Poisson solver, strong scaling

Filter regionsRegions

all 5

Depth

all levels

The same problem on 1 to 8 MPI ranks.

Tags: cpu, strong-scaling

Runs

run	file	recorded	ranks	wall time	vs baseline	page
np1	np1.h5	2026-10-17 12:00 UTC	1	414 ms	baseline	run page
np2	np2.h5	2026-10-17 12:00 UTC	2	215 ms	−48.1%	run page
np4	np4.h5	2026-10-17 12:00 UTC	4	125 ms	−69.9%	run page
np8	np8.h5	2026-10-17 12:00 UTC	8	74.8 ms	−81.9%	run page

Each run's page is linked on the right.

What changed

Baselinenp1np2np4np8Candidatenp1np2np4np8⇄ swap every run against the baseline

Durations

Showtotalownavg/callcallsOrdercall treelargest change

Charts

Expand all chartsCollapse all charts

Region durations

Grouped bars compare each region's mean call duration across the profiled runs.

bar labels

run MPI ranks OpenMP threads nodes ranks × threads wall time bar value

Open in new tab ↗

Speedup

Each line is one region's speedup: its mean call duration on the run with the fewest MPI ranks divided by its mean call duration on each run. The dashed line is ideal scaling; a region below it gains less than its extra MPI ranks would allow. The 8 regions with the most time on that first run are shown. Valid for a strong-scaling study: the same total problem size on every run.

x-axisMPI ranksnodesOpen in new tab ↗

Rank imbalance

Each line compares a region's total duration by rank; its dashed line marks the mean across ranks. Points far from the mean identify stragglers.

Open in new tab ↗

Rank heatmap

This heatmap uses exclusive timings. Exclusive duration is the time spent in a region itself, excluding time spent in nested child regions; this prevents the enclosing session region from dominating the heatmap.

Open in new tab ↗

Metadata

commit	0000000
system	synthetic