SiFive launches RISC-V chip design for high-performance AI workloads

SiFive, a chip designer primarily based on the RISC-V computing platform, introduced the launch of a collection of latest AI chips for high-performance AI workloads.

The SiFive Intelligence XM Sequence is designed to speed up high-performance synthetic intelligence workloads. That is SiFive’s first IP and features a extremely scalable AI Matrix Engine that accelerates semiconductor firms constructing system-on-a-chip options for edge IoT, client units, next-generation electrical and/or autonomous automobiles, knowledge facilities and extra The plan’s time to market.

As a part of SiFive’s plans to assist clients and the broader RISC-V ecosystem, SiFive additionally introduced its intention to open supply the reference implementation of its SiFive Core Library (SKL).

The information was introduced at a SiFive press convention in Santa Clara on Tuesday, the place executives mentioned the management position RISC-V structure performs on the core of synthetic intelligence options from totally different market leaders and supplied The most recent information, roadmap and enterprise momentum from SiFive’s technique.

open options

SiFive CEO Patrick Little mentioned in an interview with VentureBeat that clients within the semiconductor, system and client markets have begun to understand the software program methods behind SiFive and RISC-V.

He famous that greater than 10 billion SiFive cores have been shipped thus far. Little famous that SiFive has invested greater than $500 million in analysis and growth and sells merchandise to prime semiconductor leaders and hyperscalers. The corporate has gained greater than 400 designs.

The RISC-V structure has an open requirements interface for software program, which implies any sort of core will be linked to it. Little mentioned this implies clients designing with SiFive can select their very own accelerators for AI and different purposes with out worrying about breaking software program compatibility.

He mentioned that whereas massive leaders in synthetic intelligence resembling Nvidia can use their very own proprietary graphics processing unit (GPU) structure, smaller firms use their very own accelerators. However Little mentioned software program programmers do not wish to be taught a brand new language each time a brand new accelerator comes out. That is why superscalar and chip firms wish to use RISC-V options like SiFive so they do not should continually rewrite software program, he mentioned.

The RISC-V open requirements software program interface permits the RISC-V commonplace to evolve gracefully over time, and it reduces the danger of options past a single proprietary vendor.

SiFive has been steadily shifting up the meals chain, beginning with embedded cores within the Nineties and including its first vector processor in 2021. Clients can use it as an information stream processor as a front-end to the processor to match the ever-changing back-end synthetic intelligence accelerator.

“They did not wish to maintain writing knowledge into the AI ​​software program. So we put a RISC-V vector processor in entrance of it. AI processors maintain altering quickly. Fashions maintain altering. Software program writers count on to be writing it 15 years from now one thing,” he mentioned. “We’re one of many few firms that may fill that hole. At the moment we introduced our personal accelerator or matrix multiplication engine, and we’re creating the XM product line to do what we do in vector processing. It’s a matrix multiplication engine.

Clients who need a substitute for Nvidia can flip to a different supply, however they do not need that competitor to turn out to be one other proprietary resolution. As an alternative, they like RISC-V as a result of it has so many competing firms behind it, Little mentioned.

“We imagine our resolution can scale to Nvidia-level efficiency,” he mentioned.

“Many firms are seeing the advantages of open processor requirements as they race to maintain up with the fast tempo of change in synthetic intelligence. AI leverages SiFive’s performance-per-watt benefit and our distinctive means to assist clients customise their options. “We have now Offering RISC-V options to 5 of the Magnificent 7 firms, we’re working with firms starting from automotive to knowledge facilities to develop new AI options as firms transfer to a “software program first” design technique. Web of Issues.

SiFive’s new XM collection gives an especially scalable and environment friendly AI computing engine. By integrating scalar, vector and matrix engines, XM Sequence clients can benefit from very environment friendly reminiscence bandwidth. The XM collection additionally continues SiFive’s custom of delivering extraordinarily excessive efficiency per watt for compute-intensive purposes.

“RISC-V was initially developed to successfully assist specialised computing engines, together with mixed-precision operations,” SiFive chief architect Krste Asanovic mentioned in an announcement. “This, coupled with the addition of environment friendly vector directions and specialised Assist for AI scaling is why lots of the largest knowledge heart firms have adopted RISC-V AI accelerators.”

As a part of the presentation, Asnovic launched extra particulars in regards to the new XM Sequence, which expands its vary of sensible merchandise. The XM collection additionally continues SiFive’s custom of delivering extraordinarily excessive efficiency per watt for compute-intensive purposes.

With 4 X cores per cluster, the cluster can ship 16 TOPS (INT8) or 8 TFLOPS (BF16) per GHz. Every XM Sequence cluster of this chip has 1TB/s of sustained reminiscence bandwidth, and the cluster can entry reminiscence by a high-bandwidth port or by the CHI port for constant reminiscence entry. SiFive envisions making a system that doesn’t embody a bunch CPU or relies on RISC-V, x86, or Arm. The corporate is now sampling its resolution.

SiFive will attend the RISC-V North America Summit on October 22-23, 2024 in Santa Clara, California. The corporate presently has 500 staff.

“We have now turn out to be the gold commonplace for RISC-V,” Little mentioned.