White box / Verification methods / Verification methods and benchmark design

Recorded structured-TET4 workloads

Verification method · assumptions, essential equations, full derivation, algorithm, code and verification references.

Browse another sheet Verification methods / Verification methods and benchmark design
Verification methodsVerification methods and benchmark designSheet 6 / 7
Verification methodHISTORICAL ENVIRONMENT-SPECIFIC RECORDS

Selected 1.029M, 3.0M, 5.013M, and approximately 10.125M active-DOF cases used declared PETSc/MPI routes.

01 Model and scope

These workload records measure selected structured meshes at different active-DOF counts and on declared hardware/backend configurations. They are observations for those runs; they do not provide a hardware-independent performance law or general mixed-mesh scaling result.

Route assumptions and calculation checks

  1. The 5.013M case records two complete Silver replays.
  2. The ~10.125M entry is a bounded observation; it does not prove a general scaling law.
  3. These workloads are separate from the compact learning fixtures.

02 Essential equations and symbols

1,029,000DOF⋅stablePETScreplays1,029,000 DOF \cdot{} stable PETSc replays3,000,000DOF⋅frozenCG/GAMGroute3,000,000 DOF \cdot{} frozen CG/GAMG route5,012,640DOF⋅9,773,946TET4,8ranks5,012,640 DOF \cdot{} 9,773,946 TET4, 8 ranks10,125,000DOF⋅boundedsolveobservation10,125,000 DOF \cdot{} bounded solve observation

Definitions and conventions

nDOF
Active degrees of freedom.
Twall
Elapsed wall-clock time for the recorded run.
memory
Observed or estimated peak storage, as declared by the record.
configuration
Hardware, backend, mesh and solver settings tied to each measurement.
Recorded structured-TET4 workloads · method mapA performance observation belongs to the frozen mesh, hardware, ranks, backend, solver, and measured timing boundary.
3 detailed steps

03 Step by step derivation

Follow the assumptions, equations and discrete operations in order. The full explanation stays visible.

1 · Treat each size as a frozen workload

The recorded active-DOF counts—1.029M, 3.0M, 5.013M, and approximately 10.125M—refer to particular structured TET4 meshes, solver settings, rank counts, and environments. They are not interchangeable samples from one hardware-independent performance law.

nactive=1,029,000;  3,000,000;  5,012,640;  ∼10,125,000n_{active}=1{,}029{,}000;\;3{,}000{,}000;\;5{,}012{,}640;\;\sim10{,}125{,}000

2 · Separate convergence, memory, and cost

For each run, capture wall time, peak memory, residual, iteration count, ranks, solver/preconditioner, source revision, and replay status. The 5.013M-DOF record includes two complete Silver replays; the approximately 10.125M entry is a bounded observation rather than a repeated scaling campaign.

Sp=T1Tp,Ep=Spponly for comparable rank-scaled runsS_p=\frac{T_1}{T_p},\qquad E_p=\frac{S_p}{p}\quad\text{only for comparable rank-scaled runs}

3 · Do not transfer scale claims to learning examples

The million-DOF workloads use separate declared PETSc/MPI routes and infrastructure. They do not establish general mixed-mesh performance, partitioned assembly, nonlinear scaling, GPU support, or reproducibility on unrelated hardware. The compact JSON examples serve a different purpose.

04 Algorithm and implementation

This sheet provides explanatory references rather than an inline source excerpt. Use the labeled sources to inspect the precise boundary.

05 Verification and limitations

HISTORICAL ENVIRONMENT-SPECIFIC RECORDS

Each reference supports its own model, observable and analysis route. Read the original contract and limitations before transferring a conclusion.

06 Sources and related sheets

Continue the reasoning

Back to sheet contents ↑