In a leap forward for AI infrastructure, Arm announced the latest advancements in its Neoverse chip technology based on the Arm computing architecture.
The company unveiled two new Arm Neoverse Compute Subsystems (CSS) built on third-generation Neoverse intellectual property. Arm designs chips or parts of chips for other third-party chip designers and manufacturers.
In this case, the Arm Neoverse family of CPUs offers server-class performance combined with low-power design for high performance-per-watt of energy used.
Arm Neoverse CSS V3
The new designs include the Arm Neoverse CSS V3 product for the high-performance V-series portfolio, boasting a 50% improvement in performance-per-socket over CSS N2.
Arm Neoverse CSS N3 is an extension of the N-series CSS roadmap, providing a 20% increase in performance-per-watt compared to CSS N2. Arm said that its Arm Total Design ecosystem has expanded to include over 20 members in just four months, delivering System-on-Chip (SoC) and chiplet designs across three leading foundries.
“Arm has built the most pervasive CPU architecture the world has ever seen and we’ve led many of the technology revolutions that have impacted the day-to-day lives of people all around the world,” said Mohamed Awad, senior vice president and general manager of the infrastructure line of business, in a press briefing. “Even so, right now is one of the most exciting times in our history. Innovation and transformation are happening across every aspect of technology. From the smallest sensor to the largest data center, compute is becoming more and more specialized. Commodity CPUs are just no longer sufficient.”
These developments showcase Arm’s commitment to pushing the boundaries of AI compute performance, a crucial aspect as the demand for AI applications continues to surge across various industries, including education, healthcare, and manufacturing.
“In infrastructure, we are continuing that transformation to more sophisticated warehouse-scale computing. You see, it’s no longer about chip or server or rack, it’s about the entire data center,” Awad said. “In 2023, we saw that transition accelerate as the world embraced generative AI adoption. And, in 2024 and beyond, we expect massive innovation as AI permeates things like education, employment, manufacturing, healthcare, and transportation. AI is transforming our economy and our daily lives and it’s all happening on Arm.”
The Neoverse CSS advantage
Neoverse CSS compute subsystems offer optimized, integrated, and verified platforms, bringing together key technologies at the heart of a System-on-Chip (SoC).
“Today we are introducing a new CSS product in the N-Series, Neoverse CSS N3,” said Dermot O’Driscoll, VP of product solutions for the infrastructure line of business at Arm, in a press briefing. “Focused on power efficiency, CSS N3 offers a tremendous 20% per core uplift in performance-per-watt over CSS N2. We tune CSS N3 to meet a gap we see in the market for highly efficient compute that meets infrastructure performance requirements.”
Arm said that Neoverse CSS serves as an ideal starting point for partners prioritizing differentiation, SoC optimization, and accelerated time to market. Arm said it engages with partners at a deeper level than other compute vendors, allowing Neoverse CSS to maximize Total Cost of Ownership (TCO) for critical workloads. This includes support for emerging technologies like chiplets, demonstrating Arm’s commitment to staying at the forefront of innovation.
“AI is becoming an integral part of things like networking, security and storage, and it applies to everything from the smallest endpoint through to things like switches, routers, and base stations, the entire infrastructure. At Arm, we are incredibly proud to be at the heart of how technology leaders are solving for the insatiable demand for data and compute brought about by the world’s transition to AI,” Awad said.
Ecosystem expansion with Arm Total Design
Arm previously introduced the Arm Total Design, an ecosystem of partners dedicated to delivering custom SoCs based on Neoverse CSS.
Today, Arm said Arm Total Design has grown to include over 20 partners, all committed to providing accessible and efficient solutions to meet the computing demands of an AI-accelerated future.
These partners collaborate on various aspects, from verifying IP to customizing firmware and building chiplets on the world’s most advanced process nodes, all utilizing Neoverse CSS.
Arm is taking steps towards defining an interchangeable chiplet ecosystem with the recently announced Chiplet System Architecture (CSA). Leveraging feedback from the Arm Total Design ecosystem, CSA sets the foundation for a robust chiplet environment, promising innovations in chip design and integration.
One Neoverse CSS partner saved 80 engineering years worth of effort and another likes to talk about how they went from project kickoff to tape out in just nine months, Awad said.
“This is the best generation of Neoverse technology yet. We are extending our leading N-Series roadmap with a brand new platform and compute subsystem, which takes perf-per-watt to the next level, Awad said. “Simultaneously, we are also launching our next-generation performance-focused V-Series platform and, for the first time, bringing compute subsystems to that product line. With our new N3 and V3 compute subsystems, we are focused on unlocking new technologies like chiplets and maximizing TCO for the real workloads that matter the most to the ecosystem, from AI to databases, networking, and more.”
Some analysts already estimate that as much as 80% of deployed AI servers are dedicated to inference and that number is expected to rise, Awad said.
“This shift will mean finding the right models and model configurations, training them, and then deploying them to the most cost-effective computing infrastructure,” Awad said. “Part of that equation is throughput and we’re seeing some very good token generation throughput on Arm silicon that is deployed and widely available today.”
Partner activity
Arm noted that at least 80 companies building AI accelerators now, including market leader Nvidia, which has a tightly coupled Neoverse V2-based compute chiplet called Grace.
“Plus we added silicon design partners in strategic geographies like Korea, Taiwan, China, and India, here we see tremendous growth opportunity,” Awad said. “We’re engaging all three foundries to ensure our CSS products are optimized for their advanced process nodes. Custom silicon demands access to a reliable supply chain and only Arm provides this level of diversity and choice in the semiconductor supply chain.”
“TSMC has been partnering with Arm for more than a decade. Our transformative partnership continued to trailblaze the industry in pushing the envelope on computing innovations,” said a spokesman for TSMC. “Arm has been a longstanding partner of TSMC and an important part of our open innovation platform, our OIP ecosystem. We look forward to working closely with Arm and our OIP ecosystem partner to bring next generation AI- and HPC-related semiconductor design to market.”
Wrapping it up
Awad said, “The platforms we have introduced today are the best generation of Arm Neoverse. Yet they’ll be the foundation of the next-generation products and services partners build as more AI workloads are adopted and permeate across the entire continuum of compute. From the largest data centers to networks, servers, and all aspects of the computing infrastructure, Neoverse V3 and N3 provide both the computing performance and efficiency needed to underpin this transformation, and our compute subsystems supported by the Arm Total Design ecosystem allow for faster time to market and more investment in innovation than ever before.”
He added, “And Neoverse is just one example of how Arm is delivering AI-optimized computing platforms into every market segment. Arm is unique because we are delivering a relentless pace of technological advancement. We enable the flexibility for partners to innovate and we are supported by a vast ecosystem. Ultimately, that’s why the future of compute, including AI, will be built on Arm.”
Author: Dean Takahashi
Source: Venturebeat
Reviewed By: Editorial Team
