NextSilicon’s Maverick-2 Challenges GPU Dominance with Revolutionary Energy Efficiency

Revolutionary Architecture Redefines Compute Efficiency

Israeli chip startup NextSilicon is making bold claims about its Maverick-2 accelerator, positioning it as a serious competitor to established players in the high-performance computing space. The company asserts that its innovative Intelligent Compute Accelerator (ICA) architecture delivers superior performance while consuming significantly less energy than competing solutions from industry giants.

Revolutionary Architecture Redefines Compute Efficiency
Energy Savings Take Center Stage
Architectural Innovation Behind the Performance
Expanding the Portfolio with RISC-V Development
Market Implications and Future Outlook

According to recently released benchmarks, the Maverick-2 demonstrates remarkable efficiency gains. NextSilicon’s internal testing reveals the accelerator achieves 4x the performance-per-watt compared to Nvidia’s HGX B200 GPU for FP64 computations, and an even more impressive 20x performance-per-watt advantage over Intel’s Xeon Sapphire Rapids CPU for the same workload type.

Energy Savings Take Center Stage

The energy efficiency narrative extends beyond performance metrics to practical cost considerations. NextSilicon claims that total energy costs for Maverick-2 installations are less than half those of equivalent Nvidia HGX B200 deployments. This substantial reduction in operational expenditure could prove compelling for data center operators and research institutions facing escalating power budgets.

In specific benchmark testing, the Maverick-2 demonstrated its capabilities across different workload types. For AI inferencing scenarios measured in GUPS (Giga-Updates Per Second), the accelerator achieved 32.6 GUPS at 460 watts – representing a 22x speed improvement over traditional CPUs and nearly 6x faster than competing GPUs. The system also showed competitive performance in HPCG (High-Performance Conjugate Gradients) benchmarks, reaching 600 gigaflops at 750 watts while consuming approximately half the power of leading GPU solutions., as covered previously, according to technology insights

Architectural Innovation Behind the Performance

Elad Raz, founder and CEO of NextSilicon, explained the fundamental shift in design philosophy: “For eight decades, rigid hardware designs have forced software to adapt, with modern CPUs dedicating roughly 98 percent of silicon to overhead and just two percent to actual computation. Maverick-2 flips this paradigm on its head by devoting the majority of hardware real estate to compute, moving runtime overhead management to intelligent algorithms and software in real-time through our dataflow architecture.”, according to additional coverage

The Maverick-2 leverages TSMC’s 5nm process technology and comes in two configurations:

Single-die PCIe format with 96GB of HBM3e memory and 300W maximum power consumption
Dual-die Open Accelerator Module (OAM) with 192GB of HBM3e memory and 600W maximum power consumption

Expanding the Portfolio with RISC-V Development

Alongside the Maverick-2 announcements, NextSilicon revealed its development of a new RISC-V test chip named Arbel. This strategic move toward the open-source instruction set architecture represents a significant departure from industry trends that have predominantly favored Arm-based designs.

The company described the RISC-V selection as a “strategic decision that goes beyond simple architectural preference,” instead characterizing it as “a fundamental shift toward an open and adaptable future.” This approach will enable NextSilicon to develop proprietary intellectual property from the ground up, potentially creating unique competitive advantages in the evolving processor landscape.

Although specific performance details for Arbel remain undisclosed, NextSilicon has already claimed that the chip “surpasses current RISC-V competition, as well as Intel LionCove and AMD Zen 5.” The company plans to release comprehensive benchmark data as the project advances beyond the test-chip phase.

Market Implications and Future Outlook

NextSilicon’s ambitious claims arrive during a period of intense competition in the accelerator market, particularly as AI and HPC workloads continue to evolve in complexity and scale. Raz emphasized the long-term value proposition: “As AI and HPC workloads evolve, Maverick-2’s intelligent, adaptive design ensures our customers’ investments remain future-proof while delivering unparalleled performance today.”

Founded in 2018, NextSilicon has secured substantial financial backing, raising over $200 million across four funding rounds from investors including Third Point Ventures, Playground Global, Aleph, and Amiti Ventures. This level of funding suggests significant confidence in the company’s technological approach and market potential.

The semiconductor industry will be watching closely as NextSilicon moves toward broader deployment of its Maverick-2 accelerators and progresses with its Arbel RISC-V development. If the company’s performance claims hold up under independent verification, it could signal a meaningful shift in the high-performance computing landscape, offering organizations new options for balancing computational power with energy efficiency in an increasingly power-conscious world.