It's mighty weird to think of Broadcom as capable of buying Qualcomm. It'll be interesting to see if RISC-V activity increases after a merger like this. NXP has been a member, Qualcomm has been a member..
I don't think this really fundamentally changes the availability devices, or their leverage on the market (aside from Qualcomm's maneuvering becoming part of Broadcom's activities). I don't see why people in here are calling for regulatory involvement. This does not form any more of a monopoly than there already has been (on part of Qualcomm being basically being the sole enabler of LTE).
I'm glad you mentioned RISC-V. Even if that doesn't displace ARM or x86, it means every company can have their own CPU designs (RISC-V makes CPUs a commodity). That means the only remaining IP blocks of high value are graphics and wireless. Every other part of a SoC is either available or easy enough to develop in-house. Got that? Graphics IP will keep a company profitable for the next 5-10 years, and both these companies have basic graphics capability. Wireless will keep pretty much THESE TWO companies profitable for the next 20 years. Their merger would result in just one company that could build a complete phone on a chip without licensing much - if anything - from outside.
Definition of commodity: A reasonably interchangeable good or material, bought and sold freely as an article of commerce.
That's just one, but none of the other definitions I can find suggest any type of IP ownership or licensing model. Don't get me wrong, what ARM did was a great thing. I just think what's happening around RISC-V is even better.
Apple refuses to sell their designs on the commodity market, so they can basically be ignored as a CPU/SoC vendor. Also, for what it's worth, Apple's CPU designs are not strictly "better" than the competition, they really just have a different target. Other vendors have more focus on dynamic and idle power, Apple seems to care a lot about single-core top line throughput.
That said, if you look at the way applications for Watch OS are submitted, you'll note that the submission format is LLVM bitcode, not machine code. This seems like an obvious and direct statement that they are willing to drop ARM at any moment for the Apple Watch. Something similar could (probably) be done for iOS.
> Other vendors have more focus on dynamic and idle power, Apple seems to care a lot about single-core top line throughput.
If that's true, it's actually more embarrassing for the other vendors, since iPhones get comparable battery life to Android phones with batteries half the size.
Apples and oranges. The Apple phones get better battery life because of stricter app store inclusion rules and native binary builds for all targets instead of generic automated ports.
Plus the rest of the materials BoM on the phones probably cherry picks parts that are very efficient and tuned to work well with each other.
Why aren't Androids in this same boat? My guess is that it's because Samsung wants to release 10 different Galaxy phones a year to have something new and shiny in every product segment instead of focusing on one REALLY polished design and leaving last year's model for their mid/low end.
> Apple phones get better battery life because of stricter app store inclusion rules and native binary builds for all targets instead of generic automated ports.
Apple does just as well in comparisons that only measure the first party browser.
Apple wins in battery life because they sell $900+ phones and invest large sums in engineering specialized chips while Qualcomm doesn't sell phones, they sell chips to customers whose main concern is price.
Not really. You'd have to go through the same motions. Either way it'd end up looking like later versions of MACRO-11 which treated PDP-11 assembly as a full language to be compiled for other architectures rather than just an assembler.
>> RISC-V is an instruction set architecture right? What do you mean by it will make CPUs a commodity?
Yep. The spec is publicly available for anyone to use. There are also several instances of CPUs that implement the instruction set available to download and incorporate into your designs at no cost. There are companies selling services to help integrate it into SoC devices and such.
Until now, if you wanted to put a CPU on a chip your choices were basically ARM, MIPS, and a couple others. You had to license the core from someone or pay to license the instruction set. Today you can integrate RISC-V with no royalties to anyone. Everyone can have a CPU on their chip.
In terms of performance, the available designs are already beating ARM in terms of power and area metrics at the same or better benchmark performance (at the low end). The BOOM cores (also available for free) are closing in on the ARM high end performance metrics. I have little doubt that a high end design could match or beat the performance of the best Intel or AMD chip, but that has yet to happen. My point is that it's not only freely available, but that it's also competitive with some of the best out there. That combination should make CPUs a commodity in the near future.
If anyone is still paying for a CPU license in 2020 I think they will have missed the boat.
>> The BOOM cores (also available for free) are closing in on the ARM high end performance metrics. ... match or beat the performance of the best Intel or AMD chip,
How does this happen, when many/most good ideas in cpu design(a mature industry) should be patented by now?
>> How does this happen, when many/most good ideas in cpu design(a mature industry) should be patented by now?
The RISC-V instruction set was designed to be free of any existing patent issues and with no specific implementation in mind. The opcode formats were designed with hardware implementation in mind however (minimizing the number of multiplexers and such). The big ideas in implementation have been around for a long time. Boom is a multiple-issue out of order core, but nothing about it looks particularly novel yet. It could still benefit from result forwarding and some other things, but that's an old concept too. I think lot of patents in processors tend to revolve around things that are specific to the ISA rather than broad concepts. The reason for that is that we've been making computers for 50 years now so many of the major concepts are rather old already.
I may be off base here too. I wouldn't be surprised if things like a micro-op cache are still patented, but I also wouldn't be surprised if that concept is 35 years old either.
It may be that getting top performance does require patented ideas, but I'd be thrilled if Intel or AMD gave it their best shot. It would probably perform better than their x86 offerings on regular code.
It's royalty free, which (in contrast to ARM's long and expensive licensing process, or MIPS's) allows designers and manufacturers to more cheaply and fairly enter the market. RISC-V eliminates the last true bottleneck/point of control in the CPU market.
>> Aren't good Graphics IP cores availble to licence from imagination or ARM ?
Yes they are. But general availability at a price is not the same thing as free. I suppose ARM had made their CPUs nearly a commodity, but it's still not the same as open source designs available for free. Graphics and RF seem to be the last holdouts with more vendors offering graphics IP as far as I can tell - or at least having IP even if they don't license it. For RF you don't have much choice.
Most of the value of RISC-V will not really be in free designs though. The GPU is already a commodity because of higher-level standardization in APIs (Vulkan, OpenGL [ES]). It's just hard to design and build good GPUs, which means fewer players in the high end space. Vivante seems like the only really independent GPU design vendor, not sure if you can license Mali for a non-ARM design, you'd at least need to support their peripheral buses.
I don't think this really fundamentally changes the availability devices, or their leverage on the market (aside from Qualcomm's maneuvering becoming part of Broadcom's activities). I don't see why people in here are calling for regulatory involvement. This does not form any more of a monopoly than there already has been (on part of Qualcomm being basically being the sole enabler of LTE).