If you're just joining us, Arm very much sees the up and coming RISC-V stack as an immediate threat to the future of their business and is taking pro-active countermeasures - doing everything from awkward, backfiring smear campaigns against RISC-V[1] to straight up license dumping their own product to prevent developers jumping ship.
It's interesting to see because RISC-V is wonderful and Arm seems to recognize that. Arm Holdings is acting rationally as someone in a privileged competitive position would do. At the same time, big players like Western Digital are migrating to RISC-V so Arm is internally freaking out [2].
What I found interesting initially was how the 32 bit CPU was disrupting the 8 bit CPU market. People using Cortex M0 chips where they used to use ATMega chips. Identical package size, faster, more resources, not too much more memory. And some unreal bargains[1].
And now as FPGAs get cheaper and cheaper, RISC-V was becoming the default 'soft' processor because everything else cost money and didn't have any sort of ecosystem. Amazing times.
Big and small players are interested indeed, from WD, Alibaba and Huami to 'some 300 companies already looking at or developing with RISC-V in China'[1], but not because RISC-V is so wonderful, it's just not bad and just enough for many tasks - and it's basically free.
It's probably impossible to satisfy the level of 'wonderfullness' you'd demand in a modern (even if minimalist) processor. There are a lot of competing demands on what it should do. RISC-V seems to have made good decisions w.r.t. modularity and efficiency at least.
What I don't understand is does it really cost a company like WD that much to license or just buy already manufactured chips?
It would seem taking risks on new architectures, and all the extra development time (both chip and tooling) would far outweigh the cost of just licensing arm.
Can someone explain the economics of it all?
Don't get me wrong, I am all for a free competitor to ARM's near monopoly on the mobile embedded space. I don't just get it.
The killer your missing is 'architecture license' versus 'IP' license. The former gives you the right to modify the architecture in your designs (add instructions, non-standard peripherals, alternate memory buses, Etc.) a standard 'IP' license just lets you essentially plop down a standard chip and innovate in the peripherals without touching the system controller at all.
Folks like WD want a processor that is cheap, and specialized like running an error correcting algorithm across spectrum. Generic CPU is too slow, custom CPU is fine. Custom CPU where there is already a lot of other support because other people are using it, is super fine.
Unlimited right to use / modify is a really valuable thing for these companies.
I think the issue is really just one of scale. If you ship 400M drives a year (as the global hard drive industry did last year), a cost savings of $1 is a lot of money that goes into your pocket/R&D. If you're making 10000 of something, making a custom chip is almost certainly going to cost you a fortune; if you're making tens or hundreds of millions of something, that cost gets amortized very quickly, I imagine.
If WD is making their own chips, it almost certainly isn't to save a couple grand a year; it's probably to save many millions.
Yeah, in consumer electronics the hardware engineers have knife fights in the hallways over resistors. Well, not that bad. Probably.
"Can't you do that in software?" 50M units later you understand the reason why that hardware guy made your life a living hell for two weeks just before release, because what he did saved ten cents on the motherboard and at scale that's real money.
I spent a little bit of time Logitech's R&D department back in the late 90's, they were doing all they could to shave two pins off a chip. It cost them a terrible amount of money, including a custom package and the difference in price per mouse must have been on the order of a few pennies. But the enormous amount of devices shipped more than made up for the R&D investment.
Well WD has total sales of $15.13bn with a cost of goods of $10.8bn if the chip and license for arm costs $3 and they can get the costs to $1 with the chip doing exactly and only what they need at that scale it’s a meaningful impact on cost of goods thus bottom line.
It is not uncommon to have lots of simple processors in one chip. One could manage low power mode, another encryption and boot, and another core functionality required in a hard drive. When you have 3-4 cores per device the license fees start to add up.
When you get large enough, doing it yourself can become cheaper than buying it, and small %0.5 improvement projects can have big benefits since they accumulate.
RISC-V feels like CPU hardware's linux moment right now.
