pith. sign in

arxiv: 2604.11615 · v1 · submitted 2026-04-13 · 💻 cs.AR · cs.AI· cs.DC· cs.LG

CUTEv2: Unified and Configurable Matrix Extension for Diverse CPU Architectures with Minimal Design Overhead

Jinpeng Ye , Chongxi Wang , Wenqing Li , Bin Yuan , Shiyi Wang , Fenglu Zhang , Junyu Yue , Jianan Xie
show 6 more authors
Yunhao Ye Haoyu Deng Yingkun Zhou Xin Cheng Fuxin Zhang Jian Wang
This is my paper

Pith reviewed 2026-05-10 14:52 UTC · model grok-4.3

classification 💻 cs.AR cs.AIcs.DCcs.LG
keywords matrix extensionCPU architectureAI accelerationGEMM utilizationdecoupled designasynchronous abstractionmixed-precisionRTL integration
0
0 comments X

The pith

A decoupled matrix unit architecture integrates into diverse CPUs with low overhead while achieving over 90% utilization and up to 2.31x speedups on AI models.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that matrix extensions can be unified and configurable by separating matrix units from the main CPU pipeline. This separation keeps coordination with existing compute and memory resources while cutting integration costs across different architectures. An asynchronous abstraction for matrix multiplication hides hardware specifics, eases kernel writing, and permits overlapping matrix work with vector operations. If these elements hold, the approach would make high-performance matrix acceleration practical for open-source CPUs without requiring per-platform redesigns or fine-grained instruction changes.

Core claim

By decoupling matrix units from the CPU pipeline and introducing an asynchronous matrix multiplication abstraction with flexible granularity, the design enables low-overhead integration across diverse CPUs, supports mixed-precision configurable operations, and maintains close coordination with existing resources. Integrated into four open-source CPU RTL platforms, the units exceed 90% utilization on GEMM workloads and deliver speedups of 1.57x on ResNet, 1.57x on BERT, and 2.31x on Llama3 when matched to Intel AMX throughput and bandwidth, with over 30% of gains from overlapped matrix-vector execution; a 4 TOPS@2GHz unit occupies 0.53 mm² in 14nm CMOS.

What carries the argument

The decoupled configurable matrix unit paired with an asynchronous matrix multiplication abstraction that conceals hardware details and enables overlap with vector execution.

Load-bearing premise

Decoupling matrix units from the CPU pipeline while keeping close coordination with compute and memory resources adds only low integration overhead and avoids hidden bottlenecks across varied architectures.

What would settle it

Integrating the design into an additional CPU architecture and measuring GEMM utilization below 90% or speedups below the reported levels due to synchronization delays or bandwidth contention would falsify the central claims.

Figures

Figures reproduced from arXiv: 2604.11615 by Bin Yuan, Chongxi Wang, Fenglu Zhang, Fuxin Zhang, Haoyu Deng, Jianan Xie, Jian Wang, Jinpeng Ye, Junyu Yue, Shiyi Wang, Wenqing Li, Xin Cheng, Yingkun Zhou, Yunhao Ye.

Figure 1
Figure 1. Figure 1: AI Model Architectures and Kernel Fusion Patterns. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: illustrates the hardware architecture. The matrix unit is decoupled from the CPU pipeline and driven by asynchronous matrix multiplication instructions. Depending on ISA and microar￾chitectural support, the CPU dispatches these instructions via a RoCC-like or CSR-based interface, with registers defined in [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Matrix Unit Microarchitecture. Memory Loader generates memory access requests and han￾dles all data reads and writes for matrix computations. Within it, the Request Generator translates the tensor described by matrix multiplication instructions into memory and Scratchpad addresses, generating and issuing the corresponding requests. The Data Re￾order module receives the returned data and reorder it as requi… view at source ↗
Figure 5
Figure 5. Figure 5: Vector and Matrix overlap. 4.4 Low Design Overhead Integration The proposed matrix extension was integrated into the four open￾source CPU RTL platforms listed in [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: GEMM Performance under Different Config. [PITH_FULL_IMAGE:figures/full_fig_p005_7.png] view at source ↗
Figure 6
Figure 6. Figure 6: GEMM Performance on Various CPU Platforms. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: GEMM Performance vs. Existing Extensions. [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 11
Figure 11. Figure 11: Llama3 Performance vs. Existing Extensions. [PITH_FULL_IMAGE:figures/full_fig_p006_11.png] view at source ↗
read the original abstract

