Table 2.

Comparison of different methods to solve a model covariance matrix, with n=70 000.

RHSiterations (low/high)time (s)average power (W)standard error (W)energy (kW⋅s)TOP500 (GFLOPS)Green500 (GFLOPS/W)
d=1
Cholesky321/1546.0190.013.5103.7214.41.11
CG1179/2115.7175.73.220.315.30.09
CG3283/1112.5197.17.522.2233.81.19
banded CG1691/811.2175.96.22.07.10.04
banded CG32692/847.0179.415.38.454.30.30
d=2
Cholesky321/1546.0190.013.5103.7214.41.11
CG183/153.5177.83.19.515.30.09
CG3285/1119.3195.211.223.3225.91.16
banded CG1173/22.9174.47.00.56.30.04
banded CG32177/213.8177.614.62.546.10.26
d=4
Cholesky321/1546.0190.013.5103.7214.41.11
CG185/153.8179.01.89.615.70.09
CG3288/1125.5195.010.824.6222.21.13
banded CG185/11.8174.14.90.35.50.03
banded CG3288/18.4172.614.21.537.80.22