Kinda off-topic, but I feel like they could gain a lot by opening their graphics stacks or simply by letting FOSS devs work on them, instead of making their work more difficult/impossible (see: history of the FOSS lima driver).
Yes but if it is patented that doesn't mean it is secret -- in fact the opposite. I can copy the code but I can't sell a product using it without getting a licence or risk being sued. Obviously I can't licence it as GPLv3. But I don't see why the driver implementation should be so closed.
Because you don't hide the driver to hide what's in your patent. You hide the driver to hide code that violates someone else's patent.
'Patent minefield' means there are so many patents for so many basic things it's impossible to do anything without violating something. You'll never know you've violated a patent until your competitor sues you. A lot of these patents are probably invalid, but the fights in court will take years, judgement will sometimes happen by someone without know how, and losing 1 fight will cause you severe damage.
The 'solution' is trying to hide all your violations by closing the driver. Reverse engineering costs a lot, so if someone is doing it to find your patent violations, you build your own patent war chest and cross license with them - they have engineers so they probably produce something that has violations for your patents.
Ah, that makes total sense. Makes you wonder how they avoided this situation with RISC-V (they say via identification of prior art), but graphics does seem to be worse patent-wise (though ARM is notoriously litigious).
Having spent a few years focused on this (and other semiconductor enablement-related problems) my guess is they have evaluated their own patent portfolios and the risk of their implementations triggering a war, and have concluded it was worth doing. Note that Intel's GPUs are relatively simple, so AMD would seem to be bolder here, but I haven't dug into the detail to see how they have limited their exposure.
I find Intel is pretty solid about open sourcing everything they provide for Linux that runs on the host — it's a pity they haven't applied the same policy to the ME.
If you're just joining us, Arm very much sees the up and coming RISC-V stack as an immediate threat to the future of their business and is taking pro-active countermeasures
By any chance, does anyone know if Intel is funding RISC-V? According to the logic of "Commoditizing the Complement" one way of hurting a competitor is to make their product a commodity.
Is there some way ARM could make money supporting an open IP like RISC-V, analagous to Red Hat for Fedora Linux? Obviously, not a perfect analogy, but...
Custom design services. Which is more of a consulting business, so does not scale in same way as IP licensing.
Could also integrate manufacturing services, but fabs is incredibly capital intensive, making it another kind of business.
Could also keep IP licensing as a model, but open the core and push it further and further out into peripherals. Problem is that would probably kill the surrounding ecosystem of partners that do this today.
This is ARM's definite response to RISC-V. As people develop their FPGA products, they will have to decide what their long term target is. If the intent is to eventually turn the code into an ASIC, RISC-V maybe the economically more sensible solution since there is no licensing costs there. However they will have to be patient with the toolchain. If the idea if the code will only be a pet project, wouldn't benefit from RISC's customization abilities, or long term is expected to use an ARM core, this is definitely the right pick. The ARM toolchain being more mature is a draw for many hardware developers.
I'm definitely very interested in trying a project with either RISC-V or ARM.
ARM is releasing the M1 for Xilinx, Altera, Actel FPGAs. These are firm (i.e. technology mapped) cores adapted for each FPGA family.
Here for example is the documentation for the Altera version. Note that it sports the Altera Avalon bus interface which would not be used in a Xilinx design.
There are lots of other FPGAs that are still available for RISC-V to target that are precluded from this announcement.
ARM used to sue anyone trying to make an ARM clone and now this. Great news for RISC-V. At least one if not two low end FPGA manufacturers will be shipping a fabric with hard blocks supporting RISC-V.
Super small CPU cores are fantastic for complex FSM replacement, especially on an FPGA which has unused block RAMs (so using them is essentially free.)
I’ve seen cases with relatively complex pure HW FSM being replace by a small CPU and 1KB of RAM where the logic used by the CPU ends up smaller than the FSM.
But now the CPU is programmable, so you can iterate much quicker in case of bugs, without the need to reaunthesize.
A good example are SDRAM controllers: almost all of them have a small CPU inside the controller that is used for calibration training.
