Andy Frame, Sr. Director, Product Marketing—June 01, 2022
SiFive - The Market Leader In RISC-V Vectors
SiFive is the market leader in RISC-V Vector processors and has gained significant market traction, with solutions being designed into a broad range of applications ranging from computer vision, mobile ISP, Edge AI, to datacenter AI.
As SiFive vector processors become the go-to workhorse processor for modern workloads, they are being chosen for stand-alone and companion processor designs in both large and small systems. As a growing list of customers use our designs, they feed back to us their requirements for next-generation devices, and SiFive is enhancing the current portfolio as well as investing heavily in new vector processors, and a long-term roadmap to take advantage of this exciting new solution. Current customers include:
- A U.S. Federal agency for next generation aerospace/defense standards
- Tier #1 global mobile image processor solution
- Tier #1 global hyperscaler for AI datacenter/learning products
- Tier #1 storage product developer
- Tier #1 global automotive infotainment solution
- Large U.S. semiconductor company implementing an eFPGA system
- U.S. eFPGA company expanding into AI DataCenter applications
- Tier #1 U.S. aerospace application requiring state of the art microelectronics
Tenstorrent, an AI semiconductor and software start-up developing next-generation computers, announced the use of the SiFive® Intelligence™ X280 processor in their future AI training and inference processor.
Kinara (formerly known as DeepVision) licensed the X280 (and other IP) to enhance the flexibility and functionality of their products to better support customers building applications for the fast-growing Internet of Things (IoT) markets like smart city, smart retail, automotive, and industrial.
SiFive vector processors are gaining momentum across these top-tier customers who are seeing the value and a number of key benefits. As customers share their requirements, in most cases the SiFive Performance™ P270 and Intelligence X280 fit perfectly well, and our roadmap instills the confidence that SiFive is looking to the future with further vector products in development now.
Whatever your vector needs, SiFive can help! In addition to bringing trusted enterprise-ready RISC-V architecture, reduced BOM costs, and open standards, there are several key benefits that SiFive vector processors bring to designers, including:
- Design flexibility
- Power and performance efficiency
- Ease of programmability
- Use of standard software
Let’s look at each of these in a bit more detail.
With RISC-V Vector Extensions, SiFive customers have tremendous freedom in both the way they design and also in the applications they pursue. The RISC-V ISA offers a compelling level of flexibility, and SiFive has capitalized on this with our product portfolio. SiFive currently has two vector products, one with 256-bit wide vectors and another with 512-bit wide vectors, along with an exciting roadmap taking us further into the higher performance market space.
Due to the nature of the RISC-V architecture, SiFive is able to build fit-for-purpose products that align with customer requirements and application workloads. The RISC-V ISA baseline instructions number just 47, with the ability to add extensions as workloads require. This extends to the RISC-V Vector ISA, with its base ISA (numbering much less than a packed-SIMD implementation) and the ability to complement this with new workload-specific instructions, and then work with the RISC-V International team to bring this into the architecture as required.
Power and Performance Efficiency
As we have seen with the flexibility of the RISC-V architecture, this also maps onto building fit-for-purpose products for our customers. Optimal system design achieves the right balance of performance for the application workload within a defined power constraint; when processing large data sets, it's about the optimum level of both scalar and vector performance. By selecting single core or multi-core, multi-cluster systems, companies are able to map this onto the required SiFive vector processor to achieve maximum performance efficiency, while meeting or exceeding their requirements for minimal area and power.
One of the most unique and powerful capabilities of the RISC-V Vector Extension is that the ISA is vector-length agnostic. This means that code written for a SiFive vector processor can run on any compatible RISC-V vector ISA processor, so it’s possible to experiment with different vector sizes to achieve the perfect balance of performance, power, and area for different application workloads. Also, with a growing base of open source software for vector processors, projects are easily portable between processors and applications for the same reason.
Ease of Programmability
A number of companies currently have designs that include one or more separate custom DSPs for performing specific application tasks. These custom DSPs needed to be designed to co-exist with the main processor, with some subtle design consequences, requiring in many cases the DSP to:
- require access to, and use of the main system bus (for data transfers), potentially causing bus latency issues
- need its own cache system, for maintaining peak performance, with an associated increase in chip area
- require a separate proprietary toolchain for programming, along with the separation of algorithm tasks
With the combined scalar and vector compute capabilities offered by SiFive vector application processors, companies have been able to consolidate the custom DSP functionality into the vector processor and still achieve performance and efficiency goals.
Use of Standard Software
RVV is an open standard, and much of the code written for RVV will be available in the open source domain. This allows developers to access the large and growing ecosystem of RVV-based algorithms, along with access to a full range of open source and commercial grade tools for compilation, modeling, debug, and trace. The upside and key benefits of this include:
- access to open source algorithms
- standardized and stable approach to algorithm development
- reduced development costs
- and, crucially, enabling much faster time to market
Want to learn more? Contact us or access the other materials on our SiFive vectors page.