Data Centers
Run cloud and enterprise workloads at high core utilization with lower licensing cost per socket than proprietary ISAs. 64 performance cores to deliver the density and efficiency that modern data centers require.
Server-Class RISC-V.
No Compromises.
Arbel is a 64-core RISC-V enterprise-grade processor, engineered to run data center, HPC, and AI workloads at throughput levels that match or beat x86 incumbents — without vendor lock-in.
Specifications
ISA
RISC-V (RVA23 Hypervisor) with Custom Extensions
Issue Width
10-wide instruction pipeline
Frequency
3.4 GHz
Vector Units
3 x 256-bit Vector Units (SIMD)
Execution
16 Parallel Scalar Instructions
Cores
64 Performance Cores
OoO Logic
480-entry Reorder Buffer (ROB)
RISC-V Leadership.Proven in Benchmarks.
On industry-standard CPU benchmarks, Arbel outperforms today's industry-leading processors per-GHz — the first RISC-V core to do so at server scale. Arbel delivers x86-competitive performance, on an open ISA, with a modern 5nm process.
Arbel target vs. Industry-leading processors, normalized per-GHz (EST)
NextSilicon-derived SPECint2017® / GHz (est.)
Latest Generation Processor
2.3 estNextSilicon-derived SPECfp2017® / GHz (est.)
Latest Generation Processor
3.0 est SPEC CPU® 2017 is a retired benchmark. SPEC is no longer reviewing or publishing new results for this benchmark. See www.spec.org/cpu2017 for more information.
Competitor figures are NextSilicon internal estimates modeled using cycle-level simulation and publicly available performance data as of May 2026. All Arbel and competitor figures are estimates and have not been reviewed or endorsed by SPEC or any third party. Actual competitor performance may differ.
SPEC® is a registered trademark of the Standard Performance Evaluation Corporation. SPECint® and SPECfp® are registered trademarks of SPEC.
Where
HPC
Serves as the host processor for Maverick accelerators in simulation, weather modeling, and genomics. Arbel keeps accelerators fed and utilized, eliminating the CPU bottleneck in heterogeneous compute environments.
AI/ML Orchestration
Handles data movement, preprocessing, and inference scheduling for large-scale training and inference pipelines. Three 256-bit vector units provide native data-parallel throughput without offload overhead.
Edge Servers
Delivers server-grade compute in space- and power-constrained deployments. Real-time analytics and inference at the edge, backed by the same architecture that runs in the data center.
Built from the ground up
Arbel was designed as a clean-sheet server processor. A 10-wide issue pipeline and 480-entry reorder buffer sustain high instruction throughput across complex workloads. Three 256-bit vector units accelerate AI inference and data-parallel computation natively.
The result: consistent, high-utilization performance in the workloads where data centers spend the most cycles.
Production-Ready Software Stack
Arbel runs standard Linux distributions and compiles with LLVM and GCC toolchains. Full RVA23 compliance means existing RISC-V software runs without modification — no porting, no lock-in, no ecosystem tax.
Native support for Ubuntu and standard Linux distributions. Full RISC-V ecosystem compliance (RVA23) enables software portability and ecosystem adoption across vendors.
Arbel already provides customer and partner silicon evaluations, including HPC program leads, AI infrastructure architects, and data center operators. Their input has validated Arbel's key performance attributes and is shaping the requirements and architecture decisions for the 64-core production processor.
