The key CPU architecture you've never heard of: RISC-V

The RISC-V CPU architecture currently accounts for under 1% of the world’s processor market, but that is going to change rapidly over the next years as its parallel processing is perfectly suited to running energy efficient AI.

Since the 90s x86 and ARM have dominated computing, ARM in the embedded markets and x86 in personal computing. Both have multiple licensees making processors using them, though with different business models; ARM's business model is based on licensing its architecture rather than on manufacturing, while Intel owns x86 and licences it to AMD largely out of necessity to avoid being a monopoly. 

However, they are not the only CPU game in town. A number of technology companies including Intel and Microsoft invested $10 million in the Parallel Computing Laboratory at University of California Berkeley to develop parallel processing systems. This research built on a lot of prior art and culminated in the RISC-V microarchitecture specification. Development continued at the UC Berkeley ASPIRE Lab with significant investments from DARPA and the US Air Force among others. 

The RISC-V Foundation formed in 2015 to build a community around the RISC-V Instruction Set Architecture (ISA), and formally launched a collaboration with the Linux Foundation in 2018. In an effort to free itself from geo-politics, the RISC-V International Association was incorporated in Switzerland with a focus on building the RISC-V ecosystem.

Here’s one of the reasons why people should pay attention to it: The RISC-V Foundation has a community process for accepting, reviewing, and ratifying updates and extensions to the RISC-V ISA. One notable example of how this works is the new crypto extensions. Though the base RISC-V ISA can perform encryption on its own, it requires over 1000 RISC-V instructions to implement an encryption algorithm such as AES128 but only 75 using the scalar Crypto extension, and the vector Crypto extension that is close to ratification lowers that to just 1. 

Since all three are standards, a company making an embedded RISC-V processor can leave out the Crypto vector extension to conserve die area and power consumption, a higher performance RISC-V processor can implement the vector extension, yet it's possible to build software that can run on both and if one of the extensions is available, use it to boost crypto performance. Which is all pretty neat.

$1 billion investment to cope with AI

Though there are over a billion RISC-V processors in the wild today, RISC-V presently accounts for under 1% of the world's processor market so it's no surprise that it rarely if ever makes the news.

That is about to change.

RISC-V International CEO, Calista Redmond, was a keynote speaker at the Open Source Summit for North America in Vancouver, BC in May 2023. She highlighted some of the recent developments in the RISC-V ecosystem and community, and shared some interesting new developments.

One of the biggest ones is a new member of the RISC-V community: Intel.

It's no secret that Intel is no longer the top CPU dog in the world. It's also no secret that Intel has been building a foundry business aiming to compete with TSMC. To that end, Intel has invested $1 billion in RISC-V and is already bringing customers on board to the nascent Intel Foundry Services (IFS).

One of the biggest markets for new processors is of course AI. Currently nVidia dominates the AI datacenter market, with Intel, AMD and several ARM licensees such as Nuvia (now part of Qualcomm) and Ampere. Google, Meta, Amazon, and AliBaba also are developing custom silicon for their own datacenters. Most of these are based on ARM, GPUs, and x86 presently, but the massive growth in this market is demanding ever more computing power and memory, and with it ever more energy. 

That is becoming a big problem, because AI is, as you are no doubt aware, growing like a supernova. However, because RISC-V is designed for parallel computing rather than for single threaded computing like ARM and x86, it's far more efficient than both of them, leading to substantial savings in energy costs when it comes to running AI.

And as a result of that in Calista Redmond's keynote, she showed a projection that over the next two years RISC-V would go from the current sub 1% market share to a whopping 26%.

All of which suggests that x86 and ARM might finally have met their match.