Maverick-2: Introducing Intelligent Acceleration
At NextSilicon, innovation drives everything we do. Today, I’m incredibly proud to share a major milestone in our journey—the emergence from stealth with our launch of Maverick-2, the industry’s first Intelligent Compute Accelerator (ICA). This launch embodies countless hours of dedication from our team and deep collaboration with our customers, marking a significant evolution in capability from our original Maverick products. Together, we are bringing a bold new vision to life, reshaping the future of computing, and setting the stage for a new era of intelligent compute acceleration.
At NextSilicon, we’ve invented a novel and original computing architecture to tackle the unique demands of compute-intensive workloads with our intelligent, software-defined hardware. Maverick-2 redefines accelerator architectures, moving beyond the limitations of traditional fixed CPU and GPU designs based upon the eight-decade-old Von Neumann architecture. This breakthrough combines sophisticated, first-of-its-kind hardware, a 5 nm dataflow ICA, and intelligent runtime software, allowing us to power a broad range of applications. Designed specifically for high-performance computing (HPC) workloads, Maverick-2 adapts in real-time for unmatched performance and power efficiency. Whether it’s HPC, AI, or vector database applications, its intelligent compute architecture is designed to handle them all. By blending the flexibility of software with our advanced hardware, Maverick-2 addresses the growing demands for efficient performance, scalability, and flexibility which are required in modern data centers. NextSilicon's patented technologies, which uniquely combine software-defined flexibility with cutting-edge hardware, are powering today's innovations–and setting the stage for tomorrow's breakthroughs.
An Inconvenient Truth - Traditional Architectures Reaching Their Limit
The demand for compute power has never been greater. Large chip vendors and countless startups are spending billions in R&D and venture capital every year to create new solutions. But despite all this investment, the latest accelerator designs are reaching over 1000 watts to operate, creating a significant challenge for organizations to deploy them at scale. On top of that, even power grids are struggling to keep up with the new power demands, pushing large organizations to explore alternatives like nuclear energy. That means that many organizations are left on the sidelines, unable to compete in a fast-paced landscape where insights and innovation from HPC and AI are the keys to success.
These architectures are also constrained by their fixed designs and features. Built on the Von Neumann architecture – an 80-year-old sequential processing model – today’s traditional many-core architectures struggle to adapt to the changing demands of new workloads. These CPU, GPU, and alternative hardware accelerator designs are defined years before they actually run their intended workloads, with rigid instruction sets that introduce significant overhead. This challenge has only gotten worse with the rapid evolution of new approaches in HPC and AI. A good example of this is the shift from neural nets to transformer models for generative AI. With this fast-paced change and growing enthusiasm around AI, GPU vendors have shifted their focus to AI capabilities for lower precision training and inference. This shift has left HPC customers - who need high-precision compute -- being forced to settle for solutions that are inefficient, costly, and inflexible.
The challenge is further amplified by closed software ecosystems. To move and port HPC applications to GPUs organizations have to rewrite large portions of code using domain-specific languages (DSLs) like NVIDIA CUDA. This process can take months–or even years and requires employees with highly specialized skills, all without any guarantee of a significant breakthrough. Once deployed, achieving and maintaining optimal performance involves a lot of trial-and-error tuning. It’s complex, time-consuming tuning, and costly, making it difficult for organizations to explore alternative platforms, and innovate.
As demand for HPC and AI systems continues to grow, these limitations will become more evident, slowing down insights and scientific progress– and impact competitiveness across industries. Ultimately, society bears a cost as well, with slower advancements in areas that really matter like climate modeling, renewable energy, and drug discovery.
A New Way Forward - Unlocking Innovation with Maverick-2 ICA
Maverick-2 is a novel and smarter approach to compute acceleration—one that is adaptive, efficient, and ready for the future. The Maverick-2 ICA is designed to focus on the critical codes and workflows that run the majority of the time in HPC and AI environments. Its self-optimizing architecture delivers immediate, scalable performance improvements by creating software-defined processor cores tailored to the performance needs of each application and then executing them on the accelerator hardware. As your application runs, telemetry data is fed to NextSilicon’s intelligent algorithms, to continuously self-optimize performance, power consumption, and utilization in real-time. This ensures maximum efficiency and sustained performance across HPC, AI, and vector database workloads. Driven by our advanced telemetry and patented algorithms, Maverick-2 ICA continuously optimizes power and performance, delivering real-world results that surpass theoretical peak performance claims.
The result? Over 4x the performance-per-watt of traditional GPUs and more than 20x the performance of high-end CPUs—all while cutting power consumption and related costs by up to 50%. This leap in efficiency reduces data center costs and extends infrastructure life, allowing organizations to sustainably innovate and stay competitive in a constantly evolving landscape.
As we have discussed in previous blogs and articles, a core challenge remains: even with access to a more efficient option, breaking free from vendor lock-in is often difficult. At NextSilicon, we knew there had to be a better way, so we set out to remove this barrier– enabling portability and eliminating the complexity of closed ecosystems.
So how does NextSilicon eliminate these seemingly insurmountable porting roadblocks that have slowed the adoption of other alternatives? We achieve this by combining the Maverick-2 ICA and its software-defined hardware architecture with intuitive, powerful developer tools. These tools include out-of-the-box support for popular languages and frameworks such as C/C++, FORTRAN, OpenMP, and Kokkos - eliminating the need for DSLs. By supporting these portable formats, customers can break free of vendor lock-in. Even better, this means that supported applications can run on Maverick-2 without extensive code modifications, allowing customers to see results faster. For large-scale scientific simulations, this means customers can save weeks, months, or even years of effort.
And this is just the beginning. NextSilicon is committed to continually lowering the barriers to adoption for other new and emerging HPC and AI use cases. This will include our planned support and compatibility with CUDA, HIP/ROCm, and leading AI frameworks– making Maverick-2 an ideal solution for the most demanding HPC-AI applications. This flexibility and ease of use make Maverick-2 a powerful acceleration choice across industries like finance, healthcare, energy, manufacturing, life sciences, and scientific research.
A New Era of High-Performance Computing
The launch of Maverick-2 isn’t just a technical achievement—it’s the start of an exciting new chapter in high-performance computing. As industries face increasingly complex workloads, the need for intelligent, adaptable solutions has never been greater. We built Maverick-2 not just to tackle today’s challenges, but to deliver an intelligent hardware and software solution that can grow and evolve alongside our customers’ needs.
Whether it’s the rapid evolution of AI models or more sophisticated scientific simulations, NextSilicon’s intelligent compute accelerator architecture is designed to push the boundaries of what’s possible. I am genuinely inspired by the breakthroughs Maverick-2 will help researchers achieve, unlocking discoveries that were once out of reach.
I want to take a moment to thank our customers, partners, investors, and the entire NextSilicon team. Your dedication, creativity, and collaborative spirit have made this vision for our company and solutions a reality. Together, we’ve built something truly special, and I can’t wait to see what comes next as we help organizations and industries unlock their next wave of innovation and discovery.
Our future is bright, and we’re just getting started - with many more exciting details and announcements coming soon!
Stay Connected
Stay tuned as we share more details in future blogs and announcements via our Nextletter.
About the Author:
Elad Raz is the founder and CEO of NextSilicon, a company pioneering a radically new approach to HPC architecture that drives the industry forward by solving its biggest, most fundamental problems.
