コンテンツにスキップ

bench-ghia1000

英語 | 日本語

« Home · « 層流・解析解による検証

Re=1000 lid-driven cavity: vertical-centerline u(0.5, y, 0.5) from three solves -- uniform coarse (n), AMR (coarse n + wall-band refinement to 2n), and uniform fine (2n) -- compared against Ghia, Ghia & Shin (1982) Table I.

概要: Ghia らの Re=1000 キャビティを、粗い一様格子・AMR・細かい一様格子の3通りで解き、中心線速度を比較する。AMR が一様細格子にどれだけ迫れるかを見るのが主眼。Ghia は 2 次元の基準なので、公平に比べるには環境変数 SPANZ で擬似2D(薄い周期スラブ)にする必要がある — 3次元立方キャビティのままだと物理的に別問題になる。

Geometry

bench-ghia1000 geometry

Animation

Fields

この計算が最後の出力ステップで書き出した場をそのまま描いたもの。3D ケースは中央断面。

bench-ghia1000 umag

bench-ghia1000 ux

Build & run

cmake -S native -B native/build && cmake --build native/build -j
./native/build/orheo-bench-ghia1000

ghia_re1000 [n_coarse] [t_end] [re] [collision] [mode] [band]

このページの図を作った計算のコスト(Miyabi-G Hopper GPU (GH200, -stdpar=gpu)): 5.2 s wall。

Validation

Tier T2 — 単点比較
QoI u_min_centerline
Reference Ghia, Ghia & Shin (1982) J.Comput.Phys 48:387 -- Re=100/400/1000 centreline
Gate — ゲートなし
Measured -0.376 (rel. err +1.80%) — cross-code run (cavity_ghia)

Source & run details

View source

Item Value
Target orheo-bench-ghia1000
Source native/benchmarks/ghia_re1000.cpp
Lattice —
GPU twin yes (orheo-bench-ghia1000-gpu)
Modules block_forest.hpp
Gallery run 2026-08-30 @ 38ad1eaa
Args 48 12 1000 reg coarse
Wall time 5.20 s
CPU time 5.08 s
Device Miyabi-G Hopper GPU (GH200, -stdpar=gpu)

Computation log

この計算の実際の出力(先頭と末尾)。

Re=1000, coarse n=48 (fine 2n=96 near walls), band=12, t_end=12.0, collision=reg, mode=coarse
  [uniform coarse] starting: n=48, 5760 steps, tau=0.5144
  [uniform coarse] step 288/5760 (t=0.600/12.0, 0s): centerline u_min=-0.04176
  [uniform coarse] step 576/5760 (t=1.200/12.0, 0s): centerline u_min=-0.03196
  [uniform coarse] step 864/5760 (t=1.800/12.0, 1s): centerline u_min=-0.05404
  [uniform coarse] step 1152/5760 (t=2.400/12.0, 1s): centerline u_min=-0.05235
  [uniform coarse] step 1440/5760 (t=3.000/12.0, 1s): centerline u_min=-0.04003
  [uniform coarse] step 1728/5760 (t=3.600/12.0, 1s): centerline u_min=-0.05113
  [uniform coarse] step 2016/5760 (t=4.200/12.0, 2s): centerline u_min=-0.04776
  [uniform coarse] step 2304/5760 (t=4.800/12.0, 2s): centerline u_min=-0.05966
  [uniform coarse] step 2592/5760 (t=5.400/12.0, 2s): centerline u_min=-0.04428
  [uniform coarse] step 2880/5760 (t=6.000/12.0, 2s): centerline u_min=-0.05348
  [uniform coarse] step 3168/5760 (t=6.600/12.0, 3s): centerline u_min=-0.04962
  [uniform coarse] step 3456/5760 (t=7.200/12.0, 3s): centerline u_min=-0.04868