Matrix extensions have emerged as an essential feature in modern CPUs to address the surging demands of AI workloads. However, existing designs often incur substantial hardware and software design overhead. Tight coupling with the CPU pipeline complicates integration across diverse CPUs, while fine-grained synchronous instructions hinder the development of high-performance kernels. This paper proposes a unified and configurable CPU matrix extension architecture. By decoupling matrix units from the CPU pipeline, the design enables low-overhead integration while maintaining close coordination with existing compute and memory resources. The configurable matrix unit supports mixed-precision operations and adapts to diverse compute demands and memory bandwidth constraints. An asynchronous matrix multiplication abstraction with flexible granularity conceals hardware details, simplifies matrix-vector overlap, and supports a unified software stack. The architecture is integrated into four open-source CPU RTL platforms and evaluated on representative AI models. Matrix unit utilization under GEMM workloads exceeds 90% across all platforms. When configured with compute throughput and memory bandwidth comparable to Intel AMX, our design achieves speedups of 1.57x, 1.57x, and 2.31x on ResNet, BERT, and Llama3, with over 30% of the gains attributed to overlapped matrix-vector execution. A 4 TOPS@2GHz matrix unit occupies only 0.53 mm\textsuperscript{2} in 14nm CMOS. These results demonstrate strong cross-platform adaptability and effective hardware-software co-optimization, offering a practical matrix extension for the open-source community.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper introduces CUTEv2, a unified and configurable matrix extension for diverse CPU architectures. By decoupling the matrix units from the CPU pipeline, it aims for low-overhead integration while maintaining coordination with compute and memory resources. The design supports mixed-precision operations and uses an asynchronous matrix multiplication abstraction to simplify software and enable overlap. It has been integrated into four open-source CPU RTL platforms, achieving over 90% matrix unit utilization for GEMM workloads. When configured similarly to Intel AMX, it delivers speedups of 1.57x on ResNet, 1.57x on BERT, and 2.31x on Llama3, with more than 30% of the gains from overlapped matrix-vector execution. The matrix unit for 4 TOPS at 2GHz occupies 0.53 mm² in 14nm CMOS.

Significance. If the claims hold, this work would be significant for the open-source hardware community by providing a practical, adaptable matrix extension that can be integrated across different CPU designs with purported minimal effort. The high utilization rates, demonstrated speedups on key AI models, and the small physical area make it attractive for enhancing CPU capabilities for AI workloads. The emphasis on hardware-software co-optimization through the asynchronous abstraction is a strength. However, the significance is tempered by the lack of detailed reporting on the actual integration overheads in the host CPUs, which is central to validating the 'minimal design overhead' aspect.

major comments (1)
  1. [Abstract] Abstract: The central claim of 'minimal design overhead' and 'low-overhead integration' across diverse CPUs is not supported by data. Only the standalone matrix unit area (0.53 mm² for 4 TOPS@2GHz in 14nm) is reported, with no quantitative information on added area, timing path changes, gate count deltas, memory interface modifications, or coordination logic overheads in the four host RTL platforms. This is load-bearing for the decoupling-based approach, as unmeasured platform-specific costs could undermine the 'no hidden bottlenecks' assertion and the cross-platform adaptability claim.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed review and for recognizing the potential significance of CUTEv2 for the open-source hardware community. We address the major comment on the abstract's claims regarding minimal design overhead below. We agree that additional quantitative details on integration costs would strengthen the manuscript and will revise accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of 'minimal design overhead' and 'low-overhead integration' across diverse CPUs is not supported by data. Only the standalone matrix unit area (0.53 mm² for 4 TOPS@2GHz in 14nm) is reported, with no quantitative information on added area, timing path changes, gate count deltas, memory interface modifications, or coordination logic overheads in the four host RTL platforms. This is load-bearing for the decoupling-based approach, as unmeasured platform-specific costs could undermine the 'no hidden bottlenecks' assertion and the cross-platform adaptability claim.

