I've built patched NWChem 6.6 with Intel 2017.4 (using ifort instead of gfortran for mpif90), configured for qm only.
As I run the QA tests (they're still going, on 2 nodes/24 ppn), I have 160 passes and 35 failures. The failures that I've looked at differ a little from the reference data, but not so much that I'd assume a total failure. For example, the cu2_m06-l test gives
< Total DFT energy = -392.32794
---
> Total DFT energy = -392.32802
So, not affecting "chemical accuracy," but still not passing validation.
Is there an option that should be passed to the Intel 2017.4 fortran compiler to tighten up the numbers?
If not, are these numbers considered unofficially OK?
Thanks; Chris
[chch5350@login01 testoutputs]$ ls *.diff
acr_lcblyp.out.diff cosmo_h3co_ecp.out.diff dft_x.out.diff hi_zora_sf.out.diff
autosym.out.diff cosmo_h3co_gp.out.diff dielsalder.out.diff mep-test.out.diff
bsse_dft_trimer.out.diff cosmo_h3co.out.diff etf_hcons.out.diff o2_zora_so.out.diff
ch3f-lc-wpbeh.out.diff cu2_m06-l.out.diff h2o_diag_to_cg_ub3lyp.out.diff oh2.out.diff
ch3f-lc-wpbe.out.diff dft_bsse_d3_opt.out.diff h2o_hcons.out.diff prop_cg_nh3_b3lyp.out.diff
ch3_m06-hf.out.diff dft_ch3ochch32.out.diff h2o_raman_3.out.diff sadsmall.out.diff
ch3radical_rot.out.diff dft_cr2.out.diff h2o_raman_4.out.diff tce_cr_eom_t_ch_rohf.out.diff
ch3radical_unrot.out.diff dft_li2freq.out.diff h2o_vscf.out.diff tropt-ch3nh2.out.diff
cho_bp_props.out.diff dft_ozone.out.diff hess_nh3_ub3lyp.out.diff
|