Brett W. Coon
Brett W. Coon has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- Arash Nasr-Esfahany
MIT CSAIL, Cambridge, MA, USA
2 shared publications - David Culler
Google
2 shared publications - Hanna Alam
Google
2 shared publications - Henry M. Levy
Google and University of Washington
2 shared publications - Martin Dixon
Google
2 shared publications - Mohammad Alizadeh
Massachusetts Institute of Technology, Cambridge, MA, USA
2 shared publications
Latest Publications
- Open Access
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 46 (5) , 29-36
Fast and Accurate CPU Performance Modeling With Compositional Analytical–Machine Learning Fusion
Cycle-level simulators like gem5 are widely used in microarchitecture design, but prohibitively slow for large-scale design space exploration. We present Concorde, a new methodology for learning fast and accurate performance models of microarchitectures. Unlike existing simulators and learning approaches that emulate each instruction, Concorde uses simple analytical models to compute per-component performance bounds, producing compact distributions that an ML model maps to program behavior. Concorde is more than five orders of magnitude faster than Google’s cycle-level simulator, with ∼2% average Cycles-Per-Instruction (CPI) prediction error across SPEC2017, open-source, and Google’s proprietary benchmarks. Concorde is already shaping large-scale architecture exploration at Google, contributing to internal CPU design projects. This enables analyses previously infeasible: in about an hour, we conducted a first-of-its-kind fine-grained performance attribution to microarchitectural components, requiring ∼150M CPI evaluations. The same compositional analytical-ML recipe extends beyond CPUs, suggesting a broader approach to building fast, interpretable performance models.
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 46 (5) , 29-36
Open AccessFast and Accurate CPU Performance Modeling With Compositional Analytical–Machine Learning Fusion
Cycle-level simulators like gem5 are widely used in microarchitecture design, but prohibitively slow for large-scale design space exploration. We present Concorde, a new methodology for learning fast and accurate performance models of microarchitectures. Unlike existing simulators and learning approaches that emulate each instruction, Concorde uses simple analytical models to compute per-component performance bounds, producing compact distributions that an ML model maps to program behavior. Concorde is more than five orders of magnitude faster than Google’s cycle-level simulator, with ∼2% average Cycles-Per-Instruction (CPI) prediction error across SPEC2017, open-source, and Google’s proprietary benchmarks. Concorde is already shaping large-scale architecture exploration at Google, contributing to internal CPU design projects. This enables analyses previously infeasible: in about an hour, we conducted a first-of-its-kind fine-grained performance attribution to microarchitectural components, requiring ∼150M CPI evaluations. The same compositional analytical-ML recipe extends beyond CPUs, suggesting a broader approach to building fast, interpretable performance models.
- Open Access
- 20 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 1480-1494
Concorde: Fast and Accurate CPU Performance Modeling with Compositional Analytical-ML Fusion
- 20 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 1480-1494
Open AccessConcorde: Fast and Accurate CPU Performance Modeling with Compositional Analytical-ML Fusion