According to Lei Feng, yesterday, Bloomberg reported that SiFive had received an offer from Intel for a $2 billion acquisition. Both Intel and SiFive declined to comment on the news. It is too early to discuss whether the acquisition can be finalized, but through SiFive’s current CEO James Prior’s views on RISC-V entering the AI and data center markets, this may partially explain why such rumors appear.
If scenario-specific data centers dominate in the near term, then RISC-V could shine. While RISC-V is most commonly associated with embedded devices, there is a push to use RISC-V as a foundation for AI and data centers as a springboard for this architecture into larger systems.
So far, RISC-V has achieved little in the data center, but it can serve as the underlying engine for various accelerators. For example, artificial intelligence chip startup Tenstorrent’s inference chip is based on RISC-V, and an ambitious project at the Barcelona Supercomputer Center will use RISC-V (by adopting SiFive’s RISC-V product) to build an on-premises data center — From processors to accelerators.
In an age where options for building or buying CPU cores are relatively diverse, what’s next for RISC-V entering the data center?
The Next Platform spoke with SiFive CEO James Prior, who said the odds of seeing an end-to-end RISC-V data center in the next five years are slim, but there’s definitely a big chance that custom accelerators can surpass Arm, especially In terms of software, tools and support.
“We’re taking a software-first approach to delivering IP cores. In a way that makes sense for programmers, taking into account both core development software and tools, rather than having programmers make our cores clear,” James Prior added.
“Another difference is that SiFive can do this without competing with customers because they are not in the chip design business. We have some directors for partners and companies to evaluate before building large designs. Nvidia vs Arm of acquisitions on a RISC-V trajectory and accelerated decision making – people are moving from asking if they should have a RISC-V strategy to thinking about their plans now. We can then be the leader in commercial RISC-V IP and Experience co-developing chips with architectures that meet specific needs.” While the majority of SiFive’s commercial RISC-V IP products are embedded, their RISC-V CPU IP products have been adopted by 80 companies over the past six years , has designed more than 200 products, including seven of the top 10 tech companies that have shipped more than 1 billion units.
“But as we continue to grow and move into the core areas of applications, we are entering the AI space with new product lines for more general processing and specific functions,” said James Prior.
In fact, SiFive sees AI as a kind of Trojan horse that can make its way into data centers in large numbers. They have developed some custom IP that customers can use to build their own accelerators, an approach that fits with SiFive’s view of data centers moving toward more specialized rather than general purpose.
For developers working on the next generation of AI processors, James Prior said having software and tools that can handle modern data types in a specific application processor with vector capabilities is key. Aligning it with custom AI accelerators for pre-/post-processing and AI math is also more flexible given the rapidly changing AI models. “There are tons of changes in AI models that are much faster than the chip creation process, which means that while you need a dedicated chip for speedup efficiency, you also need general programmability across a set of models.” AI is a good entry point for data center design wins, but James Prior said SiFive is looking at other opportunities, from the edge to the top of the rack network, all from a mature ecosystem with a common set of programming tools benefit from.
“If you look at how x86 scales from simple microarchitecture to complex out-of-order processes to multi-core, it takes a long time. Arm was also unknown at the beginning, and now it is everywhere because they are in our phones “When asked how RISC-V differs from Arm,” James Prior said, “killer applications are coming, it’s going to be in AI and accelerators, even purpose-built systems. If you look at How data centers are changing, people will say they don’t need a bunch of boring cores, they need all the computation balanced in a socket, and that’s where RISC-V can provide value – even processors on major operating systems Instead of running a RISC core, it is performing accelerated value work.” Compiled by Lei Feng.com, link to the original text:
The Links: EL640480-AM1 MP6752