, which will always be treated as a second class citizen, and whose selling point is the great sacrifice made to bring it into existence.
There may be no money in open source hardware NOW - but the argument is that so many people have now been shafted by closed source hardware that some might be interested in an alternative.
No one is talking about "free as in beer".
Someone who wants to break into the CPU market (can you say "Chinese") might want a sales pitch that can overcome their current image problems ("it might keep phoning home") - and open source is probably the ONLY tool that can crack that particular nut. And "It keeps phoning the mother ship" (or hackers.ru) is becoming a blunderbuss aimed deep into the heart of closed source. Meltdown and Spectre are "me, too" arguments in reality. You don't need bugs for closed source kit to be a security risk - you cannot tell if it is a security risk even in the absence of bugs. As any reliable and honest crook/casino owner will tell you: "if you can't tell if you are being cheated: you are being cheated".
A lot of large customers buying expensive kit is a very valuable market. "Not being American" is potentially a massive potential selling point to the 90% of the world's population that is not American, but without open access to the design, very few will buy into a product from any of the potential alternative suppliers.
However, I suspect what is really needed is not just open source, but also with a credible second source (eg from two different BRIC countries).
Sent from my ASR33 using ASCII
That's more than an order of magnitude higher than the NREs we were paying for the ASICs (including sea-of-RISC network processors) the last time I was doing ASICs - abouit 5 years back.
Has it gotten that expensive? I sincerely doubt it. But even if it has:
It has IF you want the absolute top-end pervformance fabbing. You can still get fabbed off the bellding edge much more cheaply. Intel's introduction of finfets hailed a massive change in the industry which had not been seen before: it was the first time the cost per transistor increased.
SJW n. One who posts facts.
There's nothing in the spec about implementation - you're free to recreate Meltdown and Spectre and be fully compliant as far as I can tell - so I can see no benefit.
What we are going to need going forwards - if we're serious about battling malicious software - are things like more protection rings (or similar) and hence faster mode-switching, better memory protection, container-oriented virtualisation (including better support for DMA), and possibly realising that we now have sufficient memory to run kernels mostly without address translation. That will probably involve some sort of Virtual Memory system in which an Address Space ID is part of the address for both cache efficiency and protection purposes. I don't think we'll get them, because it would involve significant changes, not to the silicon, but to the mindset of Operating System developers, most of whom seem to have been desperately reinventing Multics for the last 50 years.
Designing the architecture and logic is fraction of the engineering effort necessary to design and build a modern high end microprocessor.
In addition, a high end processor needs a complicated motherboard to run it, with high speed memory, and various peripheral I/O systems, driven by separate ASICs, or integrated in the CPU. A desktop PC motherboard is a very complex design, which is only made affordable by huge volumes.
If you want cutting edge, yes.
If you can stay back a few nodes, not so bad, a few million bucks is needed (masks are expensive at about $100K/each for the older processes), so perhaps a regular 10 metal chip requires a couple million bucks.
And while most of it is autorouted and autoplaced, you still want to hand edit the designs. Remember the reason we're at 10 metal is because for most general random logic, the limiting factor is wiring. The vast majority of transistors in any design is used in memory - caches and the like. The density falls off rapidly when you get to general logic blocks like ALUs and registers and other random logic parts. And because of the low density, you edit the layouts to sprinkle in lots of new transistors that do nothing. This is because when you're fixing bugs, you're going to need extra logic parts, and having them already in the silicon itself means only changing a few masks instead of an entire mask set.
That's why there are steppings - A0 to A1 to A2 are metal-only changes and A0 meant a full mask set, but A1 and A2 only had marginal changes so perhaps 5 masks changed (so instead of spending $2M per pop, you spent an additional $1M). But go from A2 to B0 means you changed everything again, so a complete redesign. The reason for this is you ran out of spares, or more likely, the fix involves speed-sensitive paths and you just cannot route the spares in a way to keep the speed up.
Anyhow, I still don't see how an open design would avoid something like Meltdown or Spectre, because those vulnerabilities came about because of novel ways of using the weaknesses of things we did to make processors so much faster, and given the entrenchment of the designs, everyone implemented the same issues, so it's really an issue going all the way back.
(masks are expensive at about $100K/each for the older processes)
For sure. It was kind of entertaining when I was onsite at Infineon's fab in Munich many years back with a team installing one of our femtosecond-laser defect repair systems and one of our guys (not me, I swear!) got a little careless and put his thumb through the pellicle on a mask. The customer was not pleased.
Please stand clear of the doors, por favor mantenganse alejado de las puertas
Eh? Germany already has a lot of tracks deployed, that 25000 km was China *catching up*, its also helpful that as an authoritarian regime they can pay next to nothing, ignore the environment and seize land for pennies.
As far as exports, China's rail technology is effectively subsidized by the government, cheap labour, and lax labour laws.