    Authors: We acknowledge that the manuscript reports only the standalone matrix unit area and does not include explicit quantitative deltas for integration overheads (area, timing paths, gate counts, memory interface changes, or coordination logic) within the four host CPU RTL platforms. This is a valid observation and limits the strength of the 'minimal design overhead' claim as currently presented. The design intentionally decouples the matrix units to minimize pipeline modifications, and successful integration across four diverse open-source platforms without introducing reported bottlenecks provides qualitative support. However, to directly address the concern, the revised manuscript will add a dedicated subsection on integration overheads. This will include available platform-specific metrics (e.g., area and gate count comparisons where measured, timing slack analysis, and descriptions of memory/coordination changes) along with a clearer discussion of how the asynchronous abstraction and decoupling reduce hidden costs. We will also update the abstract to reflect these additions. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical architecture proposal with no derivations or self-referential predictions

full rationale

The paper proposes a hardware architecture (decoupled matrix units, configurable mixed-precision support, asynchronous abstraction), describes its integration into four open-source CPU RTL platforms, and reports empirical results (GEMM utilization >90%, speedups of 1.57x/1.57x/2.31x on ResNet/BERT/Llama3, 0.53 mm² area for 4 TOPS@2GHz in 14nm). No equations, first-principles derivations, fitted parameters, or predictions appear in the abstract or described content. Claims rest on implementation measurements and cross-platform evaluation rather than any self-definitional loops, fitted-input predictions, or self-citation chains. The central assertions about low-overhead integration and adaptability are validated by reported RTL integrations and workload results, not reduced to inputs by construction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The design relies on standard assumptions about CPU memory hierarchies and RTL synthesis tools. No new physical entities are postulated. A small number of configuration choices (throughput target, memory bandwidth match to AMX) function as free parameters tuned to demonstrate comparability.

free parameters (1)
  • target compute throughput and memory bandwidth
    Chosen to match Intel AMX for fair comparison; directly affects reported speedup numbers.
axioms (1)
  • domain assumption Decoupling matrix units from the CPU pipeline preserves coordination with existing resources without introducing new bottlenecks
    Invoked in the design description to justify low-overhead integration.

pith-pipeline@v0.9.0 · 5624 in / 1688 out tokens · 43883 ms · 2026-05-10T14:52:07.462609+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

41 extracted references · 41 canonical work pages · 1 internal anchor

  1. [1]

    Alon Amid, David Biancolin, Abraham Gonzalez, Daniel Grubb, Sagar Karandikar, Harrison Liew, Albert Magyar, Howard Mao, Albert Ou, Nathan Pemberton, Paul Rigge, Colin Schmidt, John Wright, Jerry Zhao, Yakun Sophia Shao, Krste Asanović, and Borivoje Nikolić. 2020. Chipyard: Integrated Design, Simulation, and Implementation Framework for Custom SoCs.IEEE Mi...

    work page 2020

  2. [2]

    Guido Araujo, Jose Moreira, Rafael Sene, and Erich Focht. 2025. Integrated Matrix Extension. https://github.com/riscv-admin/integrated-matrix-extension

    work page 2025

  3. [3]

    Arm. 2025. KleidiAI. https://gitlab.arm.com/kleidi/kleidiai

    work page 2025

  4. [4]

    Patterson, Brian Richards, Colin Schmidt, Stephen Twigg, Huy Vo, and Andrew Waterman

    Krste Asanović, Rimas Avizienis, Jonathan Bachrach, Scott Beamer, David Bian- colin, Christopher Celio, Henry Cook, Daniel Dabbelt, John Hauser, Adam Izraele- vitz, Sagar Karandikar, Ben Keller, Donggyu Kim, John Koenig, Yunsup Lee, Eric Love, Martin Maas, Albert Magyar, Howard Mao, Miquel Moreto, Albert Ou, David A. Patterson, Brian Richards, Colin Schmi...

