CVDP

783 engineer-authored tasks in 13 categories covering RTL design and verification (stimulus, checker, SVA generation) with hidden cocotb harnesses.

Eval / BenchmarkOpen sourceMaintainedECADVerification

What it does and how it connects

CVDP, the Comprehensive Verilog Design Problems benchmark, is a framework for evaluating LLM and agent solutions on hardware verification challenges. It includes 783 problems across 13 task categories, covering RTL generation, verification, debugging, specification alignment and technical Q and A, authored by experienced hardware engineers. Problems are offered in both non-agentic and agentic formats, and the paper reports state-of-the-art models achieving no more than 34 percent pass@1 on code generation.

Evaluation runs against hidden cocotb harnesses. The open-source simulation image bundles cocotb 2.0.1, pytest, Icarus Verilog, Yosys and Verilator; a commercial dataset variant needs Cadence Xcelium and a license network. Both a direct LLM workflow and a Docker-based agentic workflow are supported, with multi-sample pass@k evaluation.

The dataset is published on HuggingFace. The README notes 20 datapoints were omitted from the initial public release for test harness issues and licensing restrictions, and reference solutions were excluded to mitigate data contamination. The Silicon Integration Initiative (Si2) is adopting CVDP into its LLM Benchmarking Coalition, which extends the problems, refines metrics and maintains a leaderboard.

  • VerilogEval

    156 HDLBits-derived description-to-Verilog tasks verified by Icarus Verilog simulation, rather than 783 engineer-authored problems across 13 categories.

  • RTLLM 2.0

    50 realistic natural-language-described designs scored on syntax, functionality, synthesis and PPA, rather than 783 engineer-authored problems across 13 categories.

  • TuRTLe

    A unified harness running VerilogEval, RTLLM, VGen, RTL-Repo and NotSoTiny under several simulators, rather than a single 783-problem benchmark with hidden harnesses.

  • HWE-Bench

    417 real fail-to-pass bug-fix cases from six open-source hardware projects with a live leaderboard, rather than 783 engineer-authored design and verification problems.

  • ChipBench

    44 hierarchical Verilog modules, 89 injected-bug debug cases and 132 reference-model tasks, rather than 783 engineer-authored problems across 13 categories.

Sources

  1. 1READMEgithub.com
  2. 2GitHub repo statsapi.github.com
  3. 3Paper abstract (arXiv:2506.14074)arxiv.org

Not verified

  • Pricing: none stated on the repo (open source benchmark).
  • Real user voices: no real linked post read for this page.
  • No GitHub releases exist, so latest release fields are null.
  • Related Error Index entries: no matching entry verified.

Last verified November 6, 2026. Listings are informational. The evidence score measures public, verifiable evidence and is not an endorsement or a compatibility claim.