QAPLIB tai40b に関する実験結果を掲載します。実行環境は以下の九大スパコン CX400 です。
1218400 = mDIM
2 = nBlock
-1463440 1522 = bLOCKsTRUCT
param is /home/usr2/i70202a/sdpa-src/sdpara.7.6.0.src-RC1G/param.sdpa
data is /home/usr2/i70202a/data/TAI40/tai40b_ZKRW_R3_e0.dat-s : sparse
init is /home/usr2/i70202a/sdpa-src/sdpara.7.6.0.src-RC1G/sdpara.init : dense
out is /home/usr2/i70202a/sdpa-src/sdpara.7.6.0.src-RC1G/out.tai40b_ZKRW_R3_e0.12
NumNodes is set as 128
NumThreads is set as 16
Schur computation : DENSE
mu thetaP thetaD objP objD alphaP alphaD beta
67 4.9e-11 5.0e-10 1.2e-10 +5.06e-01 +5.06e-01 6.4e-01 3.4e-01 1.00e-01
68 3.2e-11 5.0e-10 1.8e-10 +5.06e-01 +5.06e-01 4.0e-01 2.4e-01 1.00e-01
69 2.5e-11 5.0e-10 1.8e-10 +5.06e-01 +5.06e-01 4.0e-01 2.4e-01 1.00e-01
phase.value = pdOPT
Iteration = 69
mu = +2.4606883310861874e-11
relative gap = +1.1102230246251565e-15
gap = -1.1102230246251565e-15
digits = +1.4658628225808677e+01
objValPrimal = +5.0586945341044354e-01
objValDual = +5.0586945341044465e-01
p.feas.error = +5.0535933300923097e-08
d.feas.error = +1.8044232790348676e-08
total time = 12999.257194
高性能演算サーバシステム (Fujitsu PRIMERGY CX400)
演算ノード 理論演算性能 345.6GFLOPS
主記憶容量 128GB
メモリバンド幅 102.4GB/s
総ノード数 1476ノード
総プロセッサ (コア) 数 2952プロセッサ (23616コア)
理論演算性能(倍精度実数)の総和 966.2TFLOPS
(CPU: 510.1TF, GPGPU[K20m&K20Xm]: 456.1TF)
![](https://blogimg.goo.ne.jp/thumbnail/4c/f3/ac3829e91dd289d784395060806adb50_s.jpg)
1218400 = mDIM
2 = nBlock
-1463440 1522 = bLOCKsTRUCT
param is /home/usr2/i70202a/sdpa-src/sdpara.7.6.0.src-RC1G/param.sdpa
data is /home/usr2/i70202a/data/TAI40/tai40b_ZKRW_R3_e0.dat-s : sparse
init is /home/usr2/i70202a/sdpa-src/sdpara.7.6.0.src-RC1G/sdpara.init : dense
out is /home/usr2/i70202a/sdpa-src/sdpara.7.6.0.src-RC1G/out.tai40b_ZKRW_R3_e0.12
NumNodes is set as 128
NumThreads is set as 16
Schur computation : DENSE
mu thetaP thetaD objP objD alphaP alphaD beta
67 4.9e-11 5.0e-10 1.2e-10 +5.06e-01 +5.06e-01 6.4e-01 3.4e-01 1.00e-01
68 3.2e-11 5.0e-10 1.8e-10 +5.06e-01 +5.06e-01 4.0e-01 2.4e-01 1.00e-01
69 2.5e-11 5.0e-10 1.8e-10 +5.06e-01 +5.06e-01 4.0e-01 2.4e-01 1.00e-01
phase.value = pdOPT
Iteration = 69
mu = +2.4606883310861874e-11
relative gap = +1.1102230246251565e-15
gap = -1.1102230246251565e-15
digits = +1.4658628225808677e+01
objValPrimal = +5.0586945341044354e-01
objValDual = +5.0586945341044465e-01
p.feas.error = +5.0535933300923097e-08
d.feas.error = +1.8044232790348676e-08
total time = 12999.257194
高性能演算サーバシステム (Fujitsu PRIMERGY CX400)
演算ノード 理論演算性能 345.6GFLOPS
主記憶容量 128GB
メモリバンド幅 102.4GB/s
総ノード数 1476ノード
総プロセッサ (コア) 数 2952プロセッサ (23616コア)
理論演算性能(倍精度実数)の総和 966.2TFLOPS
(CPU: 510.1TF, GPGPU[K20m&K20Xm]: 456.1TF)
![](https://blogimg.goo.ne.jp/thumbnail/4c/f3/ac3829e91dd289d784395060806adb50_s.jpg)