Newly Revealed RISC-V Vector Unit Could Be Used for AI, HPC, GPU Applications
semi-dynamics introduced One of the industry’s first RISC-V vector units, a highly parallel processor used for things like Artificial Intelligence (AI), High Performance Computing (HPC), and graphics processing when equipped with appropriate dedicated IP can be used for . This announcement marks an important milestone in the development of the RISC-V ecosystem.
Semidynamics vector units are fully compatible with the RISC-V vector specification 1.0 and offer a number of flexible additional features that enhance data processing capabilities. A VU consists of multiple “vector cores”, equivalent to AMD, Intel, and Nvidia GPU cores, designed to perform multiple computations simultaneously. Each vector core is equipped with an arithmetic unit that can perform operations such as addition, subtraction, fused multiply-accumulate, division, square root, and logical operations.
The vector cores developed by Semidynamics can be customized to support various data types such as FP64, FP32, FP16, BF16, INT64, INT32, INT16 and INT8, depending on the specific needs of the customer’s application. Customers can also choose between 4, 8, 16, or 32 core options for the number of vector cores embedded in the vector unit, offering a wide range of potential tradeoffs between power and performance areas.
Semidynamics’ vector unit features a high-speed network connecting all vector cores to facilitate data shuffling for certain RISC-V instructions. Semidynamics uniquely allows customization of the Vector Register Bit Size (VLEN), offering 2X, 4X, and 8X ratios in addition to the standard 1X. Vector operations take multiple cycles when VLEN is greater than the total data path width (DLEN), which helps manage memory latency and reduce power usage.
“This will allow vector units to process unprecedented amounts of data bits,” said Roger Espasa, CEO and founder of Semidynamics. “And to fetch all this data from memory, our Gazzillion technology has the ability to handle up to 128 concurrent data requests and track them to the correct location regardless of the order in which they are returned. Combining our technologies will complete RISC-V.” A new level of fastest big data processing available today for high-performance computing such as video processing, AI and ML. opportunities open up in many application areas of “
Semidynamics says its vector unit can be combined with Atrevido’s out-of-order cores (and upcoming in-order cores) to build CPUs and other applications that require both vector and general-purpose processing. That’s it.
“Our recently announced Atrevido core is unique in that it allows us to perform ‘open core surgery,'” said Espasa. “This is unlike other vendors’ cores that can only be configured from a set of options, we actually expose the core, modify the inner workings to add features and special instructions, and make it completely bespoke. It means creating a solution, and we’re taking the same approach: the new vector unit perfectly complements the core’s ability to quickly process large amounts of data.”
Semidynamics is one of the few companies developing IP for AI and HPC applications. Besides SiFive, only Tenstorrent develops high-performance RISC-V IP that can be used to build processors and AI accelerators. On the other hand, unlike Semidynamics, Tenstorrent offers not only licensable IP, but also its own hardware processor.
