SUBSCRIBE ON:
Introduction to Guest
James Prior is Senior Director of Product Marketing Communications at SiFive, the industry leader in RISC-V processors and silicon solutions, where he leads global communications for the company, helping to plan and manage product and corporate marketing.
Before SiFive, James worked at AMD as part of the Client product management and business development team, helping to define and launch the Ryzen Threadripper HEDT and Ryzen desktop processors.
Previously, James has covered technology companies as a journalist and worked in a University datacenter as a systems programmer for smart grid operations. James received his degree in Electrical and Electronic Engineering from the University of Portsmouth, U.K.
About SiFive
SFive is the first fabless semiconductor company, now enabling modern machine learning architectures. They have 15 design centers across the world with about 550 employees and growing.
SiFive was founded by the inventors of RISC-V, which is a new instruction set architecture. RISC-V was first invented at UC Berkeley in 2010 – researchers at the time were trying to figure out new ways to make computers more efficient and programming more flexible. There wasn’t anything on the market they could just use, all because of proprietary licensing agreements and expense.
RISC-V is a free and open specification that is developed by the RISC-V international organization, which is made up of a bunch of different technology companies, researches, and universities. They guide the RISC-V specification and the extensions to make the set of instructions and the manual on how to build computers. SiFive is exploring the commercial opportunities that the RISC-V instruction set allows. SiFive can create micro architectures, processor cores, and accelerator IP’s that other companies can license from them to build and solve different problems.
SiFive has had a lot of success in the embedded market. Now they’re moving into the application and AI markets with new higher performance and more efficient and accelerated designs.
If you look at industry trends, it’s clear that what got RISC-V/SiFive so far was having a general purpose GPU. A general purpose GPU has done a great job of kickstarting AI and making it into something real, where the next level of power efficiency and performance is going to come from customized solutions. SiFive is seeing this development of custom silicon across all areas of the industry. You have places like Google and Amazon designing their own chips to increase efficiency, performance, and to accelerate parts of the workload. SiFive took a step back and thought: “The people who are writing AI, what are they trying to accelerate? What are their commonalities, how much flexibility do they need?” The big problem for James and his team was that they could easily create an AI model, but by the time they built a chip, the AI model had changed. You need a programmable flexible solution with a little bit of generality, but still the workload acceleration, which is what SiFive created.
SiFive is primarily focused on selling the IP to companies who want to design a chip. SiFive builds a reference platform where it’s just ready for the third-party IP to come in. SiFive supports companies that have a great software stack but want to move away from general purpose computing. SiFive can also support companies that are already doing so, but need help accelerating their roadmaps.
SiFive also partnered with companies like Barcelona Supercomputing Center in Europe where they’re working on the European processor initiative. They have adopted RISC-V and the vector extension for their AI supercomputer. James and his teamwork with them on contributing to the global software (that everyone will use), using libraries and other tools to help ensure the standards and software works.
Challenges Encountered
The day-to-day challenges experienced are software challenges with AI. With RISC-V being a new instruction set, they’re developing a new set of tools, while migrating the existing set of industry supported libraries, compilers, and tool chains.
Day to day the company is really focused and working on the software. There is also a lot of work in deciding which sets of workloads need to be accelerated. SiFive is discovering the biggest amount of impact they can have to accelerate AI, whether it’s in the data center as an accelerator or building a device is through software
What Has Helped the Company Be Successful
It’s a little bit of freedom and a little bit of excellence. By freedom, SiFive is adopting an open standard. It’s about the excellence of their offering, not uniqueness. SiFive has a refreshing approach of offering collaboration: the understanding and the expertise of workload, software development/implementation, and working through challenges. This particular approach has gathered a lot of momentum, especially as they see companies across the world adopt RISC-V. RISC-V is a strategic imperative. Companies realize they need to be able to defend their product roadmap against outside influences. Because RISC-V has open standards, they’re free to adopt. Being an IP provider, SiFive spends a lot of time reaching out to different technology and semiconductor companies asking them: “What challenges are you trying to solve? What are the problems you have today? Where can we help?” This is how SiFive begins to find and discover ways to solve problems, whether working with a big or small company.
https://www.sifive.com/
About the Host
Ari Yacobi is a data scientist, a teacher and a storyteller who has spent his career at…Read the Bio