This commit is contained in:
2026-10-23 03:00:20 +02:00
parent 589d22aa5e
commit 7a5c77c1c8
31 changed files with 1196 additions and 44 deletions

View File

@@ -88,11 +88,13 @@ shaft power. There is no forced power gain. Effective area/efficiency, gas
properties and the base engine's approximate combustion/overlap affect these
numbers strongly. They are prototype comparisons, not real engine predictions.
The benchmark used four solver substeps per output frame. The turbo cost .81
The original benchmark used four solver substeps per output frame. The turbo cost .81
CPU seconds per simulated second at 4000 RPM and 1.28 at 7000 RPM, versus .53
for the unboosted twin. Sustained high-load real-time playback is an open issue;
monitor underruns. The mixed transient cost .84 wall seconds per simulated
second. Tests run headlessly; latest Windows runtime is unverified.
second. The subsequent [optimization audit](OPTIMIZATION.md) resolves the
subnormal-tracer slowdown and measures about .65 seconds per simulated second
in its 7,000-RPM workload. Tests run headlessly; latest Windows runtime is unverified.
## Run