    work page 2016

  5. [5]

    Hongtao Chen, Weiyu Xie, Boxin Zhang, Jingqi Tang, Jiahao Wang, Jianwei Dong, Shaoyuan Chen, Ziwei Yuan, Chen Lin, Chengyu Qiu, Yuening Zhu, Qingliang Ou, Jiaqi Liao, Xianglin Chen, Zhiyuan Ai, Yongwei Wu, and Mingxing Zhang. 2025. KTransformers: Unleashing the Full Potential of CPU/GPU Hybrid Inference for MoE Models. InProceedings of the ACM SIGOPS 31st...

    work page 2025

  6. [6]

    Francesco Conti, Gianna Paulin, Angelo Garofalo, Davide Rossi, Alfio Di Mauro, Georg Rutishauser, Gianmarco Ottavi, Manuel Eggiman, Hayate Okuhara, and Luca Benini. 2024. Marsellus: A Heterogeneous RISC-V AI-IoT End-Node SoC With 2–8 b DNN Acceleration and 30%-Boost Adaptive Body Biasing.IEEE Journal of Solid-State Circuits59, 1 (2024), 128–142

    work page 2024

  7. [7]

    Tri Dao. 2023. Flashattention-2: Faster attention with better parallelism and work partitioning.arXiv preprint arXiv:2307.08691(2023)

    work page internal anchor Pith review arXiv 2023

  8. [8]

    Fu, Stefano Ermon, Atri Rudra, and Christopher Ré

    Tri Dao, Daniel Y. Fu, Stefano Ermon, Atri Rudra, and Christopher Ré

    work page

  9. [9]

    InProceedings of the 36th International Conference on Neural Informa- tion Processing Systems(NIPS)(New Orleans, LA, USA)

    FLASHATTENTION: fast and memory-efficient exact attention with IO- awareness. InProceedings of the 36th International Conference on Neural Informa- tion Processing Systems(NIPS)(New Orleans, LA, USA). Curran Associates Inc., Article 1189, 16 pages

    work page

  10. [10]

    OpenVINO developers. 2025. Intel Distribution of OpenVINO Toolkit. https: //github.com/openvinotoolkit/openvino

    work page 2025

  11. [11]

    ONNX Runtime developers. 2021. ONNX Runtime. https://onnxruntime.ai/

    work page 2021

  12. [12]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Associa- tion for Computational Linguistics: Human Language Technologies (NAACL-HLT) (Minneapolis, MN, USA). Association for Com...

    work page 2019

  13. [13]

    Greg Favor. 2025. Vector-Matrix Extension. https://riscv.atlassian.net/wiki/ spaces/VMEX/pages/554991628/Vector-Matrix+Extension+VME+-+PoW

    work page arXiv 2025

  14. [14]

    Peng Gao, Yang Liu, Haonan Sun, Jiang Jiang, Jun Wang, Zonghui Hong, and Jiali Qu. 2025. OASIS: A Commercial High Performance Terminal AI Processor Supporting RISC-V Tensor Extension Instructions. InProceedings of the 58th IEEE/ACM International Symposium on Microarchitecture (MICRO)(New York, NY, USA). ACM, 1264–1283

    work page 2025

  15. [15]

    Hasan Genc, Seah Kim, Alon Amid, Ameer Haj-Ali, Vighnesh Iyer, Pranav Prakash, Jerry Zhao, Daniel Grubb, Harrison Liew, Howard Mao, Albert Ou, Colin Schmidt, Samuel Steffl, John Wright, Ion Stoica, Jonathan Ragan-Kelley, Krste Asanovic, Borivoje Nikolic, and Yakun Sophia Shao. 2021. Gemmini: Enabling Systematic Deep-Learning Architecture Evaluation via Fu...

    work page 2021

  16. [16]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. 2024. The llama 3 herd of models

    work page 2024

  17. [17]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep Residual Learning for Image Recognition. In2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)(Las Vegas, NV, USA). IEEE, 770–778

    work page 2016

  18. [18]

