Changes between Initial Version and Version 1 of Ticket #2563, comment 12
- Timestamp:
- 2020-11-05T14:52:44+01:00 (3 years ago)
Legend:
- Unmodified
- Added
- Removed
- Modified
-
Ticket #2563, comment 12
initial v1 1 1 I have tried running a back-to-back of the current trunk vs r4.0.4 on the Met Office HPC. Looking at the timing.outputs for ORCA2_ICE_PISCES/LONG, running SPMD (attached XIOS servers), there is ''no significant increase in solve time''. There is a slight increase (total solve time: trunk = 19,774s vs r4.0.4 = 18,516s) but this is solely due to the increase in dia_ptr time. The order of the timings is roughly the same between the two with tra_adv still the largest contributor. It would seem that we could narrow it down to some kind of architecture/compiler/environment specific issue which affects: sbc, zdf_phy, zps_hde and icb_stp. 2 3 Job Specifications 4 Compiler: Intel (version 18.0.5.274) 5 Flags: %FCFLAGS -r8 -i4 -init=zero -init=arrays -traceback -debug minimal -debug inline-debug-info -O2 -fp-model consistent 6 Cores: 32 (jnpi=4, jpnj=8) 7 XIOS: Attached (SPMD)