“AMD congratulates SPEC on the release of its new SPEC CPU® 2026 benchmark suite and applauds its continued leadership in advancing open, standards-based benchmarking. Enterprise buyers, CIOs and IT decision makers need data that is transparent, reproducible, and grounded in real applications. SPEC’s vendor-neutral methodology gives customers a fair way to compare platforms, align processor choices to their specific needs, and make better- informed, workload-optimized infrastructure decisions.”

– Robert Hormuth, Corporate Vice President, Architecture and Strategy, AMD

“As modern servers scale to hundreds of cores and support diverse workloads simultaneously, benchmarking must evolve to reflect the heterogeneous environments seen in real data centers. The SPEC CPU 2026 benchmark suites represent an important step forward for the industry, expanding the scope and relevance of performance measurement across modern hardware and software environments. Ampere was proud to contribute to the development of the RRR heterogeneous execution model and to work with the SPEC community to advance more representative performance evaluation.”

– Naren Nayak, VP of Customer Engineering at Ampere Computing

“As computing systems continue to evolve to support increasingly complex and diverse workloads, trusted, vendor-neutral benchmarks are essential for establishing a common baseline for CPU performance. The SPEC CPU 2026 benchmark suites reflect the growing scale, diversity and architectural innovation across modern computing, and will play an important role in helping the industry evaluate and optimize next-generation infrastructure.”

– Frédéric Piry, Vice President of CPU Technology & Fellow at Arm

"The SPEC CPU 2026 benchmark suites represent a critical step forward in standardizing how the industry measures CPU, memory, and compiler performance against the demands of modern software ecosystems. As computing environments grow more complex, having a vendor-neutral, up-to-date benchmark gives organizations the clarity they need to make smarter infrastructure decisions."

– David Schmidt, VP, Product Management, Dell Technologies

“As Agentic AI applications experience explosive growth, the industry's demand for intelligent computing power continues to surge. Serving as the central orchestrator within computing systems, diverse CPU chips assume critical functions such as data preprocessing, task scheduling, workflow management, logical operations, and ensuring stable system performance. They are the essential cornerstone for the efficient, stable, and large-scale deployment of AI workloads. Given the wide variety of application scenarios and an industry landscape where x86, ARM, and emerging instruction set architectures coexist, establishing a cross-architecture, cross-scenario, fair, and open performance benchmarking standard has become a crucial prerequisite for fostering a diversified computing ecosystem and accelerating industry innovation. The release of SPEC CPU® 2026 marks a significant milestone in the field of global CPU performance benchmarking. Its multi-architecture-compatible testing framework and real-world scenario-oriented workload design provide global computing products with a more authoritative benchmarking reference that closely aligns with actual business operations. Furthermore, it serves as a roadmap for upstream and downstream industry partners to continuously optimize server performance and enhance product competitiveness.

As a service provider deeply engaged in the development of the SPEC community and its standards over the long term, IEIT SYSTEMS remains committed to a philosophy of diverse and open technological innovation. We have actively participated in workload selection and design, test suite development and validation, the formulation of evaluation rules, and the co- construction of the testing ecosystem, providing solid support for the successful rollout of the latest generation of SPEC CPU standards. Looking ahead to a future industry landscape driven by diverse computing power, IEIT SYSTEMS will continue to actively contribute to SPEC community standards and drive innovation in diverse intelligent computing. By leveraging technological innovation to propel product upgrades and utilizing high-quality computing infrastructure to serve global customers and industry development, we are dedicated to fostering the stable, and sustainable evolution of the global computing ecosystem.”

– John Hu, General Manager, IEIT SYSTEMS

"As a longtime SPEC member, Intel values our partnership with the organization and the important role SPEC CPU benchmarks have played in advancing industry-standard performance evaluation. We're pleased to support the development of the CPU 2026 suite, which will continue to provide a reliable, consistent benchmarking framework for meaningful CPU performance assessments."

– Shervin Kheradpir, Vice President, IP & Competitive Analytics, Intel

854.graph500. David Bader: "As graph-based workloads become increasingly central to machine learning and AI—from graph neural networks to knowledge graphs powering modem LLMs—the inclusion of 854.graph500 in SPEC CPU2026 reflects a critical shift in what 'general- purpose' computing must handle. Working with the SPEC committee to adapt Graph500 was a genuinely symbiotic process: their rigorous hardening for portability and determinism produced fixes we upstreamed to the community, while SPEC gained a benchmark that captures the irregular memory access patterns and data-dependent parallelism that define this growing class of applications."

767.nest. Hans Ekkehard Plesser: "Working with the SPEC committee to prepare the NEST simulator for SPEC CPU2026 has been a great experience. The SPEC team tested NEST on a much wider range of operating systems, compilers, and hardware than we have available in our regular test and benchmark setup. The small number of incompatibilities revealed in this process provided good learning opportunities for us. Stress-testing of multithreading in NEST by the SPEC CPU team complemented our own efforts and contributed to further ascertain the thread-safety of NEST. As an added bonus for us and NEST users, the SPEC CPU team even contributed some small optimizations to the simulator. We are excited that NEST, having served as a reference for neuromorphic computing systems over the past decade, will help to drive CPU performance as part of SPEC CPU2026."

734.vpr. Vaughn Betz: "Working with SPEC has helped ensure the large VPR code base is fully compliant with recent C++ standards and achieves consistent results across platforms. The collaboration with SPEC identified several instances of platform dependent or compiler dependent code and result differences, which were fixed collaboratively and upstreamed to ensure the community could use VPR on the widest range of platforms with confidence. We also believe the hardware design community will benefit from having CAD tools like VPR represented in the SPEC benchmark suite, as that will help drive future CPU improvements to speed up these compute-intensive tools."

737.gmsh. Christophe Guizaine: "Working with the SPEC committee has been a very constructive and technically rewarding experience. The process of adapting Gmsh into a benchmark—defining representative workloads, improving portability, and hardening the code for diverse platforms—has led to concrete improvements that directly benefit the project and the Gmsh user community. Overall, beyond visibility, the collaboration has helped us better understand performance-critical paths and portability constraints, with benefits that extend well beyond the benchmark itself."

753.ns3. Gabriel de Carvalho Ferreira: "The ns-3 community is delighted to have participated in the selection process for the new SPEC CPU benchmark suite. During the porting and testing phase, we identified several issues that impacted the reproducibility of simulation results, including platform related precision issues, and hindered compilation and execution across diverse platforms. These fixes were promptly integrated upstream. As a result, ns-3's platform support has been greatly expanded to a wider array of systems.

"Crafting workloads was relatively straightforward, thanks to numerous pre-existing simulation campaigns used for de-velopment and analysis under various conditions and con-figurations, and the documentation and tips provided by the SPEC committee members. The committee's analysis of how these workloads differ computationally also provided valuable insights into estimating performance for very long simulations from shorter samples, as well as opportunities to improve the ns-3 simulator code."

838.diamond. Benjamin J. Buch.fink: "DIAMOND solves a computationally hard problem fundamental to biology and contains a lot of performance-critical and carefully optimized code. Working with the committee helped me improve and harden my code with respect to microarchitectural subtleties and compiler peculiarities on different platforms. Due to the needs of scientific computation and the limited resources that most scientists have to work with, both reproducibility and correctness as well as performance are important points for the user community."

727.cppcheck. Daniel Marjamaki: "Our aim is to write truly portable code, and I remember you uncovering - and resolving - a few platform-specific issues while porting Cppcheck. That work was greatly appreciated. You even identified a case of undefined behavior along the way, which was a real bonus. It was a pleasure to be part of this program, and I'd be happy to take part again. I hope our participation will help inspire valuable optimizations in the future."

846.minizinc. Guido Tack: "Working with SPEC on integrating MiniZinc, Gecode, and Chuffed into the benchmark suite was a collaborative and highly constructive experience that we greatly appreciated. The process uncovered approximately a dozen portability, correctness and performance issues, many of which would have been difficult to identify outside SPEC's rigorous cross-platform environment, and the resulting fixes and improvements were incorporated upstream. We are grateful for the care and technical depth of the feedback from the committee, which strengthened the robustness and maturity of our software systems and will provide lasting benefits to our user community, while also helping new users discover constraint programming and adopt our solutions."

800.pot3d. Ronald Caplan: "The process of submitting, adapting, and creating workloads for our POT3D code to be included in SPEC benchmarks has been a rewarding experience for Predictive Science Inc. in several ways. For one, it launched our first major open-source code release, which has paved the way for several more since. It helped our upstream code through the testing process, finding compatibility issues and work-arounds for cutting edge features and hardware. The validation requirements have helped us craft new test suites for several of our codes. By having POT3D in SPEC, we can view the submitted results to preview how our codes will perform across new architectures, and helps guide optimizations. This also helps other groups with similar memory-bound stencil codes."