It's not unheard of to throw a soft core in there too, but in the past it'll be something like a Microblaze. An MCU vs. an AP are pretty orthogonal as it's pretty hard to cycle counting real time determinism out of an AP.
"DesignStart Eval enables completely free access to fixed-configuration Cortex-M0 and Cortex-M3 and full subsystem RTL for evaluation, prototype, teaching, and research. Available to all*, the packages can be downloaded instantly, requiring only a simple click-through agreement. "
and if you want to make a product,
"DesignStart Pro enables access to the full Cortex-M0, Cortex-M3, and subsystem RTL. Available to companies for commercial use with a no-risk, $0 license fee and success-based royalty model. Access to the IP can be requested online, requiring the signing of a simplified license agreement."
Came to the comments for the same question. The difference between beer and speech is important here, as the impact of a free-as-in-speech open-source design would be substantial.
That all being said, my sources tell me that tools exist which can decrypt the .vp files. It's still not going to be particularly readable, but at least it's available for analysis.
Hmm. Well, I get the impression the tools you mention are probably less hens-teeth than the ARM bitstream files ever were, so this development is pretty cool then.
The commetization of the CPU market was bound to happen eventually. As modern production techniques for chips and FPGA programming start to converge rapidly.
What is the ETA until we get a GNX "GNX is Not X86"? An Open source i386 core that people can run existing applications on? Much like GNU offered a standardized FOSS platform people could run existing Unix workloads on.
There's ao486 already [1] [2]. I seem to recall that there are a few corners where it's not 100% complete/compatible, but it's complete enough to boot Windows 95 and Linux (assuming a Linux distro with 486 support, now an endangered species), and run some classic DOS games/demos. It doesn't run at a speed equivalent to a "real" 486, at least on a cheap FPGA.
I'm not sure what value there would be in a 368. You can emulate that on a modern CPU if you need legacy software support. Otherwise just use RISC-V, it already runs Linux and thousands of programs. Oh, and it's got a 64bit version too.
Do you know of any worthwhile patent as in general enough to be reused in a different chip but advanced enough to be a current blocker and not replaced by another technique or is it just to protect i686 proprietary elements ?
I would love ARM releasing ARM1 and ARM2 processors (both with a 3-stage pipeline) for FPGAs, not because of economical interest, but because of historical and coolness purposes. And more, so you could e.g. run a synthesized Acorn Archimedes from a FPGA, for free, running Linux on top of it :-)
That's how the ARM1 was made in the first place: by being simulated. On a BBC Model B. First silicon worked, which nobody really expected but they surprised themselves.
Free != Open Source. Are NDA's/vendor secrecy required? Is the core, in whatever form it is provided in, not obfuscated in any way, and well documented? Does ARM permit companies using it to modify the design?
Free != Open Source.
How suitable is RISC-V for military use? If we were to redesign the F-16 from scratch in 2018, would be able to make it using hardened Arm or RISC-V chips in place of x86 processors?
Several billions of such chips are shipped per year. Many established segments are still moving from 8 bit to 32 bit micros, and new segments are still growing rapidly. So shipments of Cortex-Mx devices is likely to grow quite a bit still.
Cortex-M is still highly relevant in embedded development. While some double-digit Cortex-M cores theoretically exist (Cortex-M23, -M33, and -M35P), there are virtually no parts in production which use them; the current state of the art is Cortex-M7.
If you're just joining us, Arm very much sees the up and coming RISC-V stack as an immediate threat to the future of their business and is taking pro-active countermeasures - doing everything from awkward, backfiring smear campaigns against RISC-V[1] to straight up license dumping their own product to prevent developers jumping ship.
It's interesting to see because RISC-V is wonderful and Arm seems to recognize that. Arm Holdings is acting rationally as someone in a privileged competitive position would do. At the same time, big players like Western Digital are migrating to RISC-V so Arm is internally freaking out [2].
[1] https://www.theregister.co.uk/2018/07/10/arm_riscv_website/
[2] https://www.theregister.co.uk/2017/12/01/wdc_risc_v_edge_str...