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."
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.
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.
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 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.
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."