> It tries to undercut the ARM model of establishing how CPUs or other cores should work and how they should be valued.”
Companies are not going for ARM just to license professor ISA's. They also license and value the underlying efficient and high performance architecture ARM has developed.
What can happen with RISC-V is that there are several competing IP companies that develop competing RISC-V architectures and ask royalties and license fees for their IP. This competition can bring down the cost somewhat. If ARM royalty is 1.5% today, the cost may be 0.77% in the future. It's also possible that one company will dominate others with superior design, price and foundry connections. That company might be ARM, AMD or Intel.
> Companies are not going for ARM just to license professor ISA's. They also license and value the underlying efficient and high performance architecture ARM has developed.
Depends on the company. Apple, Qualcomm, Samsung, AMD and Nvidia in fact do go to ARM just to license the ISA and implement the CPU entirely on their own. However, the ARM implementations are licensed by a whole lot of smaller companies. The bigger players are thus pretty much forced to forever chase ARM's changes to the ISA and implement these. I doubt they mind ARM's price but I suspect that they might be concerned about ARM's influence.
One big change from RISC-V, if it materializes, would be a standard ISA not controlled by a single company like ARM, and not evolving through a knife fight between a few companies racing to extend it, like x86. Will it materialize? Not sure which way I'd bet.
I'm excited by RISC-V. It has some interesting architectural decisions.
Remember the ARM started in the 1980s was solid but not a breakout architecture until the early 2000s. The x86's roots lie in the 70s and it took a killer app (IBM PC) to make it dominant. GCC took about a decade to get any traction.
The article says that the developers expect companies to add their own features and support, so I guess we're just waiting for some established player or startup to pick up the ISA and roll something out with it?
But it specifically says "same model as Linux", and that bothers me because RISC-V is not an operating system, and I'm not sure that Linux's business model is appropriate here...
RISC-V is just an ISA, which can be thought of a s a very low level API. The memory (and most other subsystems') architecture is not tied to the ISA. Rather it has to do with load/store instruction implementation and cache organization.
People like to compare RISC-V to Linux, which I think it is just wrong.
Linux is something that you can download from kernel.org, compile it and bring it up over night. It's a package with a bunch of scripts that compiles a working kernel for your machine. All the work is already done for you.
RISC-V, on the other hand, it is just a document describing an ISA. It is far different from a working implementation.
RISC-V might shine on micro-controllers and on power management control units, since those applications are more simple and more affordable to implement from scratch.
But on high-end applications, it will be no different from ARM's path. Implementing an high performant CPU costs money, someone will have to cater those costs, either by hiring a full team of highly specialized engineers (which will cost a bunch of money) or by licensing to third parties (which will also cost money through licenses or royalties).
You can download a RISC-V core today, "compile" it and run it on an FPGA. You can tweak it or add stuff to it, test on the FPGA and go to custom silicon when ready. Not all designs can be downloaded for free, but some can and there are several you can license already. It's not as simple as software like Linux but it is the hardware equivalent.
As the article (and you) point out, the datacenter and mobile space is already lost to Intel and ARM respectively. The only path forward is through the IoT space. It's the only area where you might need something even smaller than ARM.
"Not everything exists for RISC-V that exists for the other ones, but that is filling in at incredible pace."
The RISC-V project is now 7 years old. Remember that the 6502 and its supporting hardware went from proposal to final silicon in 2 years. And that was done using a hand drawn layout.
UC Berkeley has had functional RISC-V silicon for years now as research devices. RISC-V hasn't really been a thing outside of their architecture research group for more than a few years now. The UCB-BAR also has some fairly decent RISC-V cores (BOOM and Rocket) available as open source RTL.
Another worthy contender to be rehabilitated is DEC Alpha. Indeed, the Sunway Taihu Light supercomputer (fastest the on TOP500 list) uses the Sunway SW26010 which is based on the Alpha 21164.
I didn't read the article, but I have been trying to stay vaguely informed off and on.
The two 'cons' that really need to be addressed:
* Average consumers need /access/ to purchase working solutions (which means some prosumers and some developers will).
* Working solutions probably need to include: DisplayLink, USB, Ethernet, and /maybe/ WiFi (the later two /could/ just be USB devices) ports on the hardware. Standard bulk IO (like SATA, PCI(e) bus) would be nice to have, but not required.
Issues you mention have nothing to do with RISC-V ISA.
RISC-V Is just open source ISA, hopefulĺy without any patent issues.
Nobody is going develop open source high performance RISC-V processor architecture IP and give the design away free. You pay licenses and royalties just like before.
[+] [-] Nokinside|8 years ago|reply
Companies are not going for ARM just to license professor ISA's. They also license and value the underlying efficient and high performance architecture ARM has developed.
What can happen with RISC-V is that there are several competing IP companies that develop competing RISC-V architectures and ask royalties and license fees for their IP. This competition can bring down the cost somewhat. If ARM royalty is 1.5% today, the cost may be 0.77% in the future. It's also possible that one company will dominate others with superior design, price and foundry connections. That company might be ARM, AMD or Intel.
[+] [-] yosefk|8 years ago|reply
Depends on the company. Apple, Qualcomm, Samsung, AMD and Nvidia in fact do go to ARM just to license the ISA and implement the CPU entirely on their own. However, the ARM implementations are licensed by a whole lot of smaller companies. The bigger players are thus pretty much forced to forever chase ARM's changes to the ISA and implement these. I doubt they mind ARM's price but I suspect that they might be concerned about ARM's influence.
One big change from RISC-V, if it materializes, would be a standard ISA not controlled by a single company like ARM, and not evolving through a knife fight between a few companies racing to extend it, like x86. Will it materialize? Not sure which way I'd bet.
[+] [-] cwyers|8 years ago|reply
[+] [-] gumby|8 years ago|reply
Remember the ARM started in the 1980s was solid but not a breakout architecture until the early 2000s. The x86's roots lie in the 70s and it took a killer app (IBM PC) to make it dominant. GCC took about a decade to get any traction.
So it's early days for RISC-V
[+] [-] ShannonAlther|8 years ago|reply
But it specifically says "same model as Linux", and that bothers me because RISC-V is not an operating system, and I'm not sure that Linux's business model is appropriate here...
[+] [-] tgragnato|8 years ago|reply
EDIT: coherence -> consistency (sorry)
[+] [-] jabl|8 years ago|reply
[+] [-] jackyinger|8 years ago|reply
[+] [-] sweden|8 years ago|reply
Linux is something that you can download from kernel.org, compile it and bring it up over night. It's a package with a bunch of scripts that compiles a working kernel for your machine. All the work is already done for you.
RISC-V, on the other hand, it is just a document describing an ISA. It is far different from a working implementation.
RISC-V might shine on micro-controllers and on power management control units, since those applications are more simple and more affordable to implement from scratch.
But on high-end applications, it will be no different from ARM's path. Implementing an high performant CPU costs money, someone will have to cater those costs, either by hiring a full team of highly specialized engineers (which will cost a bunch of money) or by licensing to third parties (which will also cost money through licenses or royalties).
[+] [-] phkahler|8 years ago|reply
[+] [-] throwaway40483|8 years ago|reply
[+] [-] morio123|8 years ago|reply
The RISC-V project is now 7 years old. Remember that the 6502 and its supporting hardware went from proposal to final silicon in 2 years. And that was done using a hand drawn layout.
[+] [-] baobrien|8 years ago|reply
[+] [-] yjftsjthsd-h|8 years ago|reply
[+] [-] redtuesday|8 years ago|reply
[0] https://en.wikipedia.org/wiki/SuperH
[+] [-] CalChris|8 years ago|reply
Another worthy contender to be rehabilitated is DEC Alpha. Indeed, the Sunway Taihu Light supercomputer (fastest the on TOP500 list) uses the Sunway SW26010 which is based on the Alpha 21164.
[+] [-] bobsam|8 years ago|reply
Open ISA and a full open source implementation by MIPS themselves?
[+] [-] mjevans|8 years ago|reply
The two 'cons' that really need to be addressed:
* Average consumers need /access/ to purchase working solutions (which means some prosumers and some developers will).
* Working solutions probably need to include: DisplayLink, USB, Ethernet, and /maybe/ WiFi (the later two /could/ just be USB devices) ports on the hardware. Standard bulk IO (like SATA, PCI(e) bus) would be nice to have, but not required.
[+] [-] Nokinside|8 years ago|reply
RISC-V Is just open source ISA, hopefulĺy without any patent issues.
Nobody is going develop open source high performance RISC-V processor architecture IP and give the design away free. You pay licenses and royalties just like before.