RISC-V needs more than an open architecture to compete
Arm shows us that even total domination doesn't always make stupid levels of money
Opinion Interviews with chip company CEOs are invariably enlightening. On top of the usual market-related subjects of success and failure, revenues and competition, plans and pitfalls, the highly paid victim knows that there's a large audience of unusually competent critics eager for technical details. That's you.
Take The Register's latest interview with RISC-V International CEO Calista Redmond. It moved smartly through the gears on Intel's recent Platinum Membership of the open ISA consortium ("they're not too worried about their x86 business"), the interest from autocratic regimes (roughly "there are no rules, if some come up we'll stick by them"), and what RISC-V's 2022 will look like. Laptops. Thousand-core AI chips. Google hyperscalers. Edge. The plan seems to be to do in five years what took Arm 20.
RISC-V may not be an existential risk to Intel, but Arm had better watch it.
That's a bold claim. Redmond is right to point to serious progress – getting buy-in from the industry is an outstanding achievement. RISC-V has earned its place as a contender, there's a buzz about it, and the ecosystem is building bit by bit.
All this is necessary, but prosperity is still a way away. The processor market is a very strange place, and there's no real sign that RISC-V is packing a killer punch.
The challenge isn't one of technical competence or even of being a genuinely useful concept that does things its competition cannot. Both are true: RISC-V is shipping in numbers as embedded and SoC silicon, and it's the only ISA where designers are free to add or modify architectural features at the ISA level itself. Compared to the gargantuan market advantages of its two rivals, however, these seem like running shoes where jet boots are needed.
Take Arm. In 2021 it licensed nearly 30 billion cores – four for every human alive – and made $2.7 billion. Which proves three things. There's a huge global appetite for advanced mobile and embedded cores, Arm absolutely owns that market's IP, yet that total domination doesn't make much money.
You might think $2.7 billion is a lot of money – and its profit margin is nothing to sneeze at either – but it's only 0.6 percent of the $450 billion worldwide handset revenue, which, given its completely essential nature, is not a large chunk. It's only double the revenue of the other UK chip industry, that of the British potato chip. If there's to be an upheaval of Arm's dominance, it won't be because there's piles of cash on the table.
- Only Microsoft can give open-source the gift of NTFS. Only Microsoft needs to
- Privacy pathology: It's time for the users to gather a little data – evidence
- Your AI can't tell you it's lying if it thinks it's telling the truth. That's a problem
- Twitter faces existential threat from world's richest techbro
There is a pile of cash in Intel's $80 billion annual haul, but that's because of the huge chip factories that Intel owns. Not up for grabs. What matters to RISC-V is that Intel has a very similar lock to Arm on its sector's ISA IP. And the ISA IP sectorization is very distinctive.
Intel and Arm are rulers of their own fiefdoms, ones which despite decades of trying to poach from each other have little overlap. Last year's Mac Mini M1 was the first mainstream Arm desktop computer since Acorn gave up the Archimedes fight in the early 1990s. You could write down all the Arm desktop PCs in the intervening decades on the back of a postcard. (No, the Raspberry Pi is not a mainstream desktop PC. We'll get to that later.)
As for Intel-based smartphones, choose that category if it comes up in a quiz. Ten minutes on Google and you'll be the undisputed world champion expert. Neither fiefdom has a runner-up, not even the ISA of the other side. There are no second places in the big processor markets. Incumbent inertia, the decades of momentum in sector tools, knowledge and installed base, combined with massive investment in innovation just to keep either fiefdom at bay, has turned the mainstream CPU market into two monopolies kept honest and energetic by each other. Want to compete with either? Good luck.
For RISC-V to really prosper, it needs its own sector. AI/ML could be one; there are lots of accelerator designs which are nicely heterogeneous and the industry has yet to pick a winner.
IoT? Maybe, but that's no spring chicken. Gaming has gone through its experimental early stages and is locked into high-performance standard architectures. In truth, the big players are well positioned to divide any new sector between them.
Even if a new sector does appear, RISC-V has to have a major enabling factor. Open source doesn't provide enough financial or technical advantage – most OEMs will be licensing cores, just like with Arm, not designing their own. Modern SoCs can add a lot of architectural innovation without changing the ISA.
To be blunt, RISC-V needs price-competitive performance. A month ago, the VisionFive, the first RISC-V SBC, appeared. It's just like the Pi, only bigger, slower, and more expensive. The Pi created its own sector because it was smaller, faster, and much cheaper than its rivals. It was based on a Broadcom SoC, where years of fearsome competition in mobile phone technology will get you that advantage: RISC-V has nothing to leverage.
RISC-V's not going to be a huge success by out-Pi'ing the Pi. But it needs to out-something something, and in a really arresting way. It doesn't much matter what that something is, but without it, RISC-V will always be in second place where there's only room for first. ®