    Sarda, Vikram Jain, Kodai Ueyoshi, Ioannis A

    Pouya Houshmand, Giuseppe M. Sarda, Vikram Jain, Kodai Ueyoshi, Ioannis A. Papistas, Man Shi, Qilin Zheng, Debjyoti Bhattacharjee, Arindam Mallik, Peter Debacker, Diederik Verkest, and Marian Verhelst. 2023. DIANA: An End-to-End Hybrid DIgital and ANAlog Neural Network SoC for the Edge.IEEE Journal of Solid-State Circuits58, 1 (2023), 203–215

    work page 2023

  19. [19]

    Intel. 2024. Intel 64 and IA-32 Architectures Optimization Reference Man- ual. https://cdrdv2-public.intel.com/814201/355308-Optimization-Reference- Manual-049-Changes-Doc.pdf

    work page 2024

  20. [20]

    Intel. 2025. Intel Memory Latency Checker v3.12. https://www.intel.com/content/ www/us/en/developer/articles/tool/intelr-memory-latency-checker.html

    work page 2025

  21. [21]

    Benjamin Lefaudeux, Francisco Massa, Diana Liskovich, Wenhan Xiong, Vittorio Caggiano, Sean Naren, Min Xu, Jieru Hu, Marta Tintore, Susan Zhang, Patrick Labatut, Daniel Haziza, Luca Wehrstedt, Jeremy Reizenstein, and Grigory Sizov

    work page

  22. [22]

    https: //github.com/facebookresearch/xformers

    xFormers: A modular and hackable Transformer modelling library. https: //github.com/facebookresearch/xformers

    work page

  23. [23]

    Igual, and Enrique S

    Héctor Martínez, Adrián Castelló, Francisco D. Igual, and Enrique S. Quintana- Ortí. 2026. The cambrian explosion of mixed-precision matrix multiplication for quantized deep learning inference.Future Generation Computer Systems(2026)

    work page 2026

  24. [24]

    McCalpin

    John D. McCalpin. 1995. Memory Bandwidth and Machine Balance in Current High Performance Computers.IEEE Computer Society Technical Committee on Computer Architecture (TCCA) Newsletter(Dec. 1995)

    work page 1995

  25. [25]

    José E Moreira, Kit Barton, Steven Battle, Peter Bergner, Ramon Bertran, Puneeth Bhat, Pedro Caldeira, David Edelsohn, Gordon Fossum, Brad Frey, et al

    work page

  26. [26]

    A matrix math facility for Power ISA (TM) processors.arXiv preprint arXiv:2104.03142(2021)

    work page arXiv 2021

  27. [27]

    oneDNN Contributor. 2025. oneAPI Deep Neural Network Library (oneDNN). https://github.com/uxlfoundation/oneDNN

    work page 2025

  28. [28]

    Eric Qin, Ananda Samajdar, Hyoukjun Kwon, Vineet Nadella, Sudarshan Srini- vasan, Dipankar Das, Bharat Kaul, and Tushar Krishna. 2020. SIGMA: A Sparse and Irregular GEMM Accelerator with Flexible Interconnects for DNN Training. In2020 IEEE International Symposium on High Performance Computer Architecture (HPCA)(San Diego, CA, USA). 58–70

    work page 2020

  29. [29]

    Scott Gardner, Itay Hubara, Sachin Idgunji, Thomas B

    Vijay Janapa Reddi, Christine Cheng, David Kanter, Peter Mattson, Guenther Schmuelling, Carole-Jean Wu, Brian Anderson, Maximilien Breughe, Mark Charlebois, William Chou, Ramesh Chukka, Cody Coleman, Sam Davis, Pan Deng, Greg Diamos, Jared Duke, Dave Fick, J. Scott Gardner, Itay Hubara, Sachin Idgunji, Thomas B. Jablin, Jeff Jiao, Tom St. John, Pankaj Kan...

    work page 2020

  30. [30]

    Stefan Remke and Alexander Breuer. 2024. Hello SME! Generating Fast Matrix Multiplication Kernels Using the Scalable Matrix Extension. InWorkshops of the International Conference for High Performance Computing, Networking, Storage and Analysis(SC24-W)(Atlanta, GA, USA). IEEE Press, 1443–1454

    work page 2024

  31. [31]

