RISC-V CTO: We won't dictate chip design like Arm and x86
Mark Himelstein talks about the open ISA's future and challenges
Interview Chip technologies from Arm and x86 are getting the most attention amid semiconductor shortages and trade wars. But in the background, the open-source RISC-V chip architecture is stealthily emerging as a viable third architecture that is cheaper, flexible, and free of political intrigue.
RISC-V is often referred to as the Linux of chips with engineers collaborating to design, set, and improve the architecture. RISC-V International sets the spec, which is free to license, and chip designers are free to implement it as they see fit in their processors and system-on-chips.
China and Europe are moving to RISC-V for homegrown silicon to achieve chip sovereignty. Companies are also looking at RISC-V to cut licensing and royalty costs associated with the Arm and x86 architectures.
To be sure, it may be many years until RISC-V emerges as a viable alternative to x86 and Arm in PCs, servers, and mobile phones. But it is off to a good start in microcontrollers and FPGAs, and single-board computers with RISC-V are already available. It also has the backing of companies including Google, Apple, Amazon, Intel, Qualcomm, Samsung, and Nvidia.
RISC-V International's CTO Mark Himelstein spoke with The Register about RISC-V and its progress.
How is RISC-V catching along?
People are doing things that they couldn't with other architectures before RISC-V. It's free. It's flexible. They don't have to pay anybody extra money to customize it. We take advantage of the community and stuff that's very common. We break down corporate barriers, country barriers, cultural barriers, time-zone barriers, and we all share that piece because we are part of this community.
You can innovate in a particular area where we allow for non-standard extensions. It's kind of the best of both worlds.
What are the challenges you face as an open-source chip effort?
It's not like we're dictating like Arm and x86. I ran Solaris for Sun. I've run other big organizations. You tell people what to do. You set a goal and then you go do it. Here, we're making sure we have something coherent that the members can use, and moving forward is important.
We're introducing two concepts, one called profiles and the other common platforms. This is to reduce fragmentation. Profiles are a group of instructions that work together, just like you see generational stuff and other architectures. And those are the things that will tell the upstream projects like compilers and optimizers as well as the distros, "Hey, target this profile, this is the application profile for 2022. These instructions work together." The same thing we're doing for operating systems, so a bunch of things around platforms like discovery, ACPI, SPI.
We've had a whole bunch of individual extensions, and you could either adopt them or not, but there wasn't a whole group where we said, "Look, in this generation these all work together. In this generation, these operating system components, these all work together." [In 2021] we had those. Those are new, and that's very exciting. We're not done. That's in order to make sure we're not duplicating effort, share as much as we can and don't fragment as a community.
We're still learning how contributor culture works in an instruction set architecture, because some of it is architects who are defining this ISA, but has to do with fostering the software ecosystem. We have to make sure that compilers, libraries, operating systems are there, and we're still learning how to do that.
What's coming up for RISC-V?
We had a big backlog and we're knocking that out. The next thing is setting the stage for the next-generation stuff. For example, both our AI-ML special interest group and our graphics special interest group are both interested in advanced matrix operations. That will probably start being under development in 2022. A lot of stuff sits in accelerators, but not in core architectures. You're going to see some of that stuff start going up.
Google just became a premium premier member. Amazon's working on Greengrass. Not to mention Seagate, Western Digital, Nvidia, Huawei, Alibaba. People who produce real products are putting out real numbers – hundreds of millions of cores based on RISC-V in 2021.
Are you chasing GPUs? I think people are just wondering about gaming on RISC-V. Gaming is getting better on Linux.
We're not necessarily going after GPUs. We go after workloads. We have a bunch of members who want the same thing. "Let's go ahead and figure out what it would take to be successful in graphics on RISC-V, prioritize the work and try to get additional stuff done first." People running the special interest group or in graphics said shading is most important. In order to do shading, you need matrix operations. At the same time, the ML guys are coming back and going, "In order to do these ML operations, I need matrix operations." We are combining that effort.
What's the difference between RISC-V and other open-source chip efforts like OpenSPARC or OpenPower?
If the companies behind them went away, they disappear. They can't stand on their own.
If you look at Linux, there were a lot of other operating systems that were open source at the time. Why was [Linux] successful? Part of it was Linus [Torvalds]. The second thing is, people felt they owned it. It wasn't shoved on them. They own Linux. Same thing is going to happen with RISC-V. People feel they own RISC-V, they're helping make it.
Is there any kind of interest in support for Windows on RISC-V? What about Android?
I talked to somebody from Microsoft. They're not there yet. But they're in a bunch of Linux Foundation things. When we get in the room, all we want to do is be good engineers. Is Windows in RISC-V's future? Probably. Is it sometime soon? Probably not.
It's a lot easier to talk about all the variants of Linux – Zephyr, FreeRTOS, Alpine – all the things that you see in the open-source world are going to go first. It's easy for people to port them, and because the tool chains are the same and so on. We've set the structure to be able to support Windows in the future. But again, it really depends on what the community wants. That's where we spend our time and money.
- Intel to put SiFive's latest CPU cores into 7nm dev system to woo customers to RISC-V
- If you wanna make your own open-source chip, just Google it. Literally. Web giant says it'll fab them for free
- Europe completes first phase of silicon independence project
- Intel ‘regrets' offending China with letter telling suppliers to avoid Xinjiang
Android 10 is there, [RISC-V members] are working on Android 12. In the end, Google makes money out of the Google Play Store. Until some handset manufacturer says we're doing it, [RISC-V] will be a second-class citizen. That's the way it works. And we don't feel bad about that. Google's now a premier member – we certainly have talked to them about Android and it just takes time for this ecosystem.
More non-tech industries like automotive are taking chip development in their own hands. Where does RISC-V fit in?
I don't think [chip development] is for the meek at heart. I think you will see people partnering up with RISC-V IP providers, like SiFive or Andes. A bunch of these companies will take your requests and design something for you. Or they'll give you the IP so you can burn it on an FPGA. There's been a renaissance in EDA tools, so it's easier to create an FPGA with a RISC-V processor, or something doing some specialized processing and to pump out an FPGA. That's a lot less daunting.
There are lots of people who want to get involved in RISC-V, but actually need samples of the chip.
Let's separate out the problem. There are some who need real hardware in order to even develop their system or their chip because they need to experiment. From that perspective, this year is a big inflection point because there's a half a dozen or more development boards out there. We've started a seed program, we're trying to give 1,000 boards out to academics, early adopters and distros by June 2022.
We have the RISC-V Exchange on the RISC-V website. There's a bunch of discrete chips, SOCs, boards, software services, that anybody can take advantage of today.
RISC-V is about cutting costs. Are you arranging for RISC-V chips manufactured at low cost?
We're talking about boards, you're talking about chips. We're always working on adding more people into the Exchange. I'm always working on recruiting people. Do we have everybody who's providing RISC-V services? They're absolutely not. Do we have a bunch there already? Absolutely. Will we continue to attract people and add more people in? Yes. RISC-V Exchange is also getting tags so you'll be able to search some of those entries and find what you really want.
Can you cost out a RISC-V chip versus x86 and Arm?
From a cost perspective, it doesn't matter which one you're doing, you have to pay for silicon. That's not going to change. Very clearly places like x86, like Intel, you're paying for both the design and for the chip.
Then you have fabless guys, they will sell you a design and they're going to go to some foundry, potentially one that they have a relationship with, and you're going to negotiate based on volume. That's not going to change. It's going to cost you.
I can't tell you that "here's what it cost to deploy this particular part in this particular industry if you're using Arm versus RISC-V versus x86." It's too complicated.
I suggest you look back at Linux as the example here. Is it cheaper to ship Linux than AIX? How do you do that pricing? I think there are just other attributes of Linux that attract people to using it. It's the feeling they're not going to ever feel out of control. There's a feeling they can influence the product and industry. The same thing is true around RISC-V. ®