We spent a lot of time at the RISC V Summit this week, and want to share our overall impressions of the project. Here is some background on the project, we have been writing a lot about it lately. We think it is definitely interesting, almost certainly important, but probably not the revolution that some make it out to be.
A few highlights of the conference to set the scene. The Summit was fairly crowded. Not quite the wall to wall madness of CES, but a large crowd, and we found we were constantly bumping into people we know, which is a sign of something. There were about 50 companies on the exhibition floor. They were roughly divided into three product types: chip designers, IP providers and software tools. Notable among the chip designers were Microchip and T Head, Alibaba’s internal chip division, who were demonstrating Android running on their RISC V TH1520 processor running Linux and Android. There were also a large number of companies announcing “Edge AI” chips.
The software providers included companies like Imperas who make EDA design and verification tools for RISC V. And probably the most significant announcement of the event was the keynote from a Synopsys executive launching their full support for RISC V. This is important in that the EDA/IP vendors like Synopsys are hugely influential in the process of taking chips from design to manufacture.
However, Synopsys’s entrance into the project also presents a problem for another large contingent of Summit exhibitors – the IP providers. We counted at least a dozen companies providing some form of IP for making RISC V chips. These range from the largest firms like SiFive, Andes and CodaSip to many much smaller ones providing specific features or custom designs. The sheer number of them is a sign of both RISC V’s progress and some pitfalls that lay in store. There is clearly a lot of interest in RISC V from all kinds of chip designers, and the presence of these IP providers show that the ecosystem is building a commercial layer around the “open source” roots. However, it is clear that there are too many of these vendors. Synposys’ entry into the market does not mean that all the rest are doomed, but it is likely to promote a wave of consolidation resulting in fewer vendors.
Earlier this week, we published a note on the prospects for RISC V for mobile phones. This is pretty good proxy for the overall state of RISC V. There is tremendous energy and interest in the initiative, but still a lot of bugs to work out of the system.
Our core thesis on RISC V holds that it is doing well in some end markets and not others:
- Embedded and IoT: The market for chips for industrial systems, low-powered computers, random IoT devices like cameras and the like is going to see a lot of RISC V. Arm’s upfront licensing costs are too high for these products to justify the economics, and RISC V’s flexibility allows for just the right amount of customization needed in the category.
- Mobile phones: We will see pilot projects some time soon, but their impact will be limited. Mobile will probably never move to RISC V, but if it ever does, it is going to take a decade.
- Edge AI: It’s 2023, so every chip company that wants to get funded needs to mention AI. There were about a half dozen AI accelerators on display at the show. This is a good use case for RISC V, but it already a very crowded market.
- Data Center: There are about a dozen companies making some form of processor for data centers (including CPUs, GPUs and AI accelerators). We wish them luck, but this in going to be a very hard market to crack for merchant chip vendors. The key challenge in this market is providing software support in the form of thousands of infrastructure software companies porting their software to run RISC V. We write about this subject a lot, and while there are clear signs of progress in the broader ecosystem, exactly zero of these companies came to the event. We will reassess this view when the RISC Summit has a whole hall dedicated to software companies. We are likely to see RISC V cores sitting along side Arm and x86 cores in future products (e.g. Apple A Series, Google’s TPU), but stand alone RISC V processors face giant obstacles.
- Automotive: We have tended to be open-minded about RISC V in cars. Our sense is that many of the electric vehicle (EV) companies in China who are building their own automotive processors want to lean heavily on RISC V. If this is true, there was no sign of it at the summit. Outside of China, we think none of the auto incumbents are going to spend too much time on RISC V, it is just too far from their core capabilities. So overall, we are lowering our expectations here.
Overall, we remain upbeat about RISC V. It is not going to displace Arm or x86 everywhere, but that is the wrong benchmark. It is going to become much more prevalent in all kinds of devices, even if we are not totally aware of it.
That being said, the ecosystem still faces two significant, related challenges. We wrote earlier this year about the software fragmentation that is essentially built into the “open” roots of RISC V. Every time we speak with someone from a RISC V company or group, we ask them about the potential for fragmentation, where everyone’s RISC V is a little different. They all tend to respond with some version of the prisoners’ dilemma – it is in no one’s interest to get too fragmented from the rest of the project, which will dampen the centrifugal forces and keep the herd together. This is a fair argument, but it dovetails into another problem we see peaking its head over the horizon – politics. RISC V is run by a committee and every other joint planning body in the history of technology ultimately trips into some form of political war. The best example of this was the standards wars fought in the 1990’s around the wireless standards, but there are many other examples. The decisions as to which features and extensions get adopted into the core standard are going to become more important and more valuable, and this is going to lead to conflict. It is too soon to tell how the sides will shape up for this, let alone predict a winner, but we are already seeing signs of small scale tension building. It is tempting to portray this as a ‘war’, but it probably will not play out in quite so dramatic a fashion. Nonetheless, it is important to keep an eye on how the group holds together in coming years.