    Paul Rosenfeld, Elliott Cooper-Balis, and Bruce Jacob. 2011. DRAMSim2: A Cycle Accurate Memory System Simulator.IEEE Computer Architecture Letters10, 1 (2011), 16–19

    work page 2011

  32. [32]

    Rafael Sene and Philipp Tomsich. 2023. Attached Matrix Extension. https: //github.com/riscv-admin/attached-matrix-extension

    work page 2023

  33. [33]

    Starke and Brian W

    William J. Starke and Brian W. Thompto. 2020. IBM’s POWER10 Processor. In IEEE Hot Chips 32 Symposium, HCS 2020, Palo Alto, CA, USA, August 16-18, 2020 (Palo Alto, CA, USA). IEEE, 1–43

    work page 2020

  34. [34]

    Josse Van Delm, Anton Lydike, Joren Dumoulin, Jonas Crols, Xiaoling Yi, Ryan Antonio, Jackson Woodruff, Tobias Grosser, and Marian Verhelst. 2025. The Con- figuration Wall: Characterization and Elimination of Accelerator Configuration Overhead. To appear in Proceedings of ASPLOS 2026

    work page 2025

  35. [35]

    Guangxuan Xiao, Ji Lin, Mickaël Seznec, Hao Wu, Julien Demouth, and Song Han. 2023. SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models. InInternational Conference on Machine Learning(ICML) (Honolulu, Hawaii, USA)(Proceedings of Machine Learning Research, Vol. 202). PMLR, 38087–38099

    work page 2023

  36. [36]

    Yinan Xu, Zihao Yu, Dan Tang, Guokai Chen, Lu Chen, Lingrui Gou, Yue Jin, Qianruo Li, Xin Li, Zuojun Li, Jiawei Lin, Tong Liu, Zhigang Liu, Jiazhan Tan, Huaqiang Wang, Huizhe Wang, Kaifan Wang, Chuanqi Zhang, Fawang Zhang, Linjuan Zhang, Zifei Zhang, Yangyang Zhao, Yaoyang Zhou, Yike Zhou, Jiangrui Zou, Ye Cai, Dandan Huan, Zusong Li, Jiye Zhao, Zihao Che...

    work page 2023

  37. [37]

    Xiaoling Yi, Ryan Antonio, Joren Dumoulin, Jiacong Sun, Josse Van Delm, Guil- herme Pereira Paim, and Marian Verhelst. 2025. OpenGeMM: A Highly-Efficient GeMM Accelerator Generator with Lightweight RISC-V Control and Tight Mem- ory Coupling. InProceedings of the 30th Asia and South Pacific Design Automation Conference(Tokyo, Japan). ACM, 1055–1061

    work page 2025

  38. [38]

    Martin Kroeker Zhang Xianyi. 2025. OpenBLAS. https://github.com/ OpenMathLib/OpenBLAS

    work page 2025

  39. [39]

    2024.The Saturn Microarchitecture Manual

    Jerry Zhao, Daniel Grubb, Miles Rusch, Tianrui Wei, Kevin Anderson, Borivoje Nikolic, and Krste Asanović. 2024.The Saturn Microarchitecture Manual. Technical Report. EECS Department, University of California, Berkeley. http://www2.eecs. berkeley.edu/Pubs/TechRpts/2024/EECS-2024-215.html

    work page 2024

  40. [40]

    Jerry Zhao, Ben Korpan, Abraham Gonzalez, and Krste Asanovic. 2020. Sonic- BOOM: The 3rd Generation Berkeley Out-of-Order Machine. InFourth Workshop on Computer Architecture Research with RISC-V (CARRV)

    work page 2020

  41. [41]

    Jerry Zhao, Jennifer Zhou, Albert Ou, Abraham Gonzalez, and Lux Zhang. 2023. Shuttle: A Rocket-based Superscalar In-order RISC-V Core. https://github.com/ ucb-bar/shuttle/tree/master